Add research documentation for marketing skills and superpowers plugins

2026-09-13 02:48:06 +07:00 · 2026-04-27 09:48:22 +01:00
172 changed files with 1702 additions and 7755 deletions
@@ -1,10 +0,0 @@
 # Explicit `/setup-matt-pocock-skills` pointer only for hard dependencies
 Engineering skills depend on per-repo config (issue tracker, triage label vocabulary, domain doc layout) seeded by `/setup-matt-pocock-skills`. Some skills cannot meaningfully function without that config — they have to publish to a specific issue tracker or apply a specific label string. Others only use it to sharpen output (vocabulary, ADR awareness) and degrade gracefully without it.
 We split these into **hard-dependency** and **soft-dependency** skills:
 - **Hard dependency** (`to-tickets`, `to-spec`, `triage`) — include an explicit one-liner: _"… should have been provided to you — run `/setup-matt-pocock-skills` if not."_ Without the mapping, output is wrong, not just fuzzy.
 - **Soft dependency** (`diagnose`, `tdd`, `improve-codebase-architecture`) — reference "the project's domain glossary" and "ADRs in the area you're touching" in vague prose only. If the docs aren't there, the skill still works; output is just less sharp.
 The split keeps soft-dependency skills token-light and avoids cargo-culting the setup pointer into places where it isn't load-bearing.
@@ -1,41 +0,0 @@
 # Ship the skill set as a native Claude Code plugin; defer a native Codex plugin
 These skills have always been installable via [skills.sh](https://skills.sh/mattpocock/skills) (`npx skills add mattpocock/skills`), which copies editable skill files into a user's project across Claude Code, Codex, and other Agent-Skills-standard harnesses. A recurring request is a **plug-and-play** distribution: subscribe to the set as a read-only, always-current bundle you don't edit, rather than a fork you own. That is exactly what native plugin systems provide.
 We ship a native **Claude Code plugin** and, for now, **defer** a native **Codex plugin**. The split is forced by how each ecosystem's plugin manifest selects skills, against this repo's bucketed layout.
 ## The constraint: bucketed skills vs. single-path selection
 Skills live in bucket folders under `skills/` — `engineering/` and `productivity/` are **promoted** (shipped); `misc/`, `personal/`, `in-progress/`, and `deprecated/` are **not**. A plugin must expose only the promoted set, which spans two of those bucket folders.
 - **Claude Code** — `.claude-plugin/plugin.json` accepts `skills` as an **array of explicit skill-directory paths**. We list the promoted skills one by one, exclude everything else with zero ambiguity, and add `.claude-plugin/marketplace.json` so the repo is its own single-plugin marketplace. Verified end to end: `claude plugin validate . --strict` passes, and `marketplace add` → `install` resolves all promoted skills.
 - **Codex** — `.codex-plugin/plugin.json` accepts `skills` only as a **single path string** (arrays are rejected with `missing or invalid plugin.json`), and Codex discovers `SKILL.md` files recursively under it. There is no way to name two bucket folders, or to curate a subset, from one path. Two escape hatches were tested and rejected:
  - Pointing at `./skills/` would also ship `deprecated/`, `in-progress/`, `personal/`, and `misc/` — retired, draft, and personal skills we deliberately don't promote.
  - A curated flat directory of **symlinks** into the buckets does not survive install: Codex copies the plugin tree into its cache and **drops symlinks**, so the skills arrive empty.
 The only robust ways to give Codex a single promoted-only path are (a) **restructure** so `skills/` contains only promoted skills (moving the non-promoted buckets out — a large blast radius across `CLAUDE.md`, `scripts/link-skills.sh`, the bucket READMEs, and the local dev workflow that relies on `in-progress/` and `personal/`), or (b) **commit duplicate copies** of promoted skills into a flat directory (a sync burden and a second source of truth). Both are structural decisions, not something to bundle into shipping the Claude plugin. This is very likely the original, half-remembered reason a plugin wasn't shipped earlier: the manifest formats didn't cleanly express a curated subset of a bucketed repo.
 ## Decision
 - Ship the **Claude Code plugin** now (`.claude-plugin/plugin.json` + `.claude-plugin/marketplace.json`), curated to the promoted set, as the headline v1.2 deliverable.
 - Keep **skills.sh** as the universal installer — it already serves Codex and other harnesses today, so no Codex user is left without an install path.
 - **Defer** the native Codex plugin until we decide between restructuring `skills/` to promoted-only vs. committing a generated flat copy. Revisit when Codex either supports a `skills` array / include-list or preserves symlinks on install.
 ## Invariants this creates
 - Every promoted skill has an entry in `.claude-plugin/plugin.json`'s `skills` array (this already stood as a `CLAUDE.md` rule; it now also gates the plugin's contents).
 - `.claude-plugin/plugin.json`'s `version` tracks `package.json`'s version — bump both together on release. Claude uses the plugin `version` to decide when installed users see an update.
 ## Update, 2026-08-05
 `mattpocock-skills` was accepted into **Claude Code's official marketplace** — configured name `claude-plugins-official`, source repo `anthropics/claude-plugins-official` — which every Claude Code install has by default. `claude plugins install mattpocock-skills` is now the documented route, and the `marketplace add` → `install` path above is superseded. The install wording lives in [.agents/install-block.md](../install-block.md).
 The official listing points at this repo's git URL and reads `.claude-plugin/plugin.json` directly, so it does not depend on `.claude-plugin/marketplace.json`. That file is retained only as a fallback for installing the repo directly (an unreleased commit, or a fork).
 Verified 2026-08-05, on Claude Code 2.1.222, against the live listing:
 - `claude plugins install mattpocock-skills` resolves with no marketplace added first, and reports `mattpocock-skills@claude-plugins-official`.
 - `claude plugin details mattpocock-skills` then reports version 1.2.0 and loads the promoted skills.
 - The listing's `source` is `{"source": "url", "url": "https://github.com/mattpocock/skills.git", "sha": …}` — the **sha is pinned**, so a release reaches installed users when that pin moves, not the moment we tag. At the time of writing the pin sits two commits behind `main`, which is why it lists 22 skills rather than the 24 in `plugin.json`.
 - The in-session `/plugin install mattpocock-skills` was **not** exercised — `/plugin` is unavailable in headless (`claude -p`) sessions. It runs the same resolver as the CLI, and the documented example form is `/plugin install <name>@claude-plugins-official`.
@@ -1,38 +0,0 @@
 # Flatten `skills/` to the promoted set, and adopt Agent Plugins 1.0
 [ADR 0002](./0002-ship-as-a-claude-code-plugin.md) shipped the Claude Code plugin and deferred everything else, because no other manifest format could express "a curated subset of a bucketed repo". It named two ways out: **(a)** restructure so `skills/` holds only promoted skills, or **(b)** commit a generated flat copy. This ADR takes **(a)**.
 ## What forced the decision
 The [Agent Plugins](https://agent-plugins.org/) standard published 1.0.0 on 6 August 2026 — Cursor, VS Code, GitHub Copilot, ChatGPT/Codex and Kiro are listed as clients. Its discovery rule is fixed and normative: a client reads the immediate children of `skills/`, "MUST NOT recursively search deeper descendants", and "`plugin.json` cannot override these locations". A bucketed `skills/engineering/<name>/` tree is therefore invisible to every conformant client.
 A spec was proposed that kept the buckets and generated a flat package under `plugins/mattpocock-skills/`, with a TypeScript builder, a validator, a promoted-skill allowlist, and a CI job failing on any rebuild diff. That is option (b) with extra steps: every skill in the repo twice, a rewrite step mangling `SKILL.md` frontmatter on the way through, and four new pieces of tooling — in a repo with no TypeScript at all. The whole apparatus exists only to work around the buckets.
 Removing the buckets removes the apparatus. We are conformant by layout instead of by tooling.
 ## Decision
 - `skills/` is **flat** and holds **exactly** the promoted set. Folder location is the only promotion mechanism.
 - Unpromoted skills move out of `skills/`: `in-progress/` → `drafts/`, `misc/` → `extras/`. `deprecated/` is deleted — it was empty, and a retired skill is deleted outright anyway.
 - Add `plugin.json` at the repo root: the canonical Agent Plugins 1.0 manifest. It lists no skills, because the standard's schema is closed and discovery is fixed.
 - Keep `.claude-plugin/plugin.json` for Claude Code, which does not implement the standard yet, and **delete its `skills` array** — Claude auto-discovers a flat `skills/`, and a 25-entry list is a second promotion mechanism that can drift.
 - No generated package, no build script, no duplicate skill tree, no drift CI, no TypeScript.
 ## What we deliberately did not do
 - **No `.codex-plugin/plugin.json` or `.cursor-plugin/plugin.json`.** Cursor documents dual support ("A plugin that conforms to the Agent Plugins specification loads in Cursor without changes"), and the Codex source reads the root `plugin.json` first, falling back to the legacy paths only if it is absent. The proposed spec was more conservative than either client.
 - **No new marketplace manifests.** Marketplaces are not part of Agent Plugins at all. Copilot reads `.claude-plugin/marketplace.json`, and Codex treats it as legacy-compatible, so the file we already have may serve all three. Add `.agents/plugins/marketplace.json` or `.github/plugin/marketplace.json` only where a real install fails.
 - **No rename to `mp`.** The public name stays `mattpocock-skills`. It is live in Claude Code's official marketplace with auto-update, so renaming would break every existing install for a shorter slash command.
 - **No flattening of `docs/`.** `docs/<category>/` is now the only place a skill's category lives. Flattening it would force 27 manual `githubSource` repoints on aihero.dev, one CLI call each, with no bulk command — for zero technical gain, because no client reads `docs/`.
 ## Consequences
 - Promoting a skill is `git mv drafts/<name> skills/<name>` plus a README entry and a docs page. Retiring one is `git mv` out. No manifest to keep in step.
 - `npx skills@latest add mattpocock/skills` stops offering all 35 skills. It walks `skills/` to depth 3 regardless of any manifest, so it has been offering every draft and extra all along; moving them out of `skills/` narrows the offer to the promoted 25. A fix we got for free.
 - `disable-model-invocation` is carried by 14 of the 25 promoted skills and is **not** part of the Agent Skills standard — there is an open, unanswered proposal to make it one. We keep it in one shared tree and rely on non-Claude clients ignoring an unknown frontmatter key. If a portable client ever fires a user-invoked skill on its own, that assumption is broken and this decision needs revisiting; the alternative is a second skill tree, which is the duplication we just removed.
 - Claude Code was **not** on the compatible-clients list when this was written, and neither the docs nor a 5,370-line changelog mention the standard. `.claude-plugin/plugin.json` therefore stays as the working manifest, not as legacy baggage. Delete it only once Claude reads the root manifest.
 - The standard is one day old and still labelled a Working Draft. The flatten is right regardless of what the standard does next; the root `plugin.json` is 20 lines and costs nothing if the spec moves.
 ## Status of ADR 0002
 Superseded in part. Its Claude-plugin decision stands. Its deferral of every other ecosystem is resolved here by taking the restructure it described.
@@ -1,61 +0,0 @@
 # The canonical install block
 One install story, one wording. `README.md`, `.changeset/*`, and every page under `docs/` must say **this** and nothing else. Change it here first, then propagate.
 `mattpocock-skills` is listed in **Claude Code's official marketplace** — configured name `claude-plugins-official`, source repo `anthropics/claude-plugins-official` — which every Claude Code install has out of the box. There is no marketplace to add first. Official Anthropic marketplaces have auto-update enabled by default ([discover-plugins](https://code.claude.com/docs/en/discover-plugins)), so "updates arrive automatically" is a true claim, not a hope.
 ## Claude Code — the plugin
 <canonical-block name="claude-code">
 ```bash
 claude plugins install mattpocock-skills
 ```
 Or, from inside a session:
 ```
 /plugin install mattpocock-skills
 ```
 It's in Claude Code's official marketplace, so there's nothing to add first, and updates arrive automatically.
 </canonical-block>
 ## Codex, and other agents — skills.sh
 The plugin is Claude Code only. Everywhere else, [skills.sh](https://skills.sh/mattpocock/skills) copies editable skill files into the project. Use the whole-set form on `README.md`:
 <canonical-block name="skills-sh-whole-set">
 ```bash
 npx skills@latest add mattpocock/skills
 ```
 Pick the skills you want, and which coding agents to install them on. **The installer lets you choose which skills to take — make sure `setup-matt-pocock-skills` is one of them.**
 </canonical-block>
 …and the single-skill form wherever one skill is named on its own. Note that **`docs/` pages are not a consumer of this block**: ai-hero renders the install widget above the body, so a page that writes the commands out duplicates it. See [writing-docs.md](./writing-docs.md).
 <canonical-block name="skills-sh-one-skill">
 ```bash
 npx skills@latest add mattpocock/skills --skill=<name>
 ```
 ```bash
 npx skills@latest update <name>
 ```
 </canonical-block>
 `skills@latest` is the pinned spelling in all three. The pages under `docs/` used to carry their own copy of these commands; those blocks are now deleted rather than corrected, because the site renders the install commands itself.
 ## The two routes are exclusive
 The plugin is a managed, read-only bundle you subscribe to. skills.sh writes files you own and edit. Installing both leaves the user with every skill twice — always say "pick one".
 ## Not the install story
 `.claude-plugin/marketplace.json` makes the repo its own single-plugin marketplace (`/plugin marketplace add mattpocock/skills`, then `/plugin install mattpocock-skills@mattpocock`). The official listing supersedes it. It is kept as a fallback for installing the repo directly — an unreleased commit, or a fork — and is **not** documented to users.
@@ -1,20 +0,0 @@
 # Model-invoked vs user-invoked
 Every `SKILL.md` in this repo is a skill. The one axis that splits them is **invocation** — who can reach it:
 - **User-invoked** — reachable **only by the human typing its name**. Set `disable-model-invocation: true` in the frontmatter (Claude Code) and `policy.allow_implicit_invocation: false` in `agents/openai.yaml` (Codex). The `description` is **human-facing**: a one-line summary read by a person browsing slash-commands. Strip trigger lists ("Use when the user says…").
 - **Model-invoked** — reachable by **model or user**. The default: omit `disable-model-invocation` and the `policy` block from `agents/openai.yaml`. The `description` is **model-facing** and keeps rich trigger phrasing ("Use when the user wants…, mentions…, asks for…") so auto-invocation fires. The test for whether a skill should stay model-invoked: _could the model usefully reach for this autonomously?_ (Reuse is the reason to extract a skill, not the test.)
 Each harness excludes a user-invoked skill from the model's reach in its own way, so nothing but the human can fire it — no other skill can. A user-invoked skill may invoke model-invoked skills, but it can never reach another user-invoked skill.
 Every skill also carries an `agents/openai.yaml` beside its `SKILL.md`. It holds Codex UI metadata — `interface.display_name` and `interface.short_description` for the skill picker — and, for user-invoked skills, the `policy.allow_implicit_invocation: false` that pairs with `disable-model-invocation`. Keep the two in sync: a skill is user-invoked in both harnesses or neither.
 Bucket `README.md`s and the top-level `README.md` group entries into **User-invoked** and **Model-invoked**.
 ## Dependencies between them
 Dependencies are expressed as **`/skill`-style prose invocation** ("Run the `/grilling` skill"), not deep `../other-skill/FILE.md` cross-references. Shared reference docs live inside the skill that owns them; other skills reach that material by invoking the skill, not by linking across folders.
 ## Passive vs active domain work
 Merely _reading_ `CONTEXT.md` for vocabulary is a one-line prose pointer, not the `domain-modeling` skill. Only the active build/sharpen discipline (challenge terms, edge-case scenarios, write ADRs, update `CONTEXT.md` inline) is `domain-modeling`.
@@ -1,96 +0,0 @@
 # Writing docs pages
 Every promoted skill — every folder under `skills/` — has a human-facing **docs page** at `docs/<category>/<skill-name>.md`, where `<category>` is `engineering` or `productivity`. The skills tree itself is flat, so this docs folder is the only place a skill's category lives. The page is published at `https://aihero.dev/skills-<skill-name>`; the URL is always `skills-<skill-name>` regardless of category, so the docs path is repo organisation only. The page is not the skill and not a copy of `SKILL.md`. Skills in `drafts/` and `extras/` ship no docs page.
 Most of these skills are **user-invoked**: the agent will never fire them for you, so *you* are the index that has to remember they exist and when to reach for them. That memory is **cognitive load**. The job of a docs page is to relieve it — to orient one reader around one skill so they can hold it in their head, know when to reach for it, and see where it sits in the system. The pages are collectively a distributed router; each is a node.
 Act whenever a promoted skill is added, renamed, or has its behaviour changed: create or re-sync its docs page. A rename moves the file too (`docs/<category>/<old>.md` → `docs/<category>/<new>.md`), because the published URL tracks the name; recategorising a skill moves its docs file to the other folder and changes nothing else. Skills in `drafts/` and `extras/` get no page. A skill promoted into `skills/` gains a page; one moved out of `skills/` loses it.
 Because these pages are published on `aihero.dev`, **every link is absolute** — never a repo-relative path. A link to another skill points at `https://aihero.dev/skills-<name>`; a link into the repo points at its full `https://github.com/mattpocock/skills/...` URL. A relative link that works in the repo breaks once published.
 There is no H1 — the published page takes its title from the slug.
 ## Page structure
 Fill the template below, keeping its order. The **fixed frame** (`## What it does`, `## When to reach for it`, `## Where it fits`) appears on every page. `## Prerequisites` and the free-form substance sections carry only what this particular skill needs; delete the rest.
 Four sections make a page worth reading: `What it does`, `When to reach for it`, `Common questions`, `It's working if`. The first two orient the reader; the last two are where the page stops summarising the skill and starts answering the reader's own situation. Each of the last two has a bar to clear, below — but treat a page that clears neither as unfinished, not as finished-and-short.
 **A page carries no install commands.** The ai-hero page template renders the install widget itself — a copy button, the single-skill command, the whole-set command, and the update line — above the body. A page that also writes them out shows the reader the same command twice, and the two copies drift: the hand-written pair on every page went stale against the widget beside it. Install wording is a property of the site, not of the page. If it needs changing, change it in ai-hero; the canonical wording lives in [the install block](./install-block.md).
 <page-template>
 ## What it does
 One or two plain-language paragraphs. Lead with the skill's one-sentence job, then state the **defining constraint** — the single fact that makes this skill behave differently from the obvious default (for `to-spec`: it does not interview the user again, it synthesises what is already known). Write it as a plain declarative sentence — never a labelled aside like "The defining constraint:" or "The key thing:"; the formula reads as filler. This line is the most valuable on the page; never omit it.
 ## When to reach for it
 How and when you reach for the skill — two beats, both effectively always present:
 - **Invocation mode.** State whether you type it or the agent fires it. A user-invoked skill: "You invoke this by typing `/<name>` — the agent won't reach for it on its own." A model-invoked skill: "Type `/<name>`, or the agent reaches for it automatically when a task fits."
 - **Trigger boundary.** The index entry: "reach for this when …". Where the skill is confusable with a sibling, add the other half — "for <X> instead, use [<sibling>](https://aihero.dev/skills-<sibling>)."
 ## Prerequisites
 Optional — include only when the skill needs something in place to be functional; omit the heading entirely otherwise. Covers: a **workspace it writes into** (a stateful skill like `grill-with-docs` writes `CONTEXT.md` and ADRs; `teach` builds a whole directory — say what it writes and where), **prior setup** (`triage`/`to-spec`/`to-tickets` need `setup-matt-pocock-skills` to have configured an issue tracker), or **repo-specific tooling**. A stateless skill that runs anywhere has no prerequisites — drop the section.
 ## <free-form middle>
 One to three short sections, in the skill's *own vocabulary*, that make it click — choose whatever headings fit the skill: the loop it runs, the artifact it produces, the fork it makes, the one anti-pattern it kills. There is no prescribed heading; the skills are too heterogeneous for one.
 The single non-negotiable: **surface the skill's leading word / defining idea** — `tight` feedback loop, `deep module`, throwaway-code-answers-a-question, red-green. It pays off twice: the reader learns what the skill *is*, and learns the word they'll later think with to *reach for* it.
 ## Common questions
 The questions readers really ask about this skill, each in bold with the answer in the lines beneath it — no sub-headings.
 An observed question always beats an invented one, so go and find them before you write any:
 - **The wiki.** If `~/repos/matt/personal-wiki` exists on this machine, it is the richest source there is. Its `wiki/audience/` area is organised around what the audience wants, discusses, and **is confused by** — read `wiki/index.md` first for the registry of pages, then the pages bearing on this skill. Every page carries `sources:` linkbacks to the original X, Discord, GitHub, and email threads; the wiki is a secondary source, so quote the asker's own question rather than the wiki's summary of it. Skip this bullet where the directory does not exist.
 - **This repo's issues.** `gh issue list --repo mattpocock/skills --search "<skill-name>" --state all`. A question filed twice is a question the page owes an answer to.
 - **`CHANGELOG.md`.** Anything renamed, moved, or behaviourally changed generates a "where did it go?" that the page has to answer.
 Where the hunt comes up thin, the section may also carry a question a reader would plainly ask — but **the count stays honest to the evidence**. A well-discussed skill earns six; an obscure one earns one or two, or none at all. Padding a thin skill out to match a rich one is how the section fills with questions nobody has, and an invented question teaches the reader nothing.
 Order them by how often each comes up, sharpest first, and say the unflattering thing where it is true — a very long grilling session usually means the scope was too big; a model asked to write its own skill produces something verbose. Omit the heading where there is nothing worth answering.
 ## It's working if
 A few bullets naming what the reader sees when the skill is doing its job. The bar on each is that the reader can check it without opening `SKILL.md` — a signal in their own work, or in the trace in front of them. "The document gets shorter as it gets better" passes; "the library section is byte-identical to `template.sh`" is a compliance check on the skill's internals wearing this section's name. Include it wherever the tells are crisp; omit the heading where they stay vague.
 ## Where it fits
 Always present. Situate the skill in the system in a sentence or two:
 - **Role.** Name it: a **chain step** (`grill-with-docs → to-spec → to-tickets → implement → code-review`), a **run-once setup** (`setup-matt-pocock-skills`), **periodic maintenance** (`improve-codebase-architecture`, "every few days"), or a **reach-for-it-anytime standalone** (`diagnosing-bugs`, `prototype`, `handoff`). A standalone's map is one honest sentence — far better than omitting the section.
 - **Neighbours.** The one or two siblings that matter, each with a because-clause, linked absolutely.
 - **The map.** Point to [ask-matt](https://aihero.dev/skills-ask-matt), the router over the whole set, so this page stays a node and never has to redraw the graph.
 </page-template>
 ## Conventions
 - Explain the **why**, not the process. The page orients and situates the skill; it never reproduces the `SKILL.md` steps or template dumps — a human choosing a tool does not need the runbook.
 - **Never name the author.** The page is a technical document, not a record of who said what. "Matt says", "Matt's own answer", "his position is", a quoted reply — all of it goes. A finding from the question hunt is worth keeping; its attribution is not. State the substance as a plain claim about the skill ("the fix is a direct instruction: …", "the split comes down to session count") and drop the frame. The reader is deciding whether to use a tool; an opinion carries the same weight either way, and an attributed one dates as soon as the position moves. Quoting a *user* stays fine — "one user reported …" is evidence about the skill in the wild, and stays anonymous.
 - Use the skill's **leading words** (_seam_, _deep module_, _tracer bullet_) so the page and the skill speak one language.
 - **Use the [AI Coding Dictionary](https://www.aihero.dev/ai-coding-dictionary)'s term where one exists, and link its first use on the page.** The dictionary is the house vocabulary for AI coding — _context window_, _subagent_, _harness_, _primary source_, _agent mode_. Prefer its word over a synonym you invent. Link the first occurrence of each term to `https://www.aihero.dev/ai-coding-dictionary/<slug>` (the slug is the term lowercased with non-alphanumerics as hyphens: _context window_ → `context-window`), and leave every later occurrence unlinked. Link only where the word carries the dictionary's sense — a domain *model*, background *context* or an auth *token* is a different word that happens to match. Never link inside a heading, a code span, or an existing link, and never link a word that names a skill in this repo rather than the concept. For the full term list, read `~/repos/ai/ai-coding-dictionary/dictionary/` if it exists on this machine — one file per term, the filename *is* the term — and otherwise [mattpocock/dictionary-of-ai-coding](https://github.com/mattpocock/dictionary-of-ai-coding), which is the source of truth either way.
 - **Branches go in a table or a list, never in a paragraph.** Where the page presents a choice — two artifacts the skill can produce, four situations that trigger it, five options at a boundary — the reader is scanning for the one row that matches their situation. A paragraph makes them read all of it to find out. A short markdown table (condition in the left column, what to do in the right) or a bulleted list gives it back in one glance. This applies wherever the branch appears, most often in `## When to reach for it` and the free-form middle.
 - Keep the page itself low-load. It is documentation *about* low-cognitive-load skills; furniture (spare headings, restated links) is the thing it is arguing against.
 ## Done when
 - The page exists at `docs/<bucket>/<name>.md`, and no stale page survives a rename or bucket move.
 - The page carries no source link and writes no install command of its own.
 - `## What it does` states the defining constraint, as plain prose rather than a labelled aside.
 - The page names no author and quotes no author — every claim stands on its own.
 - `## When to reach for it` states invocation mode and the trigger boundary.
 - `## Where it fits` names the role and links to `ask-matt`.
 - A prerequisite (workspace, prior setup, tooling) is stated where one exists, and the section is absent where none does.
 - The middle surfaces the leading word.
 - Every AI Coding Dictionary term the page uses is spelt the dictionary's way, and its first use — and only its first use — links to the dictionary entry.
 - Every multi-way branch is a table or a list, not a paragraph the reader has to read in full.
 - The hunt for real questions ran — the wiki, the issues, the changelog — and `## Common questions` is sized to what it found, not padded to match a richer skill's page.
 - Every `## It's working if` bullet is checkable without opening `SKILL.md`.
 - The sections appear in the template's order.
 - Every link is absolute, and every one resolves.
@@ -1,8 +0,0 @@
 # Changesets
 Hello and welcome! This folder has been automatically generated by `@changesets/cli`, a build tool that works
 with multi-package repos, or single-package repos to help you version and publish your code. You can
 find the full documentation for it [in our repository](https://github.com/changesets/changesets).
 We have a quick list of common questions to get you started engaging with this project in
 [our documentation](https://github.com/changesets/changesets/blob/main/docs/common-questions.md).
@@ -1,15 +0,0 @@
 {
  "$schema": "https://unpkg.com/@changesets/config@3.1.4/schema.json",
  "changelog": [
    "@changesets/changelog-github",
    { "repo": "mattpocock/skills" }
  ],
  "commit": false,
  "privatePackages": { "version": true, "tag": true },
  "fixed": [],
  "linked": [],
  "access": "restricted",
  "baseBranch": "main",
  "updateInternalDependencies": "patch",
  "ignore": []
 }
@@ -1,15 +0,0 @@
 ---
 "mattpocock-skills": patch
 ---
 Flatten `skills/` so the repo conforms to the [Agent Plugins 1.0](https://agent-plugins.org/) standard, which reads only the immediate children of `skills/` and can't be pointed anywhere else.
 Nothing changes for installed users: same plugin name, same 25 skills, same slash commands, same install commands. **Source paths moved**, so anything pinning a `skills/engineering/…` or `skills/productivity/…` path needs updating.
 - Promoted skills are now flat at `skills/<name>/`. Which folder a skill lives in is the only thing that decides whether it ships.
 - `in-progress/` → `drafts/`, `misc/` → `extras/`, both outside `skills/`. `deprecated/` is deleted; it was empty.
 - New `plugin.json` at the repo root — the Agent Plugins 1.0 manifest, read by Codex, Cursor and Copilot.
 - `.claude-plugin/plugin.json` keeps serving Claude Code, which doesn't implement the standard yet, and loses its 25-entry `skills` array.
 - `npx skills@latest add mattpocock/skills` now offers the promoted 25 only. It was quietly offering all 35, drafts included.
 The reasoning, and the generated-package approach we rejected, are in [ADR 0003](https://github.com/mattpocock/skills/blob/main/.agents/adr/0003-flat-skills-tree-for-agent-plugins.md).
@@ -1,23 +0,0 @@
 {
  "name": "mattpocock",
  "owner": {
    "name": "Matt Pocock",
    "url": "https://www.aihero.dev"
  },
  "description": "Matt Pocock's skills for real engineering, as an installable Claude Code plugin.",
  "plugins": [
    {
      "name": "mattpocock-skills",
      "source": "./",
      "description": "Matt Pocock's agent skills for real engineering — grilling, spec/ticket flows, TDD, code review, domain modelling and more.",
      "category": "engineering",
      "keywords": [
        "engineering",
        "skills",
        "tdd",
        "code-review",
        "grilling"
      ]
    }
  ]
 }
@@ -1,21 +0,0 @@
 {
  "name": "mattpocock-skills",
  "version": "1.2.3",
  "description": "Matt Pocock's agent skills for real engineering — grilling, spec/ticket flows, TDD, code review, domain modelling and more. Plug-and-play, not vibe coding.",
  "author": {
    "name": "Matt Pocock",
    "url": "https://www.aihero.dev"
  },
  "homepage": "https://www.aihero.dev/s/skills-newsletter",
  "repository": "https://github.com/mattpocock/skills",
  "license": "MIT",
  "keywords": [
    "engineering",
    "skills",
    "tdd",
    "code-review",
    "grilling",
    "domain-modeling",
    "productivity"
  ]
 }
@@ -1,40 +0,0 @@
 name: Release
 on:
  push:
    branches:
      - main
 concurrency: ${{ github.workflow }}-${{ github.ref }}
 jobs:
  release:
    name: Version
    runs-on: ubuntu-latest
    permissions:
      contents: write
      pull-requests: write
    steps:
      - name: Checkout
        uses: actions/checkout@v4
      - name: Setup Node.js
        uses: actions/setup-node@v4
        with:
          node-version: 22
      - name: Install dependencies
        run: npm ci
      - name: Check plugin manifest versions
        run: npm run check-plugin-version
      - name: Create Version Pull Request
        uses: changesets/action@v1
        with:
          version: npm run version
          publish: npx changeset tag
          commit: "chore: version skills"
          title: "chore: version skills"
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
@@ -1 +0,0 @@
 node_modules
@@ -1,25 +0,0 @@
 # Issue tracker integrations are limited to mainstream tools
 `setup-matt-pocock-skills` only offers first-class support for **mainstream** issue trackers. Requests to add support for niche, new, or single-vendor experimental trackers are out of scope.
 ## Why this is out of scope
 Every issue-tracker backend hard-codes a CLI shape into the skills (commands, flags, output parsing). Each new backend is permanent maintenance surface — it has to keep working as the tool's CLI evolves, and it has to keep being tested against `/to-spec`, `/to-tickets`, `/triage`, and friends. That cost is only worth paying for trackers a meaningful fraction of users actually have.
 "Mainstream" is a judgment call, not a numeric bar:
 - GitHub, GitLab, and Backlog.md are the kind of tools we'd consider mainstream — broadly known, widely used, well past the experimental phase.
 - A brand-new agent-focused tool with a few hundred GitHub stars is not, no matter how interesting the design.
 Stars, age, and download counts are useful signals when making the call but none of them is the rule. The rule is: would a typical engineer recognise this tool and have plausibly chosen it for their team?
 The escape hatches for non-mainstream trackers already exist:
 - `local markdown` for lightweight in-repo tracking.
 - `other/custom` for users who want to wire something up themselves.
 Neither requires the core skills to know about the specific tool.
 ## Prior requests
 - #99 — "Add dex as an issue tracker backend" (dex was ~3 months old and ~300 stars at the time of the request)
@@ -1,18 +0,0 @@
 # Hard limits on the number of questions during grilling
 The `/grill-me` skill (and grilling sessions inside other skills) does not enforce a maximum number of questions. Requests to add a configurable cap or hard ceiling are out of scope.
 ## Why this is out of scope
 Grilling is intentionally open-ended. The point is to keep digging until each branch of the decision tree is resolved — some plans need three questions, some need fifty. A fixed cap would either cut off useful exploration on hard problems or feel arbitrary on easy ones.
 If a session feels too long, the right escape hatches already exist:
 - The user can stop the session at any time and accept the current state of the plan.
 - The user can tell the model to wrap up, summarise, and move on — natural-language steering is the intended control surface, not a numeric limit.
 Adding a hard cap would also conflate two different failure modes: a model that asks too many questions because the plan is genuinely under-specified (working as intended) vs. a model that asks redundant or low-value questions (a prompt-quality issue, not a quantity issue). The fix for the latter belongs in the skill prompt, not in a counter.
 ## Prior requests
 - #44 — "Codex just asked me 200 questions"
@@ -1,15 +0,0 @@
 # Verify/Check Mode for `setup-matt-pocock-skills`
 This project will not add a dedicated verify/check mode (or a separate verify skill) for `setup-matt-pocock-skills`.
 ## Why this is out of scope
 A second skill — or a `--verify` flag — for checking whether `docs/agents/*.md` artifacts still match the seed-template schema would duplicate work the existing setup skill already handles in conversation.
 The intended workflow is: **run `/setup-matt-pocock-skills` and tell it to verify your current setup.** The skill is prompt-driven, so the maintainer can scope it to a verification pass ("don't rewrite anything, just check my existing files against the current seed templates and report drift") without needing a separate code path. Adding a flag or a sibling skill would split the surface area of a feature that's already expressible through the natural-language entry point.
 Keeping configuration management to a single skill also avoids the maintenance cost of two skills drifting from each other when seed templates evolve.
 ## Prior requests
 - #106 — Feature request: verify/check mode for setup-matt-pocock-skills
@@ -1 +0,0 @@
 CLAUDE.md
@@ -1,270 +0,0 @@
 # mattpocock-skills
 ## 1.2.3
 ### Patch Changes
 - [#779](https://github.com/mattpocock/skills/pull/779) [`efce423`](https://github.com/mattpocock/skills/commit/efce423018fc6468a3239621f1c1bcaacc723801) Thanks [@mattpocock](https://github.com/mattpocock)! - Make `diagnosing-bugs` redact secrets.
  - Add a **Redact** section to `SKILL.md`. The skill has the agent show commands, outputs and captured artifacts; the section makes redaction the first move on each — write `<REDACTED>`, build loops against env vars so the credential stays in the environment, and quote only the signal-carrying lines of a captured artifact.
  - The Phase 1 completion criterion said "paste the invocation and its output". It now says show it redacted, and Phase 1 asks the user for a **redacted** captured artifact.
  - Note in `scripts/hitl-loop.template.sh` that `capture` prints its value back to the terminal, so it takes observations while signing in stays a `step`.
 - [#781](https://github.com/mattpocock/skills/pull/781) [`14bfbbd`](https://github.com/mattpocock/skills/commit/14bfbbd8654a8d2910299e1a004c19c1979687d8) Thanks [@mattpocock](https://github.com/mattpocock)! - Drop Claude Code's tool and agent-type names from the subagent-dispatch instructions in `code-review`, `codebase-design`, and `improve-codebase-architecture`, so the step is followable on Codex and other harnesses.
 - [#783](https://github.com/mattpocock/skills/pull/783) [`c0fd1e9`](https://github.com/mattpocock/skills/commit/c0fd1e973e040347d424e09934099f1bd6c2dee0) Thanks [@mattpocock](https://github.com/mattpocock)! - wizard: remove the time estimate. The template drops `TOTAL_MINUTES` and the time-remaining display, `stage` takes a name only, and progress is counted in stages.
 ## 1.2.2
 ### Patch Changes
 - [#766](https://github.com/mattpocock/skills/pull/766) [`4aaccb5`](https://github.com/mattpocock/skills/commit/4aaccb58d40559d7e3c59a029b2290ae5ba538de) Thanks [@mattpocock](https://github.com/mattpocock)! - Make `writing-for-agents` model-invokable in Codex again.
  - Drop `policy.allow_implicit_invocation: false` from `agents/openai.yaml`. Codex filtered the skill out of the model-visible skills list, so its description could not trigger it — only an explicit `$writing-for-agents` mention worked.
  - Update the stale `interface.display_name` and `interface.short_description`, which still named the old `writing-great-skills` skill.
  - Move the skill from the **User-invoked** list to the **Model-invoked** list in `README.md` and `skills/productivity/README.md`.
 ## 1.2.0
 ### Minor Changes
 - [#551](https://github.com/mattpocock/skills/pull/551) [`697d4ce`](https://github.com/mattpocock/skills/commit/697d4ce9742da558fd1ba6697c8e9775e2e302dd) Thanks [@mattpocock](https://github.com/mattpocock)! - Add Codex metadata alongside each skill's Claude Code frontmatter so the set works in both harnesses without generated copies.
  - Add an `agents/openai.yaml` beside every `SKILL.md` with Codex UI metadata (`interface.display_name`, `interface.short_description`).
  - Mark every user-invoked skill with `policy.allow_implicit_invocation: false`, the Codex analog of `disable-model-invocation: true`, so Codex excludes it from implicit invocation while explicit `$skill` invocation still works.
  - Document the dual-harness invocation model in `.agents/invocation.md`, `CLAUDE.md`, and the promoted-bucket READMEs.
  - Add `AGENTS.md` as a symlink to `CLAUDE.md` so Codex reads the same repo instructions.
 - [#593](https://github.com/mattpocock/skills/pull/593) [`0f2bdbd`](https://github.com/mattpocock/skills/commit/0f2bdbdb06220d2df3718b8f0483157c6c8a8600) Thanks [@mattpocock](https://github.com/mattpocock)! - Graduate **`to-questionnaire`** out of `in-progress/` into the **Productivity** bucket, so it ships in the plugin. It turns a decision you can't answer alone into a Markdown questionnaire for the one person who can — filled in async, or worked through together in a meeting.
  Its defining move is that it grills you about the **send**, not the subject: a normal grilling session interrogates the topic, which is exactly what you can't answer here, so the interview asks only who the questionnaire is going to and what you need back, then aims every question at the gap between the two.
  Now wired as a promoted skill — plugin entry, top-level + Productivity READMEs under **User-invoked**, a docs page at `docs/productivity/to-questionnaire.md`, and a Standalone route in `ask-matt` framing it as the inverse of `/grill-me` (mine someone else, not yourself).
 - [#680](https://github.com/mattpocock/skills/pull/680) [`b3376f8`](https://github.com/mattpocock/skills/commit/b3376f8d39848dd08572ec2667da4739a67c8c04) Thanks [@mattpocock](https://github.com/mattpocock)! - Graduate **`wizard`** out of `in-progress/` into the **Engineering** bucket, so it ships in the plugin — and make it model-invoked. It generates an interactive bash script that walks a human through a manual procedure — third-party setup, a one-off migration, an A→B state transition — opening each URL, saying what to click, capturing the values, and writing them into `.env` files and GitHub Actions secrets.
  The delightful UX is pre-solved by the bundled `template.sh` (progress with time-remaining, confirmation gates, cross-platform URL opening including WSL, hidden secret entry, idempotent `.env` upserts, `gh secret`/`gh variable` writes with graceful degradation, closing skip summary). Everything above the `STAGES` marker is a fixed library that's never hand-edited — the skill's job is only to scope the procedure and author its **stages**.
  Engineering rather than Productivity: it reads `.env*`, `docker-compose*`, framework config and every `secrets.*`/`vars.*` reference in `.github/workflows/` to scope itself, writes CI secrets, and verifies its output with `bash -n` and `shellcheck`.
  Because it is model-invoked, the agent can reach for it the moment it hits a step only a human can perform, instead of dumping numbered instructions into the chat and hoping you follow them. Typing `/wizard` works exactly as before — model-invocation only ever _adds_ the agent's reach. The description is written as the pointer that decides when it fires: what it produces, four trigger branches (provisioning infrastructure, setting up credentials or CI secrets, walking an unfamiliar third-party dashboard, a one-off migration or cutover), and an explicit non-trigger — don't invoke it for steps the agent can perform itself. Work an agent can do, an agent should do; the wizard is for the clicks, approvals and dashboard trips you would not hand to one. The stage-list confirmation before a line is written now doubles as the proposal when the agent fires it mid-build.
  Now wired as a promoted skill — plugin entry, top-level + Engineering READMEs under **Model-invoked**, a docs page at `docs/engineering/wizard.md`, and a Standalone route in `ask-matt` for the steps only a human can take. Model-invocation also puts it out of the reach of [#693](https://github.com/mattpocock/skills/issues/693), which drops user-invoked skills from the listing on Claude's desktop and web surfaces.
 - [#763](https://github.com/mattpocock/skills/pull/763) [`77d207e`](https://github.com/mattpocock/skills/commit/77d207ef03219cc603e2832e1159cbdd1c91818e) Thanks [@mattpocock](https://github.com/mattpocock)! - Reshape the **`prototype`** skill around two ideas: the demo is **a single shareable HTML file**, and the prototype is **a primary source**.
  The logic branch now produces one self-contained file (plain HTML/CSS/JS, no build, no server) instead of a terminal app — a non-developer can open it by double-click and drive it in their own domain language: a labelled state panel, always-available free-play buttons, and a set of tabbed **guided walkthroughs**, each a scenario with the ordered buttons to press underneath it. The portable pure-logic module still lifts into the real code; the HTML shell is the throwaway.
  Throwaway no longer means deleted. Rather than being removed once it has answered its question, the prototype is captured as runnable evidence on a throwaway branch (`prototype/<name>`) out of main, with a context pointer to it left on the implementation issue — so the main branch keeps only the validated decision while the exploration stays findable. The answer (verdict + question) is still captured durably in an issue/ADR/commit.
 - [#536](https://github.com/mattpocock/skills/pull/536) [`42a5b70`](https://github.com/mattpocock/skills/commit/42a5b70fcacc7baff1977b13f3919fb2f63af14e) Thanks [@mattpocock](https://github.com/mattpocock)! - Ship the skill set as a native **Claude Code plugin**, listed in Claude Code's official marketplace. You can now subscribe to the promoted skills as a managed, read-only bundle instead of copying editable files:
  ```bash
  claude plugins install mattpocock-skills
  ```
  Or, from inside a session:
  ```
  /plugin install mattpocock-skills
  ```
  There is no marketplace to add first — the official marketplace is configured by default.
  `.claude-plugin/plugin.json` carries the full plugin metadata (version, description, author, license, keywords) and the explicit list of promoted skills. `skills.sh` remains the universal installer (and the path for Codex and other harnesses today); a native Codex plugin is deferred — see `.agents/adr/0002-ship-as-a-claude-code-plugin.md` for why.
 - [#751](https://github.com/mattpocock/skills/pull/751) [`355fa74`](https://github.com/mattpocock/skills/commit/355fa7420b418af838998f7ec4365ceda1c8dfcc) Thanks [@mattpocock](https://github.com/mattpocock)! - Add **`wait-what`** — a one-word corrective for model verbosity. Type it the moment a message doesn't land, and the agent re-pitches it: a little context, ASD-STE100 Simplified Technical English, and the ubiquitous language from your `CONTEXT.md`. User-invoked, three lines long.
  The mechanism is the name. Concision skills fail by growing — a 400-line skill still leaves the model verbose — so this one is a single precise leading word and nothing else. Names that describe the _output_ (`/tldr`, `/no-fluff`) make the model clip words and lose you further; naming the _listener's_ state asks for both halves at once, fewer words **and** the context you were missing. It also reuses the leading words already in your global `CLAUDE.md`, so the skill, `CLAUDE.md` and every `CONTEXT.md` reach for the same tokens.
  It repairs one message; it doesn't prevent the next one. The cure for jargon is a shared language built upfront with `/grill-with-docs`; this is what you reach for when you don't have one yet.
 - [#763](https://github.com/mattpocock/skills/pull/763) [`77d207e`](https://github.com/mattpocock/skills/commit/77d207ef03219cc603e2832e1159cbdd1c91818e) Thanks [@mattpocock](https://github.com/mattpocock)! - Name the `/wayfinder` unit a **decision ticket**, and burn research tickets down with subagents.
  People kept reading a wayfinder ticket as an ordinary _implementation_ ticket — a slice of a build to execute — when wayfinder uses them as **decision tickets**: questions whose resolution is a decision. The skill description and its opening line now introduce the term (and say what makes it one), with the `ask-matt` / engineering README blurbs and the docs page matching — while "ticket" stays the everyday word once the term is established. `CONTEXT.md` records **Decision ticket** as a domain term, so the "avoid: ticket" guidance no longer contradicts wayfinder's deliberate use of the word.
  Research tickets are no longer parked for a separately-launched session. Research stays a real ticket type — it's a genuine shared blocker that downstream decisions hang on, and that dependency is exactly what the frontier's blocking edges exist to render. What changes is how it's resolved: because research is AFK, charting doesn't stop and read it. After creating the tickets, the charting session fires a `/research` subagent for each research ticket to burn it down in parallel, capturing the findings on a throwaway `research/<name>` branch with a context pointer. Research tickets are the one exception to _one ticket per session_.
 - [#763](https://github.com/mattpocock/skills/pull/763) [`77d207e`](https://github.com/mattpocock/skills/commit/77d207ef03219cc603e2832e1159cbdd1c91818e) Thanks [@mattpocock](https://github.com/mattpocock)! - **Breaking:** rename **`writing-great-skills`** → **`writing-for-agents`**, restructure it, and add a new leading word.
  The reference now covers any document an agent consumes — skills, `AGENTS.md` / `CLAUDE.md`, docs reached by a pointer — not just skills. `GLOSSARY.md` is merged into `SKILL.md` (one authoritative treatment per term; the `_Avoid_` synonym lists and the standalone Predictability definition are gone); the skill-only mechanics (frontmatter, model- vs user-invoked, router skills, the invocation cut of splitting) are disclosed to a new `SKILL-MECHANICS.md`. The skill is now **model-invoked**: it fires when creating or editing skills or modifying `AGENTS.md`/`CLAUDE.md`. `ask-matt`'s pointer updated. Reinstall under the new name; the old name is gone (no alias).
  The pruning section gains **cache**. Single source of truth now reaches past the document into the environment — `package.json` scripts, config files, directory layout, `--help` output are themselves authoritative, so a doc that restates them is a cache of a lookup, earning its load only when the lookup is expensive. The positive target: cache what the agent cannot find by looking (unwritten conventions, the reason behind a choice, gotchas no config confesses), and leave one-file, one-command lookups to the environment, where they cannot go stale.
 - [#533](https://github.com/mattpocock/skills/pull/533) [`45afd80`](https://github.com/mattpocock/skills/commit/45afd8074a8b7de5fe073845d080fa9dd6c429fa) Thanks [@mattpocock](https://github.com/mattpocock)! - Add a YAGNI scoping filter to the **`improve-codebase-architecture`** skill's Explore step. Instead of scanning the whole repo evenly, it now scopes to where change is actually landing: if you name a direction it takes it, otherwise it reads the last ~20 commit messages to bias exploration toward actively-developed paths. A deepening opportunity in code nobody touches is a refactor you'll never cash in — the leverage only pays off where you keep editing — so the report stops tidying dormant corners of the repo.
 ### Patch Changes
 - [#763](https://github.com/mattpocock/skills/pull/763) [`77d207e`](https://github.com/mattpocock/skills/commit/77d207ef03219cc603e2832e1159cbdd1c91818e) Thanks [@mattpocock](https://github.com/mattpocock)! - Sharpen `/ask-matt` — the router now covers phase boundaries, the two wayfinder mistakes, and two skills it never mentioned.
  **Phase boundaries.** A **phase** is a chunk of work inside a session — the grilling, the implementation, the QA — and the boundary between two of them is where you decide what to do with the context you've built. The two-bullet `Crossing sessions` section is replaced by a decision tree carrying all five options in order (**continue**, `/clear`, `/handoff`, **subagent**, `/compact`), with the reasoning disclosed in a new `PHASE-BOUNDARIES.md`. Three fixes come with it:
  - **`/handoff` was oversold.** It read as the general bridge between context windows. It's narrow: you need it only when something has to _travel_ — a new harness, a new directory, a colleague, or a side task forked mid-phase. What it buys is portability.
  - **`/compact` is the default, not the first reach.** It sits at the bottom of the tree, after the four cheaper or more precise questions above it. Starting there produces a session that's confidently wrong about whatever the summary flattened.
  - **Two branches were missing entirely.** **Continue** is the one to rule out first — it's the only move that keeps the conversation as a primary source rather than a summary of one — and a **subagent** handles anything scoped tightly enough to run AFK.
  Context hygiene's escape hatch now says `/compact` rather than `/handoff` (same harness, same directory, at a boundary — the handoff clause doesn't apply), and the smart zone figure is updated from ~120k to ~150k tokens.
  **Wayfinder routing.** The two mistakes people most often make with the heaviest, most cognitively demanding flow:
  - **Over-reaching for it.** It's slower and denser than a single grill, so it's flagged as the heaviest flow and reserved for the idea that genuinely won't fit one session — a well-scoped feature belongs on `/grill-with-docs`, not here.
  - **Losing the way at the handoff.** When the map clears, wayfinder hands off, it doesn't build: merge onto the main flow at `/to-spec` (which collapses the map's linked decisions into a buildable plan) rather than looping the map straight into `/implement`. Straight-to-`/implement` is only for efforts that turned out genuinely small.
  **Missing routes.** `/grilling` and `/resolving-merge-conflicts` were absent from the router altogether and are now in it, and `grill-me` splits from `grill-with-docs` on whether you are in a working directory.
 - [#502](https://github.com/mattpocock/skills/pull/502) [`44eed54`](https://github.com/mattpocock/skills/commit/44eed545186ffd0263e8004867750b80cfddd215) Thanks [@mattpocock](https://github.com/mattpocock)! - Make `/setup-matt-pocock-skills` friendlier and align the local-markdown tracker with the current spec.
  - **Triage labels** are now asked about only when the `triage` skill is installed, and then as a single recommended-yes question ("keep the default triage labels?") instead of an override interrogation. When `triage` isn't installed, the section — and `docs/agents/triage-labels.md` — are skipped.
  - **External PRs as a request surface** is no longer a setup question. The GitHub/GitLab templates still carry the flag, defaulted off; a user can flip it in `docs/agents/issue-tracker.md` later.
  - **Domain docs** default to single-context without asking; multi-context is only offered when the repo shows monorepo signals.
  - **Local-markdown tickets** are now one file per ticket under `.scratch/<feature>/issues/<NN>-<slug>.md` — never a single combined `tickets.md`. `/to-tickets` and the local issue-tracker template now agree, and the spec file is `spec.md` (not `PRD.md`) to match `/to-spec`.
  Docs pages for `setup-matt-pocock-skills` and `to-tickets` re-synced.
 - [#532](https://github.com/mattpocock/skills/pull/532) [`170ad48`](https://github.com/mattpocock/skills/commit/170ad48655825783d0193e850e31a9aac957bb95) Thanks [@mattpocock](https://github.com/mattpocock)! - Reword **`grilling`** for general use. Its description and body no longer scope the interview to a software plan: "this plan" → "this", "enact the plan" → "act on it", and "exploring the codebase" → "exploring the environment". The technique is unchanged; it now reads as a stress-test of any plan, decision, or idea.
 - [#593](https://github.com/mattpocock/skills/pull/593) [`a4b2009`](https://github.com/mattpocock/skills/commit/a4b2009a1a3ac9575506c10b4c84f08f9bba7a38) Thanks [@mattpocock](https://github.com/mattpocock)! - Rework **`grilling`** from one-question-at-a-time to round-by-round. It now maps the decision tree and asks the whole **frontier** — every question whose prerequisites are already settled — in a single numbered round, then recomputes the frontier from the user's answers and asks the next round. Same 13 questions land in ~3 rounds instead of 13. Facts the environment can answer are dispatched to background sub-agents so research never blocks the round: only questions downstream of a running exploration wait for it. The session ends when the frontier is empty.
  Every question in a round is emitted in one fixed shape — `❓ **Q1** - **<title>**`, then the body (prose or multiple choices), then the recommendation on its own `➡️` line. A round reads as a scannable numbered list with each recommendation visually separated from the question, so you can answer by number instead of quoting questions back.
  `grill-me`, `grill-with-docs` and `triage` run the frontier a round at a time as well — `triage`'s grill step and `grilling`'s Codex `short_description` now say so instead of describing the old rhythm. The opt-out for one-question-at-a-time (a line in your global `CLAUDE.md`) is unchanged.
 - [#752](https://github.com/mattpocock/skills/pull/752) [`c66bdee`](https://github.com/mattpocock/skills/commit/c66bdeeee002d81e3f8b21403c07f9a0d7bea6da) Thanks [@mattpocock](https://github.com/mattpocock)! - Remove six skills from the repo. None of them was in the Claude Code plugin, but all six were installable through [skills.sh](https://skills.sh/mattpocock/skills), which serves every skill in the repo — so this is what leaves that listing, and where each one went.
  Four retired skills, each already absorbed by a skill that does the job better:
  - **`ubiquitous-language`** → **`/domain-modeling`**, which builds and maintains the whole domain model rather than dumping a glossary from one conversation.
  - **`design-an-interface`** → **`/codebase-design`**. Nothing is lost: the "design it twice" technique — parallel sub-agents generating radically different designs, from Ousterhout — ships inside that skill as `DESIGN-IT-TWICE.md`.
  - **`qa`** → **`/triage`** and **`/to-tickets`**.
  - **`request-refactor-plan`** → **`/to-spec`** and **`/improve-codebase-architecture`**.
  And two that were only ever mine — tied to my own machine and never meant for anyone else. The `personal/` bucket goes with them:
  - **`edit-article`**
  - **`obsidian-vault`**, which hardcoded a path to my own Obsidian vault.
  `skills/deprecated/` stays as a bucket, now empty. `skills/in-progress/` is unchanged and is now described for what it actually is: a beta channel, published on purpose, installable one skill at a time through skills.sh.
 - [#734](https://github.com/mattpocock/skills/pull/734) [`a2f9333`](https://github.com/mattpocock/skills/commit/a2f9333669ff53db762c87ecda5a15442060a3be) Thanks [@mattpocock](https://github.com/mattpocock)! - Finish the `to-prd` → `to-spec` rename: "spec" is now the only term in the shipped text.
  - **`to-spec`** no longer opens with "you may know this document as a PRD" — the parenthetical is dropped from the skill and its docs page. The local-markdown tracker template drops the same hedge.
  - **`code-review`** talks about the originating issue/spec rather than issue/PRD, in its frontmatter description, its two-axis summary, and the spec-source search order. Both READMEs re-synced.
  - **The GitHub and GitLab tracker templates** now say "Issues and specs for this repo live as GitHub/GitLab issues" — they had been left on "PRDs" when the local template was updated, so the stale term propagated into every repo they were written into.
  - **`docs/engineering/research.md`** pointed at `https://aihero.dev/skills-to-prd`, a dead slug for the renamed skill; it now links `to-spec` like the other nineteen docs pages do.
  The CHANGELOG and existing changesets still name PRDs where they document the rename itself, which is correct.
 ## 1.1.0
 ### Minor Changes
 - [#406](https://github.com/mattpocock/skills/pull/406) [`930a450`](https://github.com/mattpocock/skills/commit/930a450089f77a49af09001d955db8452a4b867d) Thanks [@mattpocock](https://github.com/mattpocock)! - Bring the **`ask-matt`** router up to date with the full skill set. It now maps five skills it was missing: **`tdd`** (woven into the main flow as the red-green engine `implement` drives), **`diagnosing-bugs`** (a new "Something's broken" on-ramp — there was previously no route for a bug), **`domain-modeling`** and **`codebase-design`** (a new "Vocabulary underneath" section), and **`grilling`** (the shared interview primitive). `prototype` is fleshed out as a standalone and the description broadens from "user-invoked skills" to "the skills". A maintenance rule is added to `CLAUDE.md` so any future skill add/rename/remove or flow change triggers an `ask-matt` re-check, beside the existing docs-page re-sync rule.
 - [#464](https://github.com/mattpocock/skills/pull/464) [`639df6e`](https://github.com/mattpocock/skills/commit/639df6e7386dfddc739b2aecdeff37a876f2483b) Thanks [@mattpocock](https://github.com/mattpocock)! - Promote and harden **`code-review`**. The in-progress **`review`** skill is renamed to **`code-review`** and moved from `in-progress/` into `engineering/`: it now ships in the plugin, is listed in the top-level and Engineering READMEs (Model-invoked), and has a docs page at `docs/engineering/code-review.md`. The `/implement` skill and docs point at `/code-review`.
  It also gains an always-on **Fowler smell baseline** on its Standards axis — a curated ~12 high-signal "Bad Smells in Code" (Mysterious Name, Duplicated Code, Feature Envy, Data Clumps, Primitive Obsession, Repeated Switches, Shotgun Surgery, Divergent Change, Speculative Generality, Message Chains, Middle Man, Refused Bequest) inlined into `SKILL.md` as a fixed baseline alongside whatever the repo documents, not a new third axis. Two binding rules keep it safe: a documented repo standard overrides the baseline, and every smell is reported as a judgement call, never a hard violation.
 - [#464](https://github.com/mattpocock/skills/pull/464) [`639df6e`](https://github.com/mattpocock/skills/commit/639df6e7386dfddc739b2aecdeff37a876f2483b) Thanks [@mattpocock](https://github.com/mattpocock)! - Sharpen **`grilling`** on two fronts.
  **A confirmation gate.** The agent won't enact the plan until you confirm the shared understanding has been reached — turning the skill's existing "shared understanding" completion criterion into an explicit stop-gate. The `description` also recruits the pretrained **`grill`** leading word ("Grill the user relentlessly") to sharpen invocation, and the docs page is re-synced.
  **Facts vs. decisions.** Grilling now splits _facts_ (look them up — explore the codebase) from _decisions_ (put each one to the human and wait for their answer). The old blanket line — "if a question can be answered by exploring the codebase, explore the codebase instead" — was written for the live-human case, but once another skill runs grilling inside a resolve-the-ticket frame it read as license to answer _decisions_ autonomously too. Separating the two keeps a grilling agent from racing ahead and answering its own questions.
 - [#463](https://github.com/mattpocock/skills/pull/463) [`af6d692`](https://github.com/mattpocock/skills/commit/af6d6922c3e2b5288eef155346cbe319e4ed3bd0) Thanks [@mattpocock](https://github.com/mattpocock)! - Add two adjacent Steering failure modes to **`writing-great-skills`**, both about how language you think of as "off" still steers the agent. **Negation** — the _elephant_ — is steering by prohibition: naming what _not_ to do drags the forbidden behaviour into context and makes it _more_ available, not less (_don't think of an elephant_), so the cure is to prompt the **positive**. **Negative Space** — the void — is blindness to the steering done by what you leave _out_: every decision a skill declines is delegated to the agent's priors rather than left neutral, so the cure is to read a draft for its silences and decide each omission deliberately (fill it, or leave it open as a real **branch**). Kept as two entries, not one — they carry different diagnostics and different cures — each a full `GLOSSARY.md` entry plus a `SKILL.md` failure-mode bullet, matching how every other failure mode is carried.
 - [`850873c`](https://github.com/mattpocock/skills/commit/850873cd73d5f81826ebf512ad35d2b1e113001f) Thanks [@mattpocock](https://github.com/mattpocock)! - Make the **`prototype`** skill model-invoked, so the agent can reach for it autonomously (and other skills can too). Its description is rewritten around the leading word _prototype_ — throwaway code that answers a design question — with one trigger per branch (state/logic sanity-check, or UI exploration).
 - [#409](https://github.com/mattpocock/skills/pull/409) [`0d74d01`](https://github.com/mattpocock/skills/commit/0d74d01cbc64ca27778a49b38599f70c534e76a0) Thanks [@mattpocock](https://github.com/mattpocock)! - Add the **`research`** skill — a small, model-invoked skill that spins up a **background agent** to investigate a question against **primary sources** (official docs, source code, specs, first-party APIs), then leaves a single cited Markdown file wherever the repo keeps such notes. It's delegable reading legwork: you keep working while it reads, and get back a document to grill, plan, or design against. Listed in the top-level and Engineering READMEs (Model-invoked), added to `.claude-plugin/plugin.json`, given a docs page at `docs/engineering/research.md`, and routed as a Standalone in `ask-matt`.
 - [#469](https://github.com/mattpocock/skills/pull/469) [`a0329ba`](https://github.com/mattpocock/skills/commit/a0329ba95751f58566ed7ab484475917a68f1629) Thanks [@mattpocock](https://github.com/mattpocock)! - Split the **`to-issues`** skill into a lean **Process** and a **Reference** section, and teach it to handle a **wide refactor** — a single mechanical change (like renaming a column) whose **blast radius** fans across the whole codebase, breaking thousands of call sites at once so no vertical slice can land green. The drafting step now points at two co-located reference blocks: the **Vertical slice rules** for ordinary tracer bullets, and **Wide refactors**, which slices the change by **expand–contract** (expand the new form beside the old, migrate call sites in batches sized by blast radius, then contract the old form away) so CI stays green batch to batch — or, when it can't, only at a final integrate-and-verify issue. The issue body template moves into Reference too.
 - [#464](https://github.com/mattpocock/skills/pull/464) [`386d4ff`](https://github.com/mattpocock/skills/commit/386d4ff719a7c420ad1454232d0436b01f1b8c17) Thanks [@mattpocock](https://github.com/mattpocock)! - Unify the planning skills. **`to-prd` is renamed to `to-spec`** — "spec" is now the single through-line term (it still opens with "you may know this document as a PRD" for discoverability). **`to-plan` and `to-issues` are merged into one `to-tickets` skill, and `to-issues` is deleted.**
  `to-tickets` breaks a plan, spec, or conversation into a set of **tickets** — tracer-bullet vertical slices, each declaring its **blocking edges**. That one artifact reads two ways depending on the tracker `/setup-matt-pocock-skills` configured: a **local file** (`tickets.md`) writes the edges as text and you work it top-to-bottom by hand; a **real tracker** writes them as native blocking links, so any ticket whose blockers are done is on the frontier and several agents can run at once. The edges live in the ticket either way — the medium only decides whether anything acts on them in parallel.
  Publishing prefers the tracker's **native sub-issues** for parent → slice and **native blocking edges** for `Blocked by` where the tracker supports them, keeping the `## Parent` / `## Blocked by` body sections as the fallback. The "What to build" template points at where a `/prototype`'s code lives rather than inlining a snippet from it.
  `ask-matt`'s main flow now routes `idea → /to-spec → /to-tickets → /implement`, and there are human-facing docs pages at `docs/engineering/to-spec.md` and `docs/engineering/to-tickets.md`.
 - [#464](https://github.com/mattpocock/skills/pull/464) [`0557d57`](https://github.com/mattpocock/skills/commit/0557d57579d9b3d39839fdaf8d4a6542b17539ce) Thanks [@mattpocock](https://github.com/mattpocock)! - Settle wayfinder's place in the docs as a **situational on-ramp**, not the new main entry flow — the grill-led _idea → ship_ chain stays the front door (crowning wayfinder as the default spine is a v2-sized move, not a 1.1). The **`ask-matt`** router now names wayfinder's concrete triggers — a greenfield project or a huge feature build, too big for one session — and the two grill front doors (**`grill-me`**, **`grill-with-docs`**) signpost _up_ to wayfinder for the effort that's too big to hold in one session, so the on-ramp is discoverable from where a reader actually starts.
 - [#464](https://github.com/mattpocock/skills/pull/464) [`639df6e`](https://github.com/mattpocock/skills/commit/639df6e7386dfddc739b2aecdeff37a876f2483b) Thanks [@mattpocock](https://github.com/mattpocock)! - Graduate and reframe **`wayfinder`** — the skill for planning a huge chunk of work, more than one agent session can hold. It moves out of `in-progress/` into `engineering/` (plugin entry, top-level + Engineering READMEs under **User-invoked**, a docs page at `docs/engineering/wayfinder.md`, and a route in `ask-matt`), landing as a mature skill. The rename and reframe that got it there:
  - **`decision-mapping` is renamed to `wayfinder`**, invoked as `/wayfinder`. "Decision map" was jargony and inaccurate — only one ticket type is actually a decision. The reframe charts a route through a foggy problem instead, giving one coherent leading-word frame — **fog of war**, **frontier**, **the map** — rather than an invented term layered on top.
  - **Destination as the leading word.** Wayfinding finds the _way_ to a destination; it doesn't charge at building it. Naming the destination is the first act of charting — it fixes the scope and shapes every ticket — so the map gains a `## Destination` field every session orients to, and triage pins it before any ticket exists.
  - **Plan, don't do.** The map produces **decisions, not deliverables**; it's done when nothing is left to decide before someone builds the thing. An effort can override this in its Notes.
  - **The map is an index, not a store.** A decision lives in exactly one place — its ticket — so the map only gists and links, never restates; graduating fog into a ticket clears the graduated patch so nothing lingers in two places.
  - **Collaborative by default.** The map moves off a local Markdown file onto the repo's issue tracker: a single `wayfinder:map` issue whose tickets are its child issues — one shared URL the team can watch. Sessions load the map at low resolution and zoom into tickets on demand. Wayfinder stays tracker-agnostic (GitHub, GitLab, local-markdown) behind a pointer in `docs/agents/issue-tracker.md`, and `setup-matt-pocock-skills` seeds the "Wayfinding operations" section.
  - **Claim by assignment, not a label.** A session claims a ticket by assigning it to the driving dev — the assignee _is_ the claim — freeing the label vocabulary to `wayfinder:<type>` alone.
  - **Native blocking.** Blocking prefers the tracker's native dependency relationship, which renders the frontier visually in the tracker's own UI so the human sees what's takeable without opening the map. GitHub and GitLab templates spell out the native recipe, with a body-convention fallback.
  - **Fog vs. out of scope, split.** Two plainly-named map sections — `## Not yet specified` (in-scope fog that graduates as the frontier advances) and `## Out of scope` (work ruled beyond the destination, closed, never graduating) — so beyond-destination work no longer reads as takeable frontier.
  - **A fourth `task` ticket type.** For literal manual work that blocks a decision (provisioning access, moving data, signing up for a service) — the one type that _does_ rather than decides, earning its place by unblocking a decision.
  - **HITL / AFK ticket classification.** Every ticket type is **HITL** (human in the loop — grilling, prototype) or **AFK** (agent alone — research; task is either). A HITL ticket only resolves through the live exchange, so "wait for the human" falls out of the label — a grilling agent that answers its own questions has, by definition, broken HITL. (This fixes students' reports of `/wayfinder` grilling _itself_ instead of the human.)
  - **No-fog early exit restored.** If the opening breadth-first grilling surfaces no fog, the journey is small enough for one session — so it stops and asks how you'd like to proceed rather than building a map nobody needs.
 ### Patch Changes
 - [#464](https://github.com/mattpocock/skills/pull/464) [`639df6e`](https://github.com/mattpocock/skills/commit/639df6e7386dfddc739b2aecdeff37a876f2483b) Thanks [@mattpocock](https://github.com/mattpocock)! - Reshape **`tdd`** into a reference-only skill and add a missing anti-pattern.
  **Reference-only.** The red → green → refactor loop is anchored by leading words the model already holds, so the step-by-step Workflow was largely restating the loop. Dropped the Workflow and per-cycle checklist; folded their one durable idea — vertical slices / tracer bullets — into the Anti-patterns section and a short Rules-of-the-loop list. Introduced **seam** as the leading word for where tests go: test only at pre-agreed seams, confirmed with the user before any test is written. Also dropped the refactor stage — TDD is now red → green; refactoring belongs to the review stage, so the refactor rule and `refactoring.md` moved out (its home is `code-review`).
  **Tautological tests.** Added the tautological-test anti-pattern: a test whose assertion is recomputed the way the code computes it passes by construction and gives zero confidence — distinct from the implementation-coupling anti-pattern already covered. Added as a peer at the same sites: a Philosophy principle (expected values must come from an independent source of truth), a checklist gate, and a BAD/GOOD example pair in `tests.md`.
 - [`e00eadb`](https://github.com/mattpocock/skills/commit/e00eadb4bb32c3d5a631ead1a5ed5d6a7c5f74e2) Thanks [@mattpocock](https://github.com/mattpocock)! - Extend the **`triage`** skill to triage external pull requests, treating a PR as an issue with attached code that runs through the same roles and state machine. PRs flow inline alongside issues (gated by a per-repo setup toggle), discovery surfaces only external PRs, the bug-only "reproduce" step is generalized into a single "verify the claim" step, and a redundancy check resolves already-implemented requests to `wontfix` without polluting the out-of-scope knowledge base. `setup-matt-pocock-skills` gains the PRs-as-a-request-surface toggle for GitHub/GitLab.
 - [#472](https://github.com/mattpocock/skills/pull/472) [`d869d45`](https://github.com/mattpocock/skills/commit/d869d45afc32beab1c2d1350f8de5e81589512cd) Thanks [@mattpocock](https://github.com/mattpocock)! - Fix **`wayfinder`** hardcoding the issue-tracker doc path, which broke the indirection the rest of the suite relies on.
  `to-issues`, `to-prd`, and `triage` never name a path — they resolve the tracker through the `### Issue tracker` block that `setup-matt-pocock-skills` writes into `CLAUDE.md` / `AGENTS.md`, which points at the tracker doc wherever it lives. Wayfinder instead pinned the literal `docs/agents/issue-tracker.md`, so in a repo that keeps its agent docs elsewhere it silently fell back to the local-markdown tracker — even one whose `CLAUDE.md` clearly declares GitHub issues. It now resolves the doc via that same pointer and reads its "Wayfinding operations" section by name, keeping the indirection consistent across the suite.
 ## 1.0.1
 ### Patch Changes
 - [`d20ee26`](https://github.com/mattpocock/skills/commit/d20ee2684e2a9442698ac3c1e0f2c5b68c4cf296) Thanks [@mattpocock](https://github.com/mattpocock)! - Make the **`teach`** skill reuse-first. Lessons are now built from reusable **components** in `./assets/` — stylesheets, quiz widgets, simulators, diagram helpers. Reuse is the default: the agent reads `./assets/` before authoring a lesson, builds from what's there, and extracts anything new and reusable into a component rather than inlining it.
 ## 1.0.0
 ### Major Changes
 - [`47bde84`](https://github.com/mattpocock/skills/commit/47bde84da032afb2e5058f997f3bbca47d321dbd) Thanks [@mattpocock](https://github.com/mattpocock)! - Add the **`ask-matt`** skill — a user-invoked router that points you at the right skill or flow for your situation.
  **Breaking:** `ask-matt` routes over the other user-invoked skills in this repo, so it expects them to be installed.
 - [`47bde84`](https://github.com/mattpocock/skills/commit/47bde84da032afb2e5058f997f3bbca47d321dbd) Thanks [@mattpocock](https://github.com/mattpocock)! - Add the shared design skills and rewire existing skills onto them.
  - New **`codebase-design`** skill — the deep-module vocabulary (module, interface, depth, seam, adapter) and the principles for putting a lot of behaviour behind a small interface. The language that previously lived in `improve-codebase-architecture/LANGUAGE.md` now lives here, generalized for reuse across skills.
  - New **`domain-modeling`** skill — actively build and sharpen a project's domain model, stress-testing terms against the glossary and keeping `CONTEXT.md` and ADRs current.
  - `improve-codebase-architecture` now draws its architecture vocabulary from `/codebase-design` and its domain model from `/domain-modeling`.
  - `tdd` now leans on `/codebase-design` for interface-design guidance — its inline `deep-modules.md` / `interface-design.md` notes were removed in favour of the shared skill.
  - `grill-with-docs` now builds the domain model inline via `/domain-modeling`.
  **Breaking:** these skills now depend on the new `codebase-design` / `domain-modeling` skills, so you must install them too.
 - [`47bde84`](https://github.com/mattpocock/skills/commit/47bde84da032afb2e5058f997f3bbca47d321dbd) Thanks [@mattpocock](https://github.com/mattpocock)! - Remove the **`caveman`** and **`zoom-out`** skills.
  - `caveman` was a duplicate of another skill I was testing and was never meant to be public.
  - `zoom-out` went unused in practice, so it's been removed from the repo.
  **Breaking:** both skills have been removed.
 - [`47bde84`](https://github.com/mattpocock/skills/commit/47bde84da032afb2e5058f997f3bbca47d321dbd) Thanks [@mattpocock](https://github.com/mattpocock)! - Rename the **`diagnose`** skill to **`diagnosing-bugs`**.
  **Breaking:** invoke it as `/diagnosing-bugs` — the old `/diagnose` name no longer exists.
 - [`47bde84`](https://github.com/mattpocock/skills/commit/47bde84da032afb2e5058f997f3bbca47d321dbd) Thanks [@mattpocock](https://github.com/mattpocock)! - Replace **`write-a-skill`** with **`writing-great-skills`**.
  - Removed `write-a-skill`.
  - Added `writing-great-skills` (plus its `GLOSSARY.md`) — a reference for writing and editing skills well: the vocabulary and principles that make a skill predictable, hunting no-ops down to the sentence level.
  - Exposed `grilling` as a model-invoked skill — the reusable interview loop behind `grill-me` and `grill-with-docs`.
  **Breaking:** `write-a-skill` has been removed; use `writing-great-skills` instead.
 ### Minor Changes
 - [`47bde84`](https://github.com/mattpocock/skills/commit/47bde84da032afb2e5058f997f3bbca47d321dbd) Thanks [@mattpocock](https://github.com/mattpocock)! - Add the **`resolving-merge-conflicts`** skill — a loop for resolving an in-progress git merge or rebase conflict. Standalone, with no dependencies on other skills.
 - [`47bde84`](https://github.com/mattpocock/skills/commit/47bde84da032afb2e5058f997f3bbca47d321dbd) Thanks [@mattpocock](https://github.com/mattpocock)! - Rename the skill taxonomy from **Commands / Skills** to **User-invoked / Model-invoked** across the docs, and add `docs/invocation.md` defining the split: user-invoked skills are reachable only when you type them and exist to orchestrate; model-invoked skills can also be reached automatically when the task fits. A user-invoked skill may invoke model-invoked skills, but never another user-invoked one.
 ### Patch Changes
 - [`47bde84`](https://github.com/mattpocock/skills/commit/47bde84da032afb2e5058f997f3bbca47d321dbd) Thanks [@mattpocock](https://github.com/mattpocock)! - Tighten the **`review`** skill: fail-fast ref check, single-sourced rules, and no-op cuts.
@@ -1,23 +0,0 @@
 A skill's folder decides whether it ships. There are three top-level folders, and no buckets inside them:
 - `skills/` — the **promoted** set. Exactly what the plugin ships, flat, one folder per skill.
 - `drafts/` — beta: public on purpose, feedback wanted, not shipped.
 - `extras/` — kept around but rarely used, not shipped.
 `skills/` is flat because the [Agent Plugins](https://agent-plugins.org/) standard reads only the immediate children of `skills/` and cannot be pointed elsewhere. Never add a folder between `skills/` and a skill, and never put a non-promoted skill under `skills/` — it would ship.
 Promoting a skill is a `git mv` into `skills/`, plus a reference in the top-level `README.md`. Retiring one is a `git mv` out. There is no manifest list to keep in step.
 Install commands are copied verbatim from [.agents/install-block.md](./.agents/install-block.md). Two manifests carry the same metadata and must agree: `plugin.json` at the root (the Agent Plugins 1.0 manifest, read by Codex, Cursor and Copilot) and `.claude-plugin/plugin.json` (read by Claude Code, which does not implement the standard yet). Neither lists skills. `.claude-plugin/marketplace.json` makes the repo its own single-plugin marketplace — a fallback the install block explains, not the documented route. Run `claude plugin validate . --strict` and `node scripts/sync-plugin-version.mjs --check` after touching any manifest. Why the repo is laid out this way lives in [.agents/adr/0003-flat-skills-tree-for-agent-plugins.md](./.agents/adr/0003-flat-skills-tree-for-agent-plugins.md).
 Each skill entry in the top-level `README.md` must link the skill name to its `SKILL.md`.
 `drafts/` and `extras/` each have a `README.md` listing every skill in the folder with a one-line description, name linked to its `SKILL.md`, as a flat list. The promoted skills are listed only in the top-level `README.md`, grouped into **User-invoked** and **Model-invoked**.
 Every promoted skill also has a human-facing docs page at `docs/<category>/<skill-name>.md`, where `<category>` is `engineering` or `productivity`. That docs folder is the **only** place a skill's category lives — no client reads it, and it drives nothing but the grouping on the site and in the top-level `README.md`. The published URL is `https://aihero.dev/skills-<skill-name>` regardless of category. When you add, rename, or change the behaviour of a promoted skill, create or re-sync its docs page following [.agents/writing-docs.md](./.agents/writing-docs.md). A finished page carries four sections — **What it does**, **When to reach for it**, **Common questions**, **It's working if** — and `writing-docs.md` holds the template, the section order, and where to hunt for the questions. Skills in `drafts/` and `extras/` get **no** docs page.
 Every `SKILL.md` is either user-invoked (`disable-model-invocation: true` plus `policy.allow_implicit_invocation: false` in `agents/openai.yaml`, reachable only by the human) or model-invoked (model- or user-reachable). See [.agents/invocation.md](./.agents/invocation.md).
 [`ask-matt`](./skills/ask-matt/SKILL.md) is the router that maps every user-reachable skill and how they relate. The same trigger that re-syncs a docs page applies to it: whenever you add, rename, remove, or change how a user-reachable skill fits the flows, re-read `ask-matt`'s `SKILL.md` and update it so the map stays accurate — a new skill it never mentions, or a stale one it still routes to, is a router that lies.
 To (re)link every skill into the local harness skill directories (`~/.claude/skills`, `~/.agents/skills`), run `scripts/link-skills.sh`. Each entry is a symlink into this repo, so a `git pull` keeps installed skills current; re-run the script after adding, removing, or renaming a skill.
@@ -1,30 +0,0 @@
 # Matt Pocock Skills
 A collection of agent skills (slash commands and behaviors) loaded by Claude Code. Skills are organized into buckets and consumed by per-repo configuration emitted by `/setup-matt-pocock-skills`.
 ## Language
 **Issue tracker**:
 The tool that hosts a repo's issues — GitHub Issues, Linear, a local `.scratch/` markdown convention, or similar. Skills like `to-tickets`, `to-spec`, and `triage` read from and write to it.
 _Avoid_: backlog manager, backlog backend, issue host
 **Issue**:
 A single tracked unit of work inside an **Issue tracker** — a bug, task, spec, or slice produced by `to-tickets`.
 _Avoid_: ticket (use only when quoting external systems that call them tickets, or for a **Decision ticket** — see below)
 **Decision ticket**:
 A `wayfinder` unit — a child **Issue** of a `wayfinder:map` holding a *question* whose resolution is a decision, not a slice of a build to execute. The **decision** qualifier is what keeps it distinct from an implementation ticket; `wayfinder` introduces the term, then uses "ticket".
 **Triage role**:
 A canonical state-machine label applied to an **Issue** during triage (e.g. `needs-triage`, `ready-for-afk`). Each role maps to a real label string in the **Issue tracker** via `docs/agents/triage-labels.md`.
 ## Relationships
 - An **Issue tracker** holds many **Issues**
 - An **Issue** carries one **Triage role** at a time
 - A **Decision ticket** is an **Issue** (a child of a `wayfinder:map`)
 ## Flagged ambiguities
 - "backlog" was previously used to mean both the *tool* hosting issues and the *body of work* inside it — resolved: the tool is the **Issue tracker**; "backlog" is no longer used as a domain term.
 - "backlog backend" / "backlog manager" — resolved: collapsed into **Issue tracker**.
@@ -1,231 +1,115 @@
-<p>
+# Agent Skills For Real Engineers
  <a href="https://www.aihero.dev/s/skills-newsletter">
    <picture>
      <source media="(prefers-color-scheme: dark)" srcset="https://res.cloudinary.com/total-typescript/image/upload/v1777382277/skills-repo-dark_2x.png">
      <source media="(prefers-color-scheme: light)" srcset="https://res.cloudinary.com/total-typescript/image/upload/v1777382277/skill-repo-light_2x.png">
      <img alt="Skills" src="https://res.cloudinary.com/total-typescript/image/upload/v1777382277/skill-repo-light_2x.png" width="369">
    </picture>
  </a>
 </p>
 # Skills For Real Engineers
 [![skills.sh](https://skills.sh/b/mattpocock/skills)](https://skills.sh/mattpocock/skills)
 My agent skills that I use every day to do real engineering - not vibe coding.
 Developing real applications is hard. Approaches like GSD, BMAD, and Spec-Kit try to help by owning the process. But while doing so, they take away your control and make bugs in the process hard to resolve.
 These skills are designed to be small, easy to adapt, and composable. They work with any model. They're based on decades of engineering experience. Hack around with them. Make them your own. Enjoy.
 If you want to keep up with changes to these skills, and any new ones I create, you can join ~60,000 other devs on my newsletter:
 [Sign Up To The Newsletter](https://www.aihero.dev/s/skills-newsletter)
-## Installation (30-second setup)
+## Planning & Design
-Two ways in, two philosophies. **The [Claude Code plugin](https://code.claude.com/docs/en/plugins)** installs the whole set as a managed, read-only bundle that updates when I ship — you subscribe rather than fork. **[skills.sh](https://skills.sh/mattpocock/skills)** copies editable skill files into your project, so you can hack on them and make them your own. Pick one — installing both leaves you with every skill twice.
+These skills help you think through problems before writing code.
-### 1. Get the skills
+- **to-prd** — Turn the current conversation context into a PRD and submit it as a GitHub issue. No interview — just synthesizes what you've already discussed.
-<details>
+  ```
-<summary><strong>Claude Code</strong></summary>
+  npx skills@latest add mattpocock/skills/to-prd
  ```
-```bash
+- **to-issues** — Break any plan, spec, or PRD into independently-grabbable GitHub issues using vertical slices.
 claude plugins install mattpocock-skills
 ```
-Or, from inside a session:
+  ```
  npx skills@latest add mattpocock/skills/to-issues
  ```
-```
+- **grill-me** — Get relentlessly interviewed about a plan or design until every branch of the decision tree is resolved.
 /plugin install mattpocock-skills
 ```
-It's in Claude Code's official marketplace, so there's nothing to add first, and updates arrive automatically.
+  ```
  npx skills@latest add mattpocock/skills/grill-me
  ```
-</details>
+- **design-an-interface** — Generate multiple radically different interface designs for a module using parallel sub-agents.
-<details>
+  ```
-<summary><strong>Codex, and other agents</strong></summary>
+  npx skills@latest add mattpocock/skills/design-an-interface
  ```
-```bash
+- **request-refactor-plan** — Create a detailed refactor plan with tiny commits via user interview, then file it as a GitHub issue.
 npx skills@latest add mattpocock/skills
 ```
-Pick the skills you want, and which coding agents to install them on. **The installer lets you choose which skills to take — make sure `setup-matt-pocock-skills` is one of them.**
+  ```
  npx skills@latest add mattpocock/skills/request-refactor-plan
  ```
-A native Codex plugin is on the roadmap — see [`.agents/adr/0002-ship-as-a-claude-code-plugin.md`](./.agents/adr/0002-ship-as-a-claude-code-plugin.md).
+## Development
-</details>
+These skills help you write, refactor, and fix code.
-<details>
+- **tdd** — Test-driven development with a red-green-refactor loop. Builds features or fixes bugs one vertical slice at a time.
 <summary><strong>For tinkerers</strong></summary>
-Use the same installer, on any agent — including Claude Code:
+  ```
  npx skills@latest add mattpocock/skills/tdd
  ```
-```bash
+- **triage-issue** — Investigate a bug by exploring the codebase, identify the root cause, and file a GitHub issue with a TDD-based fix plan.
 npx skills@latest add mattpocock/skills
 ```
-It writes the skills into your repo as ordinary files you own and can edit. Nothing updates behind your back; pull my latest changes when you want them with `npx skills update`.
+  ```
  npx skills@latest add mattpocock/skills/triage-issue
  ```
-</details>
+- **improve-codebase-architecture** — Find deepening opportunities in a codebase, informed by the domain language in `CONTEXT.md` and the decisions in `docs/adr/`.
-### 2. Run `/setup-matt-pocock-skills`
+  ```
  npx skills@latest add mattpocock/skills/improve-codebase-architecture
  ```
-In your agent, run it once per repo. It will:
+- **migrate-to-shoehorn** — Migrate test files from `as` type assertions to @total-typescript/shoehorn.
- Ask you which issue tracker you want to use (GitHub, Linear, or local files)
+  ```
- Ask you what labels you apply to tickets when you triage them (`/triage` uses labels)
+  npx skills@latest add mattpocock/skills/migrate-to-shoehorn
- Ask you where you want to save any docs we create
+  ```
-### 3. Bam - you're ready to go.
+- **scaffold-exercises** — Create exercise directory structures with sections, problems, solutions, and explainers.
-## Why These Skills Exist
+  ```
  npx skills@latest add mattpocock/skills/scaffold-exercises
  ```
-I built these skills as a way to fix common failure modes I see with Claude Code, Codex, and other coding agents.
+## Tooling & Setup
-### #1: The Agent Didn't Do What I Want
+- **setup-pre-commit** — Set up Husky pre-commit hooks with lint-staged, Prettier, type checking, and tests.
-> "No-one knows exactly what they want"
+  ```
->
+  npx skills@latest add mattpocock/skills/setup-pre-commit
-> David Thomas & Andrew Hunt, [The Pragmatic Programmer](https://www.amazon.co.uk/Pragmatic-Programmer-Anniversary-Journey-Mastery/dp/B0833F1T3V)
+  ```
-**The Problem**. The most common failure mode in software development is misalignment. You think the dev knows what you want. Then you see what they've built - and you realize it didn't understand you at all.
+- **git-guardrails-claude-code** — Set up Claude Code hooks to block dangerous git commands (push, reset --hard, clean, etc.) before they execute.
-This is just the same in the AI age. There is a communication gap between you and the agent. The fix for this is a **grilling session** - getting the agent to ask you detailed questions about what you're building.
+  ```
  npx skills@latest add mattpocock/skills/git-guardrails-claude-code
  ```
-**The Fix** is to use:
+## Writing & Knowledge
- [`/grill-me`](./skills/grill-me/SKILL.md) - for non-code uses
+- **write-a-skill** — Create new skills with proper structure, progressive disclosure, and bundled resources.
 - [`/grill-with-docs`](./skills/grill-with-docs/SKILL.md) - same as [`/grill-me`](./skills/grill-me/SKILL.md), but adds more goodies (see below)
-These are my most popular skills. They help you align with the agent before you get started, and think deeply about the change you're making. Use them _every_ time you want to make a change.
+  ```
  npx skills@latest add mattpocock/skills/write-a-skill
  ```
-### #2: The Agent Is Way Too Verbose
+- **edit-article** — Edit and improve articles by restructuring sections, improving clarity, and tightening prose.
-> With a ubiquitous language, conversations among developers and expressions of the code are all derived from the same domain model.
+  ```
->
+  npx skills@latest add mattpocock/skills/edit-article
-> Eric Evans, [Domain-Driven-Design](https://www.amazon.co.uk/Domain-Driven-Design-Tackling-Complexity-Software/dp/0321125215)
+  ```
-**The Problem**: At the start of a project, devs and the people they're building the software for (the domain experts) are usually speaking different languages.
+- **ubiquitous-language** — Extract a DDD-style ubiquitous language glossary from the current conversation.
-I felt the same tension with my agents. Agents are usually dropped into a project and asked to figure out the jargon as they go. So they use 20 words where 1 will do.
+  ```
  npx skills@latest add mattpocock/skills/ubiquitous-language
  ```
-**The Fix** for this is a shared language. It's a document that helps agents decode the jargon used in the project.
+- **obsidian-vault** — Search, create, and manage notes in an Obsidian vault with wikilinks and index notes.
-<details>
+  ```
-<summary>
+  npx skills@latest add mattpocock/skills/obsidian-vault
-Example
+  ```
 </summary>
 Here's an example [`CONTEXT.md`](https://github.com/mattpocock/course-video-manager/blob/076a5a7a182db0fe1e62971dd7a68bcadf010f1c/CONTEXT.md), from my `course-video-manager` repo. Which one is easier to read?
 - **BEFORE**: "There's a problem when a lesson inside a section of a course is made 'real' (i.e. given a spot in the file system)"
 - **AFTER**: "There's a problem with the materialization cascade"
 This concision pays off session after session.
 </details>
 This is built into [`/grill-with-docs`](./skills/grill-with-docs/SKILL.md). It's a grilling session, but that helps you build a shared language with the AI, and document hard-to-explain decisions in ADR's.
 It's hard to explain how powerful this is. It might be the single coolest technique in this repo. Try it, and see.
 > [!TIP]
 > A shared language has many other benefits than reducing verbosity:
 >
 > - **Variables, functions and files are named consistently**, using the shared language
 > - As a result, the **codebase is easier to navigate** for the agent
 > - The agent also **spends fewer tokens on thinking**, because it has access to a more concise language
 ### #3: The Code Doesn't Work
 > "Always take small, deliberate steps. The rate of feedback is your speed limit. Never take on a task that’s too big."
 >
 > David Thomas & Andrew Hunt, [The Pragmatic Programmer](https://www.amazon.co.uk/Pragmatic-Programmer-Anniversary-Journey-Mastery/dp/B0833F1T3V)
 **The Problem**: Let's say that you and the agent are aligned on what to build. What happens when the agent _still_ produces crap?
 It's time to look at your feedback loops. Without feedback on how the code it produces actually runs, the agent will be flying blind.
 **The Fix**: You need the usual tranche of feedback loops: static types, browser access, and automated tests.
 For automated tests, a red-green-refactor loop is critical. This is where the agent writes a failing test first, then fixes the test. This helps give the agent a consistent level of feedback that results in far better code.
 I've built a **[`/tdd`](./skills/tdd/SKILL.md) skill** you can slot into any project. It encourages red-green-refactor and gives the agent plenty of guidance on what makes good and bad tests.
 For debugging, I've also built a **[`/diagnosing-bugs`](./skills/diagnosing-bugs/SKILL.md)** skill that wraps best debugging practices into a disciplined loop, gated phase by phase.
 ### #4: We Built A Ball Of Mud
 > "Invest in the design of the system _every day_."
 >
 > Kent Beck, [Extreme Programming Explained](https://www.amazon.co.uk/Extreme-Programming-Explained-Embrace-Change/dp/0321278658)
 > "The best modules are deep. They allow a lot of functionality to be accessed through a simple interface."
 >
 > John Ousterhout, [A Philosophy Of Software Design](https://www.amazon.co.uk/Philosophy-Software-Design-2nd/dp/173210221X)
 **The Problem**: Most apps built with agents are complex and hard to change. Because agents can radically speed up coding, they also accelerate software entropy. Codebases get more complex at an unprecedented rate.
 **The Fix** for this is a radical new approach to AI-powered development: caring about the design of the code.
 This is built in to every layer of these skills:
 - [`/to-spec`](./skills/to-spec/SKILL.md) quizzes you about which modules you're touching before creating a spec
 And crucially, [`/improve-codebase-architecture`](./skills/improve-codebase-architecture/SKILL.md) surveys a codebase for deepening opportunities and hands you the candidates. I recommend running it on your codebase once every few days. It is a survey, not a rescue: on a genuinely old codebase it will find real candidates, but it won't untangle the mud for you.
 ### Summary
 Software engineering fundamentals matter more than ever. These skills are my best effort at condensing these fundamentals into repeatable practices, to help you ship the best apps of your career. Enjoy.
 ## Reference
 These split on one axis — who can invoke them. **User-invoked** skills are reachable only when you type them (e.g. `/grill-me`); their job is to orchestrate. **Model-invoked** skills can be invoked by you _or_ reached for automatically by the agent when the task fits; they hold the reusable discipline. A user-invoked skill may invoke model-invoked skills, but never another user-invoked one.
 ### Engineering
 Skills I use daily for code work.
 **User-invoked**
 - **[ask-matt](./skills/ask-matt/SKILL.md)** — Ask which skill or flow fits your situation. A router over the user-invoked skills in this repo.
 - **[grill-with-docs](./skills/grill-with-docs/SKILL.md)** — Grilling session that also builds your project's domain model, sharpening terminology and updating `CONTEXT.md` and ADRs inline.
 - **[triage](./skills/triage/SKILL.md)** — Move issues through a state machine of triage roles.
 - **[improve-codebase-architecture](./skills/improve-codebase-architecture/SKILL.md)** — Scan a codebase for deepening opportunities, present them as a visual HTML report, then grill through whichever one you pick.
 - **[setup-matt-pocock-skills](./skills/setup-matt-pocock-skills/SKILL.md)** — Configure this repo for the engineering skills (issue tracker, triage labels, domain doc layout). Run once per repo before using the other engineering skills.
 - **[to-spec](./skills/to-spec/SKILL.md)** — Turn the current conversation into a spec and publish it to the issue tracker. No interview — just synthesizes what you've already discussed.
 - **[to-tickets](./skills/to-tickets/SKILL.md)** — Break any plan, spec, or conversation into a set of tracer-bullet tickets, each declaring its blocking edges — written as text in a local file, or as native blocking links on a real tracker.
 - **[implement](./skills/implement/SKILL.md)** — Build the work described by a spec or set of tickets, driving `/tdd` at pre-agreed seams and closing out with `/code-review` before committing.
 - **[wayfinder](./skills/wayfinder/SKILL.md)** — Plan a huge chunk of work, more than one agent session can hold, as a shared map of decision tickets on the issue tracker — resolve them one at a time until the way to the destination is clear.
 **Model-invoked**
 - **[prototype](./skills/prototype/SKILL.md)** — Build a throwaway prototype to answer a design question — a single shareable HTML file for state/logic questions, or several radically different UI variations toggleable from one route.
 - **[diagnosing-bugs](./skills/diagnosing-bugs/SKILL.md)** — Disciplined diagnosis loop for hard bugs and performance regressions: build a feedback loop that goes red on this bug → minimise → hypothesise → instrument → fix → regression-test.
 - **[research](./skills/research/SKILL.md)** — Investigate a question against high-trust primary sources and capture the findings as a cited Markdown file in the repo, run as a background agent.
 - **[tdd](./skills/tdd/SKILL.md)** — Test-driven development with a red-green-refactor loop. Builds features or fixes bugs one vertical slice at a time.
 - **[domain-modeling](./skills/domain-modeling/SKILL.md)** — Actively build and sharpen a project's domain model — challenge terms against the glossary, stress-test with edge-case scenarios, and update `CONTEXT.md` and ADRs inline.
 - **[codebase-design](./skills/codebase-design/SKILL.md)** — Shared discipline and vocabulary for designing deep modules: a lot of behaviour behind a small interface, placed at a clean seam, testable through that interface.
 - **[code-review](./skills/code-review/SKILL.md)** — Two-axis review of the diff since a fixed point: **Standards** (does it follow the repo's coding standards, plus a Fowler smell baseline?) and **Spec** (does it faithfully implement the originating issue/spec?), run as parallel sub-agents so neither pollutes the other.
 - **[resolving-merge-conflicts](./skills/resolving-merge-conflicts/SKILL.md)** — Work through an in-progress git merge or rebase conflict hunk by hunk, resolving by intent traced to each side's primary source, then finish the operation — never `--abort`.
 - **[wizard](./skills/wizard/SKILL.md)** — Generate an interactive bash wizard that walks a human through steps only they can perform: provisioning infrastructure, setting up credentials or CI secrets, walking an unfamiliar third-party dashboard, or running a one-off migration or cutover.
 ### Productivity
 General workflow tools, not code-specific.
 **User-invoked**
 - **[grill-me](./skills/grill-me/SKILL.md)** — Get relentlessly interviewed about a plan or design until every branch of the design tree is resolved.
 - **[handoff](./skills/handoff/SKILL.md)** — Compact the current conversation into a handoff document so another agent can continue the work.
 - **[teach](./skills/teach/SKILL.md)** — Teach the user a new skill or concept over multiple sessions, using the current directory as a stateful teaching workspace.
 - **[to-questionnaire](./skills/to-questionnaire/SKILL.md)** — Turn a decision you can't answer alone into a Markdown questionnaire for the one person who can — filled in async, or together over a meeting. It grills you about the send (who it's for, what you need back), not the subject.
 - **[wait-what](./skills/wait-what/SKILL.md)** — Fire this the moment a message doesn't land. The agent re-pitches it with the context you're missing, in plain English, using your `CONTEXT.md` vocabulary.
 **Model-invoked**
 - **[grilling](./skills/grilling/SKILL.md)** — Interview the user relentlessly about a plan, decision, or idea until every branch of the design tree is resolved. The reusable interview primitive behind `grill-me`, `grill-with-docs`, `triage`, `wayfinder` and `improve-codebase-architecture`.
 - **[writing-for-agents](./skills/writing-for-agents/SKILL.md)** — Writing documents for agents: skills, AGENTS.md/CLAUDE.md, and any doc an agent reaches by a pointer.
@@ -0,0 +1,49 @@
 ---
 name: caveman
 description: >
  Ultra-compressed communication mode. Cuts token usage ~75% by dropping
  filler, articles, and pleasantries while keeping full technical accuracy.
  Use when user says "caveman mode", "talk like caveman", "use caveman",
  "less tokens", "be brief", or invokes /caveman.
 ---
 Respond terse like smart caveman. All technical substance stay. Only fluff die.
 ## Persistence
 ACTIVE EVERY RESPONSE once triggered. No revert after many turns. No filler drift. Still active if unsure. Off only when user says "stop caveman" or "normal mode".
 ## Rules
 Drop: articles (a/an/the), filler (just/really/basically/actually/simply), pleasantries (sure/certainly/of course/happy to), hedging. Fragments OK. Short synonyms (big not extensive, fix not "implement a solution for"). Abbreviate common terms (DB/auth/config/req/res/fn/impl). Strip conjunctions. Use arrows for causality (X -> Y). One word when one word enough.
 Technical terms stay exact. Code blocks unchanged. Errors quoted exact.
 Pattern: `[thing] [action] [reason]. [next step].`
 Not: "Sure! I'd be happy to help you with that. The issue you're experiencing is likely caused by..."
 Yes: "Bug in auth middleware. Token expiry check use `<` not `<=`. Fix:"
 ### Examples
 **"Why React component re-render?"**
 > Inline obj prop -> new ref -> re-render. `useMemo`.
 **"Explain database connection pooling."**
 > Pool = reuse DB conn. Skip handshake -> fast under load.
 ## Auto-Clarity Exception
 Drop caveman temporarily for: security warnings, irreversible action confirmations, multi-step sequences where fragment order risks misread, user asks to clarify or repeats question. Resume caveman after clear part done.
 Example -- destructive op:
 > **Warning:** This will permanently delete all rows in the `users` table and cannot be undone.
 >
 > ```sql
 > DROP TABLE users;
 > ```
 >
 > Caveman resume. Verify backup exist first.
@@ -0,0 +1,94 @@
 ---
 name: design-an-interface
 description: Generate multiple radically different interface designs for a module using parallel sub-agents. Use when user wants to design an API, explore interface options, compare module shapes, or mentions "design it twice".
 ---
 # Design an Interface
 Based on "Design It Twice" from "A Philosophy of Software Design": your first idea is unlikely to be the best. Generate multiple radically different designs, then compare.
 ## Workflow
 ### 1. Gather Requirements
 Before designing, understand:
 - [ ] What problem does this module solve?
 - [ ] Who are the callers? (other modules, external users, tests)
 - [ ] What are the key operations?
 - [ ] Any constraints? (performance, compatibility, existing patterns)
 - [ ] What should be hidden inside vs exposed?
 Ask: "What does this module need to do? Who will use it?"
 ### 2. Generate Designs (Parallel Sub-Agents)
 Spawn 3+ sub-agents simultaneously using Task tool. Each must produce a **radically different** approach.
 ```
 Prompt template for each sub-agent:
 Design an interface for: [module description]
 Requirements: [gathered requirements]
 Constraints for this design: [assign a different constraint to each agent]
 - Agent 1: "Minimize method count - aim for 1-3 methods max"
 - Agent 2: "Maximize flexibility - support many use cases"
 - Agent 3: "Optimize for the most common case"
 - Agent 4: "Take inspiration from [specific paradigm/library]"
 Output format:
 1. Interface signature (types/methods)
 2. Usage example (how caller uses it)
 3. What this design hides internally
 4. Trade-offs of this approach
 ```
 ### 3. Present Designs
 Show each design with:
 1. **Interface signature** - types, methods, params
 2. **Usage examples** - how callers actually use it in practice
 3. **What it hides** - complexity kept internal
 Present designs sequentially so user can absorb each approach before comparison.
 ### 4. Compare Designs
 After showing all designs, compare them on:
 - **Interface simplicity**: fewer methods, simpler params
 - **General-purpose vs specialized**: flexibility vs focus
 - **Implementation efficiency**: does shape allow efficient internals?
 - **Depth**: small interface hiding significant complexity (good) vs large interface with thin implementation (bad)
 - **Ease of correct use** vs **ease of misuse**
 Discuss trade-offs in prose, not tables. Highlight where designs diverge most.
 ### 5. Synthesize
 Often the best design combines insights from multiple options. Ask:
 - "Which design best fits your primary use case?"
 - "Any elements from other designs worth incorporating?"
 ## Evaluation Criteria
 From "A Philosophy of Software Design":
 **Interface simplicity**: Fewer methods, simpler params = easier to learn and use correctly.
 **General-purpose**: Can handle future use cases without changes. But beware over-generalization.
 **Implementation efficiency**: Does interface shape allow efficient implementation? Or force awkward internals?
 **Depth**: Small interface hiding significant complexity = deep module (good). Large interface with thin implementation = shallow module (avoid).
 ## Anti-Patterns
 - Don't let sub-agents produce similar designs - enforce radical difference
 - Don't skip comparison - the value is in contrast
 - Don't implement - this is purely about interface shape
 - Don't evaluate based on implementation effort
@@ -1,90 +0,0 @@
 ## What it does
 `ask-matt` is the router over the skills in this repo. You describe the situation you are in — an idea you cannot start, a pile of incoming bug reports, a [session](https://www.aihero.dev/ai-coding-dictionary/session) that has run long — and it names the skill or the sequence of skills that fits, plus where the human decisions in that sequence sit.
 It recommends and stops. It does not grill, write a [spec](https://www.aihero.dev/ai-coding-dictionary/spec), open a file or fire the skill it just named; what you get back is the next thing to type, and you type it. It is also a hand-written map of the skills in this repo rather than a scan of what you have installed, so it will not route you over your own skills or another author's.
 ## When to reach for it
 You invoke this by typing `/ask-matt` — the agent won't reach for it on its own.
 | Your situation | What the router gives back |
 | --- | --- |
 | An idea, and no idea where to start | The head of the main flow, and whether the build is small enough to skip the spec |
 | Bugs and requests arriving from other people | The [triage](https://aihero.dev/skills-triage) on-ramp, and why [tickets](https://www.aihero.dev/ai-coding-dictionary/ticket) you generated yourself don't belong on it |
 | Two skills that look interchangeable | The line between them, and it is usually one concrete test rather than a matter of taste. [grill-me](https://aihero.dev/skills-grill-me) or [grill-with-docs](https://aihero.dev/skills-grill-with-docs) turns on whether you are in a working directory; [grill-with-docs](https://aihero.dev/skills-grill-with-docs) or [wayfinder](https://aihero.dev/skills-wayfinder) turns on whether the effort fits one session |
 | A long session and a decision about the [context](https://www.aihero.dev/ai-coding-dictionary/context) | The ordered tree over the five options at a phase boundary |
 | A skill you have already picked | Nothing useful. Invoke that skill directly. |
 ## Prerequisites
 The router names skills; it does not install them. Everything it points at has to be installed for the recommendation to be actionable, and it only knows the promoted skills in this repo.
 The tracker-dependent routes — triage, `to-spec`, `to-tickets`, `implement` — assume [setup-matt-pocock-skills](https://aihero.dev/skills-setup-matt-pocock-skills) has already configured an issue tracker in the repo. The router will happily recommend them before that has happened.
 ## Flows, not skills
 The word the skill gives you to think with is **flow**: a path *through* the skills, not a single one. Naming your situation places you on a flow at a step, which is a different answer from "here is the skill that matches your keywords". Four kinds of route exist, and the skill itself carries them in full:
 - **The main flow**, idea to ship. Grill, spec, tickets, implement, review, with two branches inside it: a prototype detour when a question needs runnable code to settle, and the spec-and-tickets split, which only earns its cost when the build spans more than one session.
 - **On-ramps**, for a situation that generates work and then merges onto the main flow: incoming bug reports, something broken, or an effort too foggy and too large to hold in one session.
 - **Standalones**, off every flow, reached for on their own terms — the prototype, the questionnaire, the merge conflict you are already sitting in.
 - **A vocabulary layer underneath**, the two references the other skills pull in when the words rather than the process are the problem.
 ## The phase boundary
 The other idea it hands you is the **phase boundary**. A phase is a chunk of work inside a session — the [grilling](https://www.aihero.dev/ai-coding-dictionary/grilling), the implementation, the QA — and the boundary between two of them is the only place the question "what do I do with this context?" belongs. Mid-phase there is nothing to decide: continue, or split what is left into [subagents](https://www.aihero.dev/ai-coding-dictionary/subagent).
 | Option | Take it when |
 | --- | --- |
 | **Continue** | The next phase wants this one verbatim, or you have [smart zone](https://www.aihero.dev/ai-coding-dictionary/smart-zone) left. It is the only move that keeps the session as a [primary source](https://www.aihero.dev/ai-coding-dictionary/primary-source), so rule it out first |
 | **`/clear`** | Everything behind you is disposable. Cheapest move on the board, and one-way if you were wrong |
 | **[handoff](https://aihero.dev/skills-handoff)** | Something has to travel: a new [harness](https://www.aihero.dev/ai-coding-dictionary/harness), a new directory, a colleague, a side task forked mid-phase |
 | **Subagent** | The task is scoped tightly enough to run with you [away from the keyboard](https://www.aihero.dev/ai-coding-dictionary/afk) |
 | **`/compact`** | None of the above. The default, and it lands here often |
 Two of those are routinely got wrong, which is why the router carries the order rather than the list. `/handoff` reads like the general bridge between windows and is not: portability is the whole of what it buys. `/compact` is the bottom of the tree rather than the first reach, because the four questions above it are each cheaper or more precise.
 ## Common questions
 **Isn't there just a list of the skills in the right order?**
 People keep asking for one in the README. This skill is that list — it is what it exists for. A static table would say `wayfinder → to-spec → to-tickets → implement → code-review` and be wrong for most situations, because the interesting parts are the branches — is there a codebase, does the build span sessions, can this question be settled by talking. The honest cost is that the router is hand-maintained and lags the repo. `/grilling` and `/resolving-merge-conflicts` both shipped long before the router named them.
 **It told me half the skills aren't installed.**
 A known bug, unfixed. Most of the skills the router routes you through set `disable-model-invocation: true`, which means the harness leaves them out of the skill list it injects into the agent's context. The agent reads that list as exhaustive and reports them missing. One reported session had it declare the whole spec-and-tickets flow absent and reroute to bare `/grilling` and `/tdd`. Thirteen of the plugin's twenty-two skills carry the flag, so this is the common case rather than an edge. They are installed. Type the slash command anyway, or check `.claude-plugin/plugin.json`, which is the authority on what is present.
 **It described a skill's behaviour, and the skill doesn't do that.**
 Also real, also unfixed. The router answers from its own one-line summary of each skill rather than from the skill. One detailed report tracked three instances in a single session, including a recommendation to skip [to-spec](https://aihero.dev/skills-to-spec) on the strength of the gloss "turn the thread into a spec" — `to-spec/SKILL.md` was never opened. In every case it verified only after the user pushed back, and never on its own initiative. Skipping `to-spec` there cost a real seam check, and the tickets that came out undercounted the work. When the router asserts something load-bearing about another skill, ask it to open that `SKILL.md` first. The same applies to questions the map does not cover at all, such as whether to use [plan mode](https://www.aihero.dev/ai-coding-dictionary/agent-mode): that answer is the [model](https://www.aihero.dev/ai-coding-dictionary/model)'s inference, not something written down here.
 **Why is it prose instead of a numbered checklist?**
 A fair complaint, filed as an open issue arguing that most of the routing is deterministic and the narrative makes it hard to scan. Nothing stops you asking for the compressed form — "just give me the sequence" gets you the sequence. What the prose is carrying is the conditional half: the branches, where a human decision is expected, and where to clear or compact between steps. A flat checklist drops exactly that.
 **Can it route over my own skills, or another author's?**
 No. Three separate proposals have asked for a router that reads your local `skills/` directory and recommends from whatever is installed. `ask-matt` is not that. It is a map of one set, maintained by hand, and it knows nothing about skills you wrote or installed from elsewhere.
 **It told me to edit a SKILL.md.**
 That advice is often correct and rarely durable. Someone asked it how to make [implement](https://aihero.dev/skills-implement) close tickets, got told to add a line to the skill, and immediately spotted the problem: `npx skills update` overwrites the file, and the plugin install is read-only. Put standing behaviour in your own `CLAUDE.md` or `AGENTS.md`, or say it in the invocation. Prompt-level adaptations survive updates — pointing the flow at Linear instead of GitHub, or asking it which open tickets could run in parallel, are both things people do this way.
 **It named a skill I don't have, or missed one I do.**
 Check the changelog for a rename before assuming it is gone. `writing-great-skills` became [writing-for-agents](https://aihero.dev/skills-writing-for-agents) with no alias, `to-prd` became [to-spec](https://aihero.dev/skills-to-spec), and `pathfinder` became [wayfinder](https://aihero.dev/skills-wayfinder). Four skills were retired outright into the skills that absorbed them: `ubiquitous-language`, `design-an-interface`, `qa` and `request-refactor-plan`. The reverse case is the router's own lag, above.
 ## It's working if
 - It ends by naming what to type and stops there, instead of starting the work itself.
 - The route it gives back mentions where to clear or compact context and where you are expected to review, not just a list of skill names.
 - Where two skills are close, it says which one and why the other is wrong for you.
 - Any claim it makes about another skill's behaviour shows up in the trace as it reading that skill's `SKILL.md`.
 - You recognise your own situation in what it hands back, rather than the nearest generic scenario.
 ## Where it fits
 `ask-matt` is a **standalone router** that sits over the whole set. It is never a step in a chain; it points into every chain, and it is the node the other docs pages link back to so none of them has to redraw the graph. From here you most often land on [grill-with-docs](https://aihero.dev/skills-grill-with-docs), the head of the main flow, or [triage](https://aihero.dev/skills-triage), the on-ramp for work that arrived rather than work you started.
 It is a [secondary source](https://www.aihero.dev/ai-coding-dictionary/secondary-source) over the skills it describes. Where the router and a `SKILL.md` disagree, the `SKILL.md` is right.
@@ -1,94 +0,0 @@
 ## What it does
 `code-review` reviews the diff between `HEAD` and a fixed point you name — a commit, a branch, a tag, `main`, `HEAD~5` — along two axes. **Standards** asks whether the code follows how this repo writes code. **Spec** asks whether the code does what the originating issue or [spec](https://www.aihero.dev/ai-coding-dictionary/spec) asked for. Each axis runs in its own [sub-agent](https://www.aihero.dev/ai-coding-dictionary/subagent) so neither sees the other's reasoning.
 The two axes are never merged and never re-ranked. The report ends with a worst issue *per axis* and refuses to name a single winner across them, because a change can pass one axis and fail the other: code that follows every convention while implementing the wrong thing passes Standards and fails Spec; code that does exactly what the [ticket](https://www.aihero.dev/ai-coding-dictionary/ticket) asked while breaking the repo's conventions does the reverse. A blended verdict lets the passing axis hide the failing one.
 ## When to reach for it
 Type `/code-review`, or the agent reaches for it automatically when you ask to review a branch, a PR, work in progress, or anything "since X".
 | Your situation | Reach for |
 | --- | --- |
 | A diff exists and you want to know if it is built right *and* is the right thing | `code-review` |
 | You want bugs hunted in the diff — null paths, races, off-by-one | Claude Code's own built-in review, not this one (see the name clash below) |
 | Nothing is written yet and you want it written test-first | [tdd](https://aihero.dev/skills-tdd) |
 | A whole spec needs building, review included | [implement](https://aihero.dev/skills-implement), which calls this skill itself |
 | The whole codebase has drifted, not one diff | [improve-codebase-architecture](https://aihero.dev/skills-improve-codebase-architecture) |
 | Something is broken and you do not know why | [diagnosing-bugs](https://aihero.dev/skills-diagnosing-bugs) |
 You must supply the fixed point. If you do not, the skill asks for one rather than guessing; it then checks the ref resolves and the diff is non-empty before spawning anything, so a typo'd branch name fails in front of you instead of inside two sub-agents.
 ## Prerequisites
 The Standards axis needs nothing. It reads whatever the repo documents (`CODING_STANDARDS.md`, `CONTRIBUTING.md`, and the like) and falls back on a built-in baseline when the repo documents nothing.
 The Spec axis needs a spec to exist and be findable. It looks in this order:
 1. Issue references in the commit messages (`#123`, `Closes #45`, a GitLab `!67`), fetched through `docs/agents/issue-tracker.md`.
 2. A path you pass in as an argument.
 3. A spec file under `docs/`, `specs/`, or `.scratch/` matching the branch or feature name.
 4. Asking you.
 Step 1 depends on `docs/agents/issue-tracker.md`, which [setup-matt-pocock-skills](https://aihero.dev/skills-setup-matt-pocock-skills) writes. Without it the axis still works if you hand it a path. With no spec at all, the Spec sub-agent is skipped and the report says "no spec available" rather than inventing requirements.
 ## The two axes
 | | Standards | Spec |
 | --- | --- | --- |
 | Question | Is it built right? | Is it the right thing? |
 | Reads | The repo's documented standards, plus the smell baseline | The originating issue or spec |
 | Reports | Documented breaches (can be hard), and smells (always judgement calls) | Missing or partial requirements, scope creep, requirements implemented wrongly |
 | Every finding cites | The standards file and the rule, or the named smell plus the hunk | The line of the spec |
 A generic review skill that does not know your standards is the thing this design is trying to avoid — it flags what is deliberate in your codebase and misses the invariants your codebase actually depends on. So the repo's own documentation is the [primary source](https://www.aihero.dev/ai-coding-dictionary/primary-source) on the Standards axis, and **the repo always overrides**.
 The **smell baseline** is the floor underneath it: twelve Fowler code smells from _Refactoring_ ch.3 — Mysterious Name, Duplicated Code, Feature Envy, Data Clumps, Primitive Obsession, Repeated Switches, Shotgun Surgery, Divergent Change, Speculative Generality, Message Chains, Middle Man, Refused Bequest. Each is a labelled heuristic ("possible Feature Envy"), never a hard violation, and each is stated as *what it is* → *how to fix*, so a finding arrives with a move attached rather than a complaint. Anything your linter already enforces is skipped by both axes.
 ## Common questions
 **It collides with Claude Code's own `/code-review`. What do I do?**
 This is the most reported problem with the skill, and it is not fixed. Claude Code ships its own `/code-review`, which does something different — it hunts bugs in the diff, where this one checks spec compliance and repo standards. Installing this library means one of them wins, and which one wins depends on how you installed. Via the plugin marketplace, everything is aliased under a `mattpocock-skills:` prefix and the built-in becomes hard to reach at the unqualified name; via a plain skills install, the local file wins and this skill shadows the built-in. One clean answer is to remove Claude Code's built-in skills entirely: a large [context](https://www.aihero.dev/ai-coding-dictionary/context) saving, and the collision stops mattering. The shadowing itself is arguably a Claude Code [harness](https://www.aihero.dev/ai-coding-dictionary/harness) bug — a skill author should be free to name a skill anything — so the other answer is to rename the local copy. Editing the frontmatter or renaming the directory gets undone by `npx skills update`; the durable workaround reported by users is to fork the skill to a new name and drop `code-review` from the managed set, keeping a note of the commit you forked from so you can re-sync by hand.
 **Its sub-agents keep invoking `/code-review` again and spawn more agents.**
 Known open bug, reproduced by several people and in more than one harness. The Standards and Spec prompts do not forbid delegation, so a sub-agent can rediscover the skill and fan out again — one report reached 50-plus agents. The fix people have applied on forks is one line appended to both sub-agent briefs: "Do not invoke `/code-review` or spawn additional agents — perform this review directly." Some prefer to handle it at the harness level so every skill inherits the guard. Neither is in the shipped skill yet. If you run this unattended, watch the agent count.
 **Should I run it in the same [session](https://www.aihero.dev/ai-coding-dictionary/session) that wrote the code?**
 Prefer a fresh one. As one reader put it: "Same context reviewing itself isn't review, it's confirmation bias with a slash command." The reviewing agent in the authoring session holds every assumption that shaped the code, which is exactly the context an independent reviewer would not have. This is also why people ask for [implement](https://aihero.dev/skills-implement) without its built-in review step — it runs the review inside the session that just wrote the diff. Invoking `/code-review` yourself from a clean session is the honest version.
 **After every ticket, or once at the end?**
 Both work, and the skill does not decide for you. Per-ticket keeps each diff small enough that the Spec axis has one clear spec to check against, which is the mode `implement` uses. Batching to the end of a branch catches interactions between tickets that the per-ticket passes each miss. If you are unsure, review per ticket and run one final pass against the branch point.
 **Can I trust the findings?**
 Not without checking. Sub-agent output is a hypothesis, not evidence — one team reported a dozen breaking changes that prose-based reviews had waved through. The skill aggregates the two reports verbatim or lightly cleaned rather than re-verifying each claim against the files, so a finding can cite the wrong location or overstate an impact. Read the citation on each finding before acting on it. That every finding is required to carry one — a standards rule, a smell plus its hunk, or a spec line — is what makes this checkable at all.
 **Why does it find new problems every single time I run it?**
 Because fixes create new surface, and because the judgement-call half of the Standards axis is not deterministic between runs. One reader described the loop plainly: "/code-review and /improve-code-architecture always find new stuff every time. I implement fixes, rerun these skills, and again and again." There is no convergence guarantee. Treat a pass as a list of leads, act on the ones with a cited rule behind them, and stop — do not run it in a loop until it comes back clean, because it will not.
 **Does it review my uncommitted work?**
 No. It diffs `<fixed-point>...HEAD`, three-dot, which is measured from the merge-base and excludes staged and working-tree changes. If `implement` has not made an interim commit, the work about to be committed is invisible to the review. Commit first, then review, then amend or add a fixup.
 ## It's working if
 - It refuses to start on a bad ref or an empty diff, before any sub-agent is spawned.
 - The report arrives as two separate blocks under `## Standards` and `## Spec`, not one merged list.
 - Every Standards finding names either a rule in one of your repo's files or one of the twelve smells, with the hunk quoted; every Spec finding quotes a line of the spec.
 - The closing summary gives a worst issue per axis and declines to pick an overall winner.
 - With no spec available, the Spec block says so instead of listing requirements it inferred from the code.
 ## Where it fits
 `code-review` is the review step at the tail of the build chain — `grill-with-docs → to-spec → to-tickets → implement → code-review` — and also stands alone on any branch or PR you point it at.
 - [implement](https://aihero.dev/skills-implement) is the closest neighbour: it drives the build and calls this skill as its own closing review before committing.
 - [to-spec](https://aihero.dev/skills-to-spec) and [to-tickets](https://aihero.dev/skills-to-tickets) produce the document the Spec axis checks against; a vague spec makes that axis vague.
 - [improve-codebase-architecture](https://aihero.dev/skills-improve-codebase-architecture) is the whole-codebase counterpart — this skill only ever looks at one diff.
 [ask-matt](https://aihero.dev/skills-ask-matt) routes across the whole set when you are unsure which skill the situation wants.
@@ -1,88 +0,0 @@
 ## What it does
 `codebase-design` fixes the words you use to design a module: **module**, **interface**, **depth**, **seam**, **adapter**, **leverage**, **locality**. It defines each one precisely, bans the loose substitutes ("component", "service", "API", "boundary"), and states the handful of principles that follow from them.
 It is a reference, not a process. There is no loop to run, no artifact it produces, no checkpoint where it asks you a question. Every other skill that touches design borrows its vocabulary; on its own it gives you the language and stops. That is the thing to know before you invoke it, because a skill with no process and no stopping rule will improvise one if you point a [session](https://www.aihero.dev/ai-coding-dictionary/session) at it and say "go" — see the questions below.
 ## When to reach for it
 Type `/codebase-design`, or the agent reaches for it automatically when a design task fits.
 Reach for it when you already know which code you're redesigning and you need to think about its shape: where the seam goes, how small the interface can get, whether an extraction is earning its keep. It is also what you reach for to settle an argument about what a word means.
 Several skills sit close to it. Which one you want depends on what the actual problem is:
 | The problem | The skill |
 |---|---|
 | The shape of one module — its interface, its seam, its depth | `codebase-design` |
 | The *words of the domain* — "account" means three things, two people mean different things by "cancellation" | [domain-modeling](https://aihero.dev/skills-domain-modeling) |
 | You don't yet know *which* module to redesign | [improve-codebase-architecture](https://aihero.dev/skills-improve-codebase-architecture) — the survey that finds candidates |
 | You want the design argued with, not just named | [grilling](https://aihero.dev/skills-grilling) |
 | There's a concrete behaviour to build and you want tests that survive a refactor | [tdd](https://aihero.dev/skills-tdd) |
 ## The vocabulary
 The glossary is the skill. Every term is defined against the others, and each one comes with the word it replaces.
 | Term | What it means | Don't say |
 |---|---|---|
 | **Module** | Anything with an interface and an implementation. Deliberately scale-agnostic — a function, a class, a package, a slice spanning tiers. | unit, component, service |
 | **Interface** | Everything a caller must know to use it correctly: the type signature, plus invariants, ordering constraints, error modes, required config, performance characteristics. | API, signature |
 | **Depth** | Leverage at the interface — how much behaviour a caller or a test can exercise per unit of interface they have to learn. **Deep**: a lot of behaviour behind a small interface. **Shallow**: the interface is nearly as complex as the implementation. | — |
 | **Seam** | Michael Feathers' term: a place you can alter behaviour without editing in that place. It is the *location* of an interface, and where to put it is its own decision, separate from what goes behind it. | boundary |
 | **Adapter** | A concrete thing satisfying an interface at a seam. Names a role, not a substance — an in-memory fake and a Postgres repo are both adapters. | — |
 | **Leverage** | What callers get from depth: more capability per unit of interface learned. | — |
 | **Locality** | What maintainers get from depth: change, bugs and verification concentrate in one place. Fix once, fixed everywhere. | — |
 Depth is deliberately *not* defined as the ratio of implementation lines to interface lines, which is Ousterhout's own definition. That metric rewards padding the implementation. Depth-as-leverage is used instead.
 ## The four principles
 - **Depth is a property of the interface, not the implementation.** A deep module can be built internally from small swappable parts. They just don't surface to callers. A module can have internal seams its own tests use, and one external seam at its interface.
 - **The deletion test.** Imagine deleting the module. If complexity vanishes, it was a pass-through. If it reappears across N callers, it was earning its keep.
 - **The interface is the test surface.** Callers and tests cross the same seam. If you want to test *past* the interface, the module is the wrong shape.
 - **One adapter means a hypothetical seam. Two adapters means a real one.** Don't cut a seam until something actually varies across it. A single-adapter seam is just indirection.
 Two supporting files go further, and the skill reads them on demand rather than up front. [DEEPENING.md](https://github.com/mattpocock/skills/blob/main/skills/codebase-design/DEEPENING.md) classifies a candidate's dependencies — in-process, local-substitutable, remote-but-owned, true-external — because the category decides how the deepened module gets tested across its seam. [DESIGN-IT-TWICE.md](https://github.com/mattpocock/skills/blob/main/skills/codebase-design/DESIGN-IT-TWICE.md) spins up parallel [sub-agents](https://www.aihero.dev/ai-coding-dictionary/subagent) to produce three or more radically different interfaces for the same module, then compares them on depth, locality and seam placement.
 ## Common questions
 **How do I actually build a deep module in TypeScript?**
 This is the most-asked question about the skill and the skill does not answer it. It defines what a deep module *is*; it says nothing about how to stop a stray import from reaching past the interface. [Issue #458](https://github.com/mattpocock/skills/issues/458) put it plainly: "let's say we're happy with the interface, it hides the details, etc. But how do we enforce it? I think without linting or clear guardrails, humans and LLMs alike will start making it messy over time." Matt's answer, in that thread, was three options: wrap it in a class or IIFE and accept that the class gets enormous; make it a package in a monorepo and accept the monorepo tooling; or use a linter like [dependency-cruiser](https://github.com/sverweij/dependency-cruiser) to forbid imports that bypass the interface. He has separately called Effect the best mechanism and dependency-cruiser the second-best. There is a `setup-ts-deep-modules` skill in the repo's `drafts/` folder that lays down a `src/packages/<name>/index.ts` convention, but it is a beta-channel skill with no docs page, and it has no lint rule shipped with it.
 **I pointed a session at it and it burned 100k [tokens](https://www.aihero.dev/ai-coding-dictionary/token) redesigning things I never asked about.**
 Known, and filed as [issue #449](https://github.com/mattpocock/skills/issues/449). The skill is model-invoked and describes itself as vocabulary, but nothing in it hard-stops an agent from treating it as a runnable process. Told to "resume in /codebase-design and drive the open decisions", an agent reached for the most action-shaped content it could find — the parallel sub-agents in `DESIGN-IT-TWICE.md` — re-explored code a previous session had already mapped, and ran a long way before asking anything. None of the guardrails a driver skill has (checkpoints, one question at a time, no auto-advance) are present here, because a reference has none. The workaround is to name a driver skill and let this one sit underneath it: `/grill-with-docs`, `/improve-codebase-architecture` or `/tdd` with `codebase-design` as the vocabulary. The issue is open.
 **Where did `design-an-interface` go? And is there an `/interface-design` skill?**
 `design-an-interface` was removed and absorbed into this skill. Nothing was lost: its "design it twice" technique — parallel sub-agents generating radically different designs, from Ousterhout — ships here as `DESIGN-IT-TWICE.md`. Separately, several people have asked for a dedicated `/interface-design` skill for the deep-module/thin-interface philosophy; that philosophy already lives here, and no separate skill is planned. If you came looking for either name, this is the page.
 **Isn't this a file-structure convention — folders, barrel files, feature slices?**
 No, and the skill has held that line under repeated pushback. [Issue #95](https://github.com/mattpocock/skills/issues/95) proposed a formalised fractal-tree file structure as the concrete implementation of deep modules; the reply was that the two are orthogonal — "deep modules are about the design of the interface and accessing through a strict interface, no matter what the file system looks like. It seems perfectly possible that you could have shallow modules with this approach." The same came up in #458: "I think you might be tying the concept of modules too closely to the file system. The file system can certainly be a useful hint to the shape of modules, but there's no need to use the file system in the construction of deep modules." The glossary defines **module** as scale-agnostic on purpose.
 **Does `tdd` actually use this vocabulary?**
 It does now. For a long time it did not. The inline deep-module notes that used to live inside `tdd` were removed in v1.0 in favour of this shared skill, but the pointer replacing them was never added — so `tdd` defined "seam" for itself and referenced nothing. The gap is closed: the pointer is now in the skill, reached when the shape of the interface is the open question rather than the tests. `tdd` still owns "seam" as the boundary you *test* at; this skill owns the module shape behind it.
 **Does the design-it-twice pattern work outside Claude Code?**
 Not cleanly. `DESIGN-IT-TWICE.md` says "spawn 3+ sub-agents in parallel using the Agent tool", which is Claude Code's [tool](https://www.aihero.dev/ai-coding-dictionary/tool) by Claude Code's name. The repo ships metadata for other [harnesses](https://www.aihero.dev/ai-coding-dictionary/harness), including Codex, and those may expose nothing under that name — so the parallel-design phase is less portable than the skill's metadata suggests. Tracked in [issue #564](https://github.com/mattpocock/skills/issues/564), open.
 **Can I add my own concepts to the glossary — connascence, module secrets, [progressive disclosure](https://www.aihero.dev/ai-coding-dictionary/progressive-disclosure)?**
 People have proposed exactly those. [Issue #180](https://github.com/mattpocock/skills/issues/180) adds Parnas's module secrets and Page-Jones's connascence as a naming layer for *what* is leaking across a seam, with a working diff attached; [issue #303](https://github.com/mattpocock/skills/issues/303) proposes progressive disclosure inside the implementation, so a module that is deep at its public interface isn't one undifferentiated slab underneath. Both are open and unmerged. The glossary as shipped is deliberately small, and the reason it stays small is stated in the skill itself: consistent language is the whole point, and a term nobody uses consistently is worse than no term.
 ## It's working if
 - The design conversation stops producing the words "component", "service" and "boundary", and starts producing "module", "interface" and "seam".
 - Someone can point at a proposed extraction and say whether it passes the deletion test, without hedging.
 - A proposed seam comes with a second adapter named, not just the first one.
 - Discussion of an interface covers invariants, ordering and error modes — not only the type signature.
 - Invoking it does not start a session. If the agent begins reading files and proposing refactors off the back of `/codebase-design` alone, it has mistaken the reference for a driver.
 ## Where it fits
 `codebase-design` is a **reach-for-it-anytime standalone**, and the vocabulary layer underneath the engineering skills rather than a step in any chain. Its closest neighbour is [domain-modeling](https://aihero.dev/skills-domain-modeling), the parallel reference for the *problem domain*'s words rather than the module's shape — the two are usually wanted together, since naming a deep module well needs both. [improve-codebase-architecture](https://aihero.dev/skills-improve-codebase-architecture) is the other: it surveys a codebase for deepening candidates and writes every one of them in this glossary, so it finds the module and this skill is the bench you design it on. When you're unsure which skill or flow fits, [ask-matt](https://aihero.dev/skills-ask-matt) routes you.
@@ -1,93 +0,0 @@
 ## What it does
 `diagnosing-bugs` runs a six-phase diagnosis on a hard bug or a performance regression: build a repro, minimise it, rank hypotheses, instrument, fix with a regression test, clean up.
 It will not let the agent form a theory until a **tight** feedback loop exists — one named command, already run once, that goes red on *this* bug and green when it is fixed. The default behaviour of a coding agent handed a bug report is to read code and guess; this skill blocks that. If no red-capable command exists, there is no Phase 2. That single gate is what the skill is for. Everything after it — bisection, hypothesis-testing, instrumentation — is mechanical once the signal exists.
 ## When to reach for it
 Type `/diagnosing-bugs`, or the agent reaches for it on its own when a task fits — it is model-invoked, and fires on "diagnose" / "debug this" or on a report that something is broken, throwing, failing, or slow.
 Reach for it on the hard ones: a bug that resists a first look, an intermittent flake, a regression that crept in between two known-good states. It is heavy by design, and the wrong tool for a question you want answered in one message.
 | Your situation | Where to go |
 | --- | --- |
 | A specific defect you can describe as a symptom | This skill |
 | A slow endpoint or a timing regression with a known before-and-after | This skill — it has a performance branch (measure a baseline, then bisect) |
 | "Where are the bottlenecks in this codebase?" — no specific symptom | Not this skill. It diagnoses one known failure, it does not audit |
 | A raw bug report from someone else, not yet confirmed or written up | [triage](https://aihero.dev/skills-triage) first |
 | Throwaway code to answer a design question, not chase a defect | [prototype](https://aihero.dev/skills-prototype) |
 | Building a planned behaviour test-first | [tdd](https://aihero.dev/skills-tdd) |
 | No good seam exists to lock the bug down | [improve-codebase-architecture](https://aihero.dev/skills-improve-codebase-architecture) — this skill hands off there itself |
 ## The tight loop is the skill
 Phase 1 gets disproportionate effort because it is the only phase that is hard. The skill gives a ladder of ways to construct the loop, roughly in order of preference:
 1. A failing test at whatever seam reaches the bug.
 2. A curl or HTTP script against a running dev server.
 3. A CLI invocation with a fixture input, diffed against a known-good snapshot.
 4. A headless browser script asserting on DOM, console, or network.
 5. A replayed capture — a saved request, payload, or event log, run through the code path in isolation.
 6. A throwaway harness: a minimal subset of the system, one function call.
 7. A property or fuzz loop, for "sometimes wrong output".
 8. A bisection harness you can hand to `git bisect run`.
 9. A differential loop — same input, old version against new.
 10. A [human-in-the-loop](https://www.aihero.dev/ai-coding-dictionary/human-in-the-loop) bash script, last resort. The skill ships `scripts/hitl-loop.template.sh` for this: the agent runs the script, you follow prompts in your terminal, and your answers come back as parseable output.
 *A* loop is not the goal. **Tight** is: fast (seconds), deterministic (same verdict every run), sharp (asserts your exact symptom, not "didn't crash"), and agent-runnable unattended. A 30-second flaky loop is barely better than none. For a bug that only shows up sometimes, the target is not a clean repro but a **higher reproduction rate** — loop the trigger, parallelise, add stress, inject sleeps, until the flake rate is high enough to debug against.
 When it genuinely cannot build one, it is instructed to stop and say so, list what it tried, and ask you for [environment](https://www.aihero.dev/ai-coding-dictionary/environment) access, a captured artifact, or permission to add temporary instrumentation. It should not proceed to hypothesise anyway.
 ## The gates between phases
 The phases are gates, not a checklist. Each one refuses to open until something specific is true.
 | Gate | What has to be true |
 | --- | --- |
 | Into Phase 2 | A named command, already run and pasted with its output, that can go red on this bug |
 | Into Phase 3 | The repro is reproduced *and* minimised — every remaining element is load-bearing |
 | Into Phase 4 | 3–5 ranked, falsifiable hypotheses exist, each stating its prediction, shown to you before any is tested |
 | Into Phase 5 | Probes map to a specific prediction, one variable at a time, every debug log tagged `[DEBUG-a4f2]`-style so cleanup is one grep |
 | Done | Original repro no longer reproduces, instrumentation gone, and the hypothesis that turned out correct is written into the commit message |
 Phase 5 has an escape hatch worth knowing about. The regression test is written before the fix, but only if a **correct seam** exists for it — one where the test exercises the real bug pattern as it occurs at the call site. Where the only available seam is too shallow, the skill is told to say so rather than write a test that gives false confidence. That absence is itself the finding, and it is what routes the post-mortem to `improve-codebase-architecture`.
 ## Common questions
 **It fires on quick questions where I just wanted a direct answer.**
 This is the most-reported problem with the skill, and it is real. On GPT-5.6-Sol especially, users report it triggering on a plain description of a problem: "the model triggers the rather formal diagnosing-bugs skill instead. It then goes on to construct a reproduction scenario — often building a mock scenario with limited value — before giving me a response or suggestion. This results in considerable reply delays." Four separate people reported the same shape on [issue #578](https://github.com/mattpocock/skills/issues/578). The accepted fix is to start with a lighter approach and graduate to the heavier one only where the problem warrants it, but that change has not landed. The skill is calibrated against Claude Code's invocation behaviour; a [model](https://www.aihero.dev/ai-coding-dictionary/model) with a lower activation threshold over-fires it. Until it is graduated, the practical fix is to say what you want ("just answer this, don't diagnose") or to disable model invocation for it in your [harness](https://www.aihero.dev/ai-coding-dictionary/harness).
 **Can I point it at a codebase and ask where the performance problems are?**
 No. It diagnoses one failure you can already name. Its performance branch is for a regression with a symptom — establish a baseline measurement, then bisect, measure first and fix second — not for a proactive sweep. A skill for the proactive version was [proposed and closed](https://github.com/mattpocock/skills/issues/431); there is currently no skill for it.
 **Does it stop and ask me before it writes the fix?**
 No. Only Phase 3 has a human checkpoint — the ranked hypothesis list is shown to you before any is tested, and it proceeds on its own ranking if you are away. There is no gate between instrumentation and the fix, so the agent can start writing code before you have agreed with its root cause. [Issue #124](https://github.com/mattpocock/skills/issues/124) asks for that gate and is still open. If you want it, say so when you invoke the skill.
 **I already ran `/triage` on this bug report. Is this the same work again?**
 Partly, and neither skill admits it. As one reader put it: "Triage's step 3 is essentially a shallow, bounded instance of diagnosing-bugs Phase 1–2, but neither file mentions the other." Triage does a bounded "is this actually a bug, and what is the surface" pass; this skill does the thorough version. Running triage first is not wasted — its verification often gives you most of Phase 1's raw material — but expect to redo it properly here, and expect no cross-reference to tell you that.
 **Will the repro output it pastes leak secrets?**
 It might. The skill asks the agent to paste the invocation and its output, and to request artifacts like HAR files, log dumps, and core dumps. None of those are sanitised by instruction. [Issue #674](https://github.com/mattpocock/skills/issues/674) raises exactly this — credentials, tokens, cookies, and personal data riding along into a chat, an issue, or a PR — and proposes a redaction guardrail. It is open and unimplemented. Treat redaction as your job for now, particularly before the output goes anywhere public.
 **My security scanner flagged this skill as high risk.**
 Snyk flags it, and the flag is a false positive. It is the only skill in the set that ships an executable shell script (`hitl-loop.template.sh`) alongside instructions to run it and to curl a dev server. Shipped `.sh` plus run-it instructions plus outbound HTTP is enough to trip a static scanner. The script itself is about 30 lines of `read -r -p` prompts that pause for human input. The scanner is rating the capability surface, not a proven exploit.
 **What happened to `/diagnose`?**
 Renamed to `/diagnosing-bugs` in v1.0.0. The old name no longer exists. Anything of yours that chains `/diagnose` — a wrapper skill, a saved prompt — needs updating.
 ## It's working if
 - It shows you a command and its red output before it offers a single theory. If theory arrives first, the skill is not running.
 - The failure it reproduces is the one you reported, not a nearby one it found on the way.
 - It shrinks the repro before it starts guessing, and can tell you why each remaining piece is load-bearing.
 - You are shown a ranked list of 3–5 hypotheses, each with a prediction you could falsify, before any of them is tested.
 - Every debug log it adds carries a tag like `[DEBUG-a4f2]`, and a grep for that tag comes back empty when it declares done.
 - The commit or PR message names which hypothesis was right.
 - When it cannot lock the bug down with a test, it says so plainly instead of writing a shallow one.
 ## Where it fits
 `diagnosing-bugs` is a reach-for-it-anytime standalone. You drop into it when something is broken and drop out when the fix and its regression test are in; it holds no state and needs no prior setup. [ask-matt](https://aihero.dev/skills-ask-matt) routes "Something's broken" here.
 Two neighbours matter. [improve-codebase-architecture](https://aihero.dev/skills-improve-codebase-architecture) takes the [handoff](https://www.aihero.dev/ai-coding-dictionary/handoff) when the real finding is that the code has no seam to lock the bug down — the recommendation is made after the fix is in, when there is more information. [triage](https://aihero.dev/skills-triage) sits upstream of it for bugs that arrive as raw reports from other people, and does a shallower version of the same first two phases.
@@ -1,86 +0,0 @@
 ## What it does
 `domain-modeling` builds and sharpens a project's **ubiquitous language** while you are designing — challenging a term that conflicts with the glossary, forcing a precise word where you used a vague one, and stress-testing a relationship with a concrete scenario until the boundaries are exact.
 It is the **active** discipline, not the passive one. Reading `CONTEXT.md` to borrow its vocabulary is a one-line habit any skill can do; this skill is for when you are *changing* the model. That is what makes it interrupt. It writes a resolved term into `CONTEXT.md` at the moment it is resolved, in the middle of the conversation, rather than producing a tidy glossary at the end — because the batched version is a summary of a [session](https://www.aihero.dev/ai-coding-dictionary/session), and the inline version is the session's actual output.
 ## When to reach for it
 Type `/domain-modeling`, or the agent reaches for it automatically when a task fits. In practice, automatic invocation is the weakest part of the skill: when `grill-with-docs` or `wayfinder` say to load it, [models](https://www.aihero.dev/ai-coding-dictionary/model) frequently load `grilling` and skip this one. If a [grilling](https://www.aihero.dev/ai-coding-dictionary/grilling) session runs and `CONTEXT.md` is untouched at the end, that is what happened — invoke it by name alongside the other skill.
 Reach for it when the *words* are the problem:
 | The situation | The move |
 | --- | --- |
 | Two people mean different things by "cancellation" | `domain-modeling` — pick the canonical term, list the other under `_Avoid_` |
 | "Account" is doing three jobs in three files | `domain-modeling` — split it into Customer and User |
 | You just made a hard-to-reverse architectural choice | `domain-modeling` — it offers an ADR, if the choice clears the bar |
 | The module's *shape* is the problem — where the seam goes, how deep the interface is | [codebase-design](https://aihero.dev/skills-codebase-design) |
 | You want the whole plan interrogated before you build | [grill-with-docs](https://aihero.dev/skills-grill-with-docs), which drives this skill underneath |
 | You want a term looked up, not changed | Nothing. Read `CONTEXT.md`. It is a file. |
 ## Prerequisites
 None up front. The skill writes into two places and creates both lazily:
 - **`CONTEXT.md`** at the repo root, created by the first resolved term. In a repo with a `CONTEXT-MAP.md` at the root, terms go into the per-context `CONTEXT.md` the map points at instead.
 - **`docs/adr/`**, created by the first ADR that clears the bar.
 Nothing needs to exist before you start, and nothing is created speculatively.
 ## Two artifacts, two bars
 The glossary and the ADR are held to different standards, and conflating them is where most of the trouble in this skill comes from.
 | | `CONTEXT.md` | `docs/adr/NNNN-slug.md` |
 | --- | --- | --- |
 | Holds | Terms. What a thing **is**, in one or two sentences, with rejected synonyms under `_Avoid_` | One decision, in one to three sentences: context, choice, reason |
 | Bar to write | A vague term became canonical | **All three**: hard to reverse, surprising without context, the result of a real trade-off |
 | Written | Inline, the moment the term is settled | Offered, not assumed |
 | Never holds | Implementation details, a [spec](https://www.aihero.dev/ai-coding-dictionary/spec), a scratch pad, general programming concepts | A diary of every choice made this session |
 Miss any one of the ADR's three tests and there is no ADR. An easily-reversed decision will just get reversed; an unsurprising one is nobody's question; one with no real alternative records that you did the obvious thing.
 The `CONTEXT.md` rule is the one to actually hold onto, because it is the one that breaks in the field. **It is a glossary and nothing else.** Left unchecked, models treat "write to `CONTEXT.md`" as permission to persist every answer you give, and the file turns into a running spec — this is the most-reported problem with the skill, across several models.
 ## Cross-referencing, and where it stops
 The move that makes the skill click: when you state how something works, it checks the code and surfaces the contradiction. *"Your code cancels entire Orders, but you just said partial cancellation is possible — which is right?"* The language and the code are made to agree, out loud, before either is changed.
 The limit is worth knowing. It cross-references **code** and the committed `CONTEXT.md`/ADRs, and nothing else. It does not search your issue tracker, so a naming collision that was argued out and deliberately settled in a closed issue months ago gets surfaced as if it were new. There is [an open request](https://github.com/mattpocock/skills/issues/717) to fix this; until then, the workaround is to put the instruction in your own `docs/agents/domain.md`, which the skills already read.
 ## Common questions
 **My `CONTEXT.md` is 500 lines. 1,000. 3,000. What do I do?**
 The size is a symptom, not the disease — the file has absorbed implementation detail and decisions that were never glossary material. The fix is a direct instruction: `/grill-with-docs make my CONTEXT.md more concise and remove any implementation details from it`. Run it against a bloated file and most of it goes. Only reach for a `CONTEXT-MAP.md` split once the file is genuinely lean and still covers two domains that a reader would not want to hold at once; splitting a bloated file just gives you several bloated files. The skill's guidance here is not yet strong enough to prevent the growth in the first place, and the issue tracking that is still open.
 **Why is it `CONTEXT.md` and not `GLOSSARY.md`?**
 This is the most-argued naming question in the whole skill set and it has no settled answer. The case against the current name is good: if it is "a glossary and nothing else", `GLOSSARY.md` says so, and — as one reader put it — "with ai agents everything is [context](https://www.aihero.dev/ai-coding-dictionary/context)". The case for it is the map: `CONTEXT-MAP.md` pointing at several `CONTEXT.md` files reads naturally in a way `GLOSSARY-MAP.md` does not, and `context` is the standing DDD word for a bounded area of the model. At least one person maintains a local fork purely to rename the file. You can do the same, but every other skill in the set looks for `CONTEXT.md`, so a rename means patching all of them.
 **Where did `/ubiquitous-language` go?**
 It was removed, and it was not deprecated. Its job moved into `domain-modeling`, which maintains the whole model continuously rather than dumping a glossary out of one conversation. Vocabulary enforcement got more load-bearing, not less — it now runs underneath grilling, triage and mapping rather than as a separate pass you remember to do.
 **How do I get a glossary for a codebase that has none?**
 Ask for it explicitly rather than waiting for it to accumulate. `/grill-with-docs help me scaffold my existing repo with a CONTEXT.md` is the documented route; expect a long interrogation — one user reported 50+ questions before the file was in shape. Incidental use builds the glossary far too slowly on a brownfield repo.
 **Can I keep the domain model and use my own ADR format?**
 Not cleanly today. The glossary half and the ADR half ship in one skill, so a team with an established ADR convention — different template, different location, different naming — gets instructions that conflict with its house style. The current options are to copy the skill locally and edit it, or to override the ADR conventions in your repo's own agent docs. Splitting the two apart is [an open request](https://github.com/mattpocock/skills/issues/557).
 **Does a glossary actually earn its keep? It is one more artifact to review, and it can go stale.**
 Sometimes it does not, and it is worth being honest about where. DDD gets less useful the closer it gets to the implementation — the payoff is upstream, in naming and concept alignment, not in aggregates and layer ceremony. Synonym control matters at naming boundaries: module names, table names, status enums, issue titles, CLI commands. It matters much less in ordinary prose. There is also a live objection that domain terms compress communication *between humans* who already share them, and that an agent responds the same way to the plain-English description — on that reading, the glossary's value is keeping you and your reviewers aligned with what the agent is doing, not making the agent better. On a one-day build, skip it. And an unreviewed, agent-authored glossary is worse than none: it becomes confident-sounding lore that later sessions treat as truth.
 **Can it turn my vague prompts into domain language for me?**
 No, and there is no plan for a skill that does. A domain language you do not understand yourself becomes meaningless drivel once written down. This skill enforces precision once you have the understanding — it does not manufacture vocabulary you do not have. The related trap is using domain words without doing the modelling: right nouns over the wrong conceptual structure produce output that reads correct and is not.
 ## It's working if
 - It stops you mid-sentence to ask which of two things you meant, instead of picking one and moving on.
 - `CONTEXT.md` changes **during** the conversation, not in a burst at the end.
 - It refuses to write an ADR for something you could undo tomorrow — and says which of the three tests failed.
 - New entries define what a thing *is* in one or two sentences and name the words you are giving up under `_Avoid_`.
 - It quotes your code back at you when your code and your sentence disagree.
 - `CONTEXT.md` gets shorter as often as it gets longer.
 ## Where it fits
 `domain-modeling` is a **model-invoked reference** that runs *underneath* other skills more often than it runs on its own. [grill-with-docs](https://aihero.dev/skills-grill-with-docs) drives it through a grilling session, [wayfinder](https://aihero.dev/skills-wayfinder) loads it while charting a map, [triage](https://aihero.dev/skills-triage) uses it to keep [tickets](https://www.aihero.dev/ai-coding-dictionary/ticket) in the project's own words, and [improve-codebase-architecture](https://aihero.dev/skills-improve-codebase-architecture) calls it as decisions crystallise. Its closest sibling is [codebase-design](https://aihero.dev/skills-codebase-design): the two are the vocabulary layer under everything else, this one for the *domain*, that one for the module's *shape*. It is also reachable directly, when you want the discipline without committing to the steps of whatever skill would normally pull it in. When you are unsure which skill fits, [ask-matt](https://aihero.dev/skills-ask-matt) routes you.
@@ -1,88 +0,0 @@
 ## What it does
 `grill-with-docs` interviews you about a plan or design until you and the [agent](https://www.aihero.dev/ai-coding-dictionary/agent) share one understanding of it, and writes the vocabulary and the hard decisions into your repo while it does. It is the same interview [grill-me](https://aihero.dev/skills-grill-me) runs — a round of questions, then wait, then the next round — pointed at a codebase.
 It is **[stateful](https://www.aihero.dev/ai-coding-dictionary/stateful)**. Every other grilling skill leaves the [session](https://www.aihero.dev/ai-coding-dictionary/session) in your head; this one leaves files on disk. A term gets resolved and it lands in `CONTEXT.md` the moment it resolves, not batched at the end. A decision passes three gates and it lands as an ADR. That is the whole difference, and it is also the source of most of the trouble people have with the skill: the artifacts are real files in a real repo, so they can be absent when you expected them, and they can drift when more than one person is writing them.
 ## When to reach for it
 You invoke this by typing `/grill-with-docs` — the agent will not reach for it on its own.
 Reach for it at the start of a change, in a repo, when the plan is still fuzzy and the words for the thing are not settled yet. It is the single-session tool. Which grilling skill you want depends on what is in front of you:
 | What you have | Reach for |
 | --- | --- |
 | You aren't working in a working directory at all | [grill-me](https://aihero.dev/skills-grill-me) |
 | A repo, and a change you can settle in one session | `grill-with-docs` |
 | An effort too big to hold in one session — a greenfield build, a large feature | [wayfinder](https://aihero.dev/skills-wayfinder) |
 | A repo with no domain docs at all, and no particular feature in mind | `grill-with-docs`, aimed at the repo rather than a change |
 | A decision blocked on knowledge in someone else's head | [to-questionnaire](https://aihero.dev/skills-to-questionnaire) |
 The wayfinder split comes down to session count: `/grill-with-docs` for single-session planning, `/wayfinder` for multi-session planning.
 ## Prerequisites
 The skill writes into your repo, so you need to be somewhere it is safe to write. Resolved terms go to a `CONTEXT.md` glossary at the root — or to the relevant context's `CONTEXT.md`, if a `CONTEXT-MAP.md` at the root marks the repo as multi-context. Decisions go to `docs/adr/`. Both are created lazily; nothing exists until the first term or decision crystallises, so there is nothing to scaffold up front.
 It also needs two other skills present, because its own `SKILL.md` is one line that delegates to them: [grilling](https://aihero.dev/skills-grilling) supplies the interview, [domain-modeling](https://aihero.dev/skills-domain-modeling) supplies the writing. Installing `grill-with-docs` alone gets you a skill that does not work.
 ## The paper trail
 Three things come out of a session, and they are not equal.
 | What resolved | Where it lands |
 | --- | --- |
 | A term — the project's own word for a thing | `CONTEXT.md`, inline, the moment it resolves |
 | A decision that is hard to reverse, surprising without context, and a real trade-off | An ADR under `docs/adr/` |
 | Everything else you decided | The conversation, and nowhere else |
 That third row is the one that catches people out. `CONTEXT.md` is a glossary and is deliberately kept as one — no implementation details, no [spec](https://www.aihero.dev/ai-coding-dictionary/spec), no scratch notes. ADRs are gated on all three conditions at once, so most decisions do not qualify and most sessions produce none. A session that yields a sharper glossary and zero ADRs is working as designed, but it means the bulk of what you agreed exists only in the [context window](https://www.aihero.dev/ai-coding-dictionary/context-window) you agreed it in. Hand that same conversation to [to-spec](https://aihero.dev/skills-to-spec) rather than [clearing](https://www.aihero.dev/ai-coding-dictionary/clearing) it.
 The glossary is the point. Domain language is the thing this skill is actually building — the project's own words, agreed once, so you, the agent and your colleagues stop paying to re-derive them. It is worth saying that not everyone agrees this buys you agent performance: the sharpest public pushback is that a term and its plain-English expansion get the same result from the [model](https://www.aihero.dev/ai-coding-dictionary/model), and that the vocabulary really compresses communication between the humans who share it. That reading still leaves the glossary valuable; it just moves the value.
 ## It assumes one writer
 The stateful outputs assume a single person curating them. A two-developer team running four months in one repo reported state drift on roughly 20% of sampled merged PRs, with ADR citations and README claims the highest-drift surfaces — deliberate, human-curated docs drifted worse than agent memory did. Pruning the stale docs did not hold; the same sweep was stale again within days. What worked was deleting shadow state outright and adding a deterministic citation and link linter to CI.
 Related: running the skill repeatedly across unrelated changes in one repo tends to accumulate mixed-topic docs, because nothing separates one session's output from another's. Neither of these is fixed in the skill today.
 ## Common questions
 **Should I use this or `/wayfinder`?**
 Scope decides it. Use this for anything you can settle in one session; use [wayfinder](https://aihero.dev/skills-wayfinder) when the effort is too big to hold in one, and it charts the work as a map of decision [tickets](https://www.aihero.dev/ai-coding-dictionary/ticket) first. Wayfinder is slower and denser, and reaching for it on a well-scoped feature is the common mistake. It does not replace this skill — it can drop into a grilling session for the parts of the map that suit one.
 **It ran, but no `CONTEXT.md` and no ADRs appeared.**
 Two known causes. The mundane one: nothing qualified. ADRs need all three gates, and a session about a change with no new vocabulary genuinely has nothing to write. The real bug: when the skill runs inside another orchestration layer — a spec-driven-development wrapper, a multi-agent framework, a rule that invokes it as a step in someone else's pipeline — the file-writing half is reported to silently not happen, while the interview still runs. This is filed and unfixed. If you are in that setup, check the working directory before you trust the session's output.
 **It asked everything at once, with no recommendations, and never mentioned `CONTEXT.md`.**
 That is the skill failing to load its two dependencies. Because `SKILL.md` is a one-line delegation, an agent that does not pick up [grilling](https://aihero.dev/skills-grilling) and [domain-modeling](https://aihero.dev/skills-domain-modeling) guesses at what grilling means, and you get an undifferentiated question dump. Partial loading is the more confusing case — `grilling` loads, `domain-modeling` does not, and you get a good interview with no paper trail. It correlates with model and [effort](https://www.aihero.dev/ai-coding-dictionary/effort) level, and it is the most reported problem with this skill. If you suspect it, ask the agent directly which skills it loaded.
 **Where did all my other decisions go?**
 Into the conversation only. This is the most substantive open complaint about the skill: the glossary is not a spec, most answers do not earn an ADR, and there is no ledger tying each resolved answer through to a spec, a ticket and a test. Precise answers — ordering guarantees, negative requirements, numeric defaults — get softened into weaker prose downstream, and the result can look complete while missing the thing you actually decided. The mitigation available today is to keep the session and feed it straight to [to-spec](https://aihero.dev/skills-to-spec), and to re-read the spec against your own answers rather than assuming it captured them.
 **Can I point it at an existing repo that has no docs at all?**
 Yes. This is the right skill for a codebase with no ADRs, no domain language and no design principles — invoke it and say "help me document my repo". The community pattern pairs it with [improve-codebase-architecture](https://aihero.dev/skills-improve-codebase-architecture) for building or repairing a `CONTEXT.md`. Expect to steer it: it will read code and ask you about what it finds, and you are the one who says which of the words already in the codebase are the right ones.
 **What should I do when the session ends?**
 The skill's closing message tends to be open-ended, which is a known rough edge. In the main flow the answer is [to-spec](https://aihero.dev/skills-to-spec), in the same conversation. If the change is small enough to build immediately, go straight to [implement](https://aihero.dev/skills-implement) instead.
 **Why is it called that?**
 Nobody is happy with the name. There is an open suggestion to rename it `grill-domain-model`, which describes the behaviour more honestly. Nothing has moved on it. If a rename ever lands, the docs page moves with it and the URL changes.
 ## It's working if
 - `CONTEXT.md` changes *during* the session, term by term, rather than appearing in one lump at the end.
 - The glossary reads as pure vocabulary — your project's words with tight definitions — and contains no implementation detail or spec-like prose.
 - Questions the codebase can answer get answered by reading the codebase, not asked of you.
 - You get few or no ADRs, and the ones you get are decisions you would be annoyed to have to re-litigate.
 - It challenges a word you used because your existing glossary defines it differently.
 ## Where it fits
 `grill-with-docs` is the head of the main build chain:
 ```txt
 grill-with-docs → to-spec → to-tickets → implement → code-review
 ```
 It comes before anything is written down as a spec — it produces the shared understanding and settled vocabulary that [to-spec](https://aihero.dev/skills-to-spec) then synthesises without interviewing you again. Its close neighbours are [grill-me](https://aihero.dev/skills-grill-me), the same interview with no repo and no files, and [domain-modeling](https://aihero.dev/skills-domain-modeling), the glossary-and-ADR discipline it drives; both sit on the [grilling](https://aihero.dev/skills-grilling) primitive. Upstream of it, [wayfinder](https://aihero.dev/skills-wayfinder) charts efforts too large for one session and can hand parts of the map back down to it. When you're unsure which skill or flow fits, [ask-matt](https://aihero.dev/skills-ask-matt) routes you.
@@ -1,97 +0,0 @@
 ## What it does
 `implement` builds work that has already been decided. You point it at a [ticket](https://www.aihero.dev/ai-coding-dictionary/ticket), a [spec](https://www.aihero.dev/ai-coding-dictionary/spec), or the plan you just agreed in the conversation, and it writes the code, drives [tdd](https://aihero.dev/skills-tdd) at the seams, typechecks as it goes, runs [code-review](https://aihero.dev/skills-code-review) at the end, and commits to the current branch.
 It never reopens the plan. There is no interview, no clarifying round, no proposal of a different approach. Whatever was settled upstream is the input, and the skill's whole job is to turn that into a commit. That is what separates it from typing "build this" at a fresh [agent](https://www.aihero.dev/ai-coding-dictionary/agent), which will happily redesign the work while it builds it.
 ## When to reach for it
 You invoke this by typing `/implement` — the agent won't reach for it on its own. It ships with `disable-model-invocation: true`, so no other skill can call it either. Wherever [ask-matt](https://aihero.dev/skills-ask-matt) or [to-tickets](https://aihero.dev/skills-to-tickets) says "then `/implement` per ticket", that is an instruction to you, not something the agent will do unprompted.
 Where the work currently lives decides whether this is the right skill:
 | The work is… | Reach for |
 | --- | --- |
 | A ticket on the tracker | `/implement #42`, one ticket per [session](https://www.aihero.dev/ai-coding-dictionary/session), [clearing](https://www.aihero.dev/ai-coding-dictionary/clearing) context between tickets |
 | A spec, not yet split up, and the build spans sessions | [to-tickets](https://aihero.dev/skills-to-tickets) first, then `/implement` per ticket |
 | A spec, and the build is small | `/implement` directly against the spec |
 | Only in the conversation you just had, and it's still small | `/implement` right there, in the same window |
 | Not written down anywhere yet | [grill-with-docs](https://aihero.dev/skills-grill-with-docs), or [grill-me](https://aihero.dev/skills-grill-me) if there's no codebase |
 | One concrete behaviour you want test-first, with no spec | [tdd](https://aihero.dev/skills-tdd) directly |
 | Already built, and you want it checked | [code-review](https://aihero.dev/skills-code-review) directly |
 The same-session case is worth naming because the skill's own first line doesn't cover it. `SKILL.md` says "the spec or tickets", which nudges the [model](https://www.aihero.dev/ai-coding-dictionary/model) to go hunting for a file that doesn't exist. If the plan lives only in the thread, say so when you invoke it.
 ## Prerequisites
 `implement` commits to the branch you are on. It does not create one, and it does not ask. Check you are on the branch you want the work on before you start.
 If the tickets came from [to-tickets](https://aihero.dev/skills-to-tickets), the tracker they live on was configured by [setup-matt-pocock-skills](https://aihero.dev/skills-setup-matt-pocock-skills). `code-review` reads the same configuration to find the originating spec at close-out.
 ## What one run does
 A run is five beats, in order:
 1. Read the ticket or spec and work out the seams.
 2. Drive [tdd](https://aihero.dev/skills-tdd) at the pre-agreed seams, one red-green slice at a time.
 3. Typecheck often, run single test files as it goes.
 4. Run the full test suite once, at the end.
 5. Run [code-review](https://aihero.dev/skills-code-review), then commit to the current branch.
 One run covers one ticket. The tickets [to-tickets](https://aihero.dev/skills-to-tickets) produces are tracer-bullet vertical slices sized to fit a single fresh [context window](https://www.aihero.dev/ai-coding-dictionary/context-window), so the intended rhythm is: clear context, implement one ticket, commit, clear again. Each ticket is self-contained, which is what makes the previous ticket's context disposable.
 ## Pre-agreed seams
 The idea the skill runs on is the **seam**: the public boundary you observe behaviour at, without reaching inside. Tests live at seams. Working at a seam agreed before any code is written is what keeps the tests durable, because the implementation underneath can be rewritten without the tests moving.
 The word "pre-agreed" is doing real work, and it is also the skill's weakest joint. Nothing inside `implement` agrees the seams. `tdd` is the skill that asks, and it refuses to write a test at an unconfirmed seam. So in practice the agreement happens either upstream in the spec, or in the first exchange of the run. If it happens nowhere, the precondition never fires and the run quietly becomes "just write the code". Naming the seams in the spec is what stops that.
 ## Common questions
 **It finished, but my ticket is still open and the acceptance criteria are still unchecked.**
 Correct, and expected. `implement` has no completion step. It ends at the commit and never touches the work item, confirmed on GitHub Issues and on the local markdown tracker, so it is not a tracker integration problem. It also does not act on the findings `code-review` produced, and does not tick the `- [ ]` boxes on the originating issue. Close the ticket and reconcile the criteria yourself. This bites hardest on a dependency chain, because `to-tickets` defines the frontier as tickets whose blockers are all closed. If nothing gets closed, nothing ever becomes visibly unblocked.
 **Can I point it at all my tickets at once, or run several in parallel?**
 No. One invocation, one ticket. Batch dispatch across a ticket queue and [subagent](https://www.aihero.dev/ai-coding-dictionary/subagent) fan-out are both requested repeatedly, and neither exists. Running several `/implement` sessions side by side in one checkout is worse than unsupported: one field report describes a `git commit --amend` in one session landing on another session's commit, a stash vanishing from `refs/stash`, and commits landing on the wrong branch, all in a single afternoon across three issues. The sessions share one working directory, one index, and one HEAD. Git worktrees are the community workaround, and note that `refs/stash` is shared across worktrees too, so worktrees alone do not fix the stash case. If you want parallelism today, you are assembling it yourself.
 **Can it open a pull request instead of committing?**
 Not built in. It commits straight to the current branch, which several people find too eager: the code lands before they have had a chance to verify it works. There is no configuration flag and no PR mode. People override it in the invocation ("commit to a branch and open a PR") or by editing their local copy of the skill.
 **`code-review` says it cannot see my changes.**
 `code-review` reviews `git diff <fixed-point>...HEAD`, which excludes staged and working-tree changes. `implement` runs it before committing, so unless an interim commit already exists there is nothing in that diff to review. Multiple people have reported this and it is unfixed on both sides. Commit first, then review against the point you branched from.
 Separately, some people deliberately do not want the review inside the run at all, because an agent reviewing the code it just wrote is biased toward its own solution. Running [code-review](https://aihero.dev/skills-code-review) in a fresh session against a fixed point is a legitimate alternative, and is the same reason that skill runs its two axes in separate sub-agents.
 **One ticket burned 150k tokens. Am I using it wrong?**
 Probably the ticket is too big rather than the skill being misused. A run does codebase exploration, a red-green loop per seam, a full suite, and a review, so a non-trivial ticket exceeding 100k [tokens](https://www.aihero.dev/ai-coding-dictionary/token) is normal rather than a sign something broke. The lever is upstream: right-size the tickets in [to-tickets](https://aihero.dev/skills-to-tickets) so each fits one fresh window. If a single ticket keeps blowing out, split it rather than raising the [effort](https://www.aihero.dev/ai-coding-dictionary/effort) level.
 **`/implement #2` in a fresh session worked on something completely unrelated.**
 `#2` is resolved against whatever numbered list the agent can see, which in a fresh session may be a todo file, a checklist, or another work list rather than the configured tracker. The resolution is confident rather than fail-closed, so the mistake is not obvious until it has started. Pass the full reference, the issue URL or `owner/repo#2`, and ask it to confirm the title back before it begins.
 ## It's working if
 - The session opens by reading the ticket or spec and restating what it will build, rather than asking you what to build.
 - You can see an actual `/tdd` invocation in the trace, not just tests appearing in the diff.
 - Typechecks and single test files run repeatedly during the run, and the full suite runs once near the end.
 - The run reaches a commit on your current branch without you prompting it to carry on.
 - The diff is one ticket's worth of change: a vertical slice through every layer, not several tickets swept together.
 ## Where it fits
 `implement` is the build step of the main chain, second from the end:
 ```txt
 grill-with-docs → to-spec → to-tickets → implement → code-review
 ```
 Its neighbours are [to-tickets](https://aihero.dev/skills-to-tickets), which produces the tickets it consumes and declares the blocking edges that decide their order; [tdd](https://aihero.dev/skills-tdd), which it drives internally at each seam; and [code-review](https://aihero.dev/skills-code-review), which it runs before committing. It sits downstream of the planning skills and trusts them. It does not re-validate the shape of what it was handed, so a badly-structured map or a horizontally-layered ticket gets built as written.
 That trust is why [wayfinder](https://aihero.dev/skills-wayfinder) merges onto the chain at [to-spec](https://aihero.dev/skills-to-spec) rather than looping its map straight into `implement`. Go straight to `implement` from a map only when the effort turned out genuinely small.
 [ask-matt](https://aihero.dev/skills-ask-matt) is the router over the whole set when you are not sure which flow you are in.
@@ -1,101 +0,0 @@
 ## What it does
 `improve-codebase-architecture` surveys a codebase for **deepening opportunities** — places where a shallow module (an interface nearly as complex as the thing it hides) could become a deep one — writes them up as a self-contained HTML report, and then [grills](https://www.aihero.dev/ai-coding-dictionary/grilling) you through whichever one you pick.
 It never changes the code. The whole run produces one HTML file in your OS temp directory and a conversation; the refactor itself happens later, in a separate [session](https://www.aihero.dev/ai-coding-dictionary/session), through the normal build flow. That is what makes it a survey rather than a refactoring tool, and it is why the skill is worth running on a codebase you are not ready to touch yet.
 Two filters keep the report from becoming generic cleanup advice. Every candidate has to pass the **deletion test** — would removing this module concentrate complexity behind a smaller interface, or just spread it across callers? Only the "concentrates" cases earn a card. And unless you point it at a specific area, it reads recent commit history first and biases the scan toward paths that are actively changing, on the grounds that a deepening in code nobody touches is a refactor you will never cash in.
 ## When to reach for it
 You invoke this by typing `/improve-codebase-architecture` — the [agent](https://www.aihero.dev/ai-coding-dictionary/agent) will not reach for it on its own.
 It sits outside the build loop — it is not a step in the main loop but something you run periodically to queue up more work to improve the codebase. The four situations it gets used in:
 | Situation | How it is used |
 | --- | --- |
 | Routine upkeep | Run it every few days, or whenever a spare moment appears, to stop structure rotting between features. |
 | Before a big build | Point it at the [spec](https://www.aihero.dev/ai-coding-dictionary/spec): "how can we make this change easy?" This is the most effective prompt for it. |
 | Brownfield audit | Run it on a large, unstructured or [vibe-coded](https://www.aihero.dev/ai-coding-dictionary/vibe-coding) repo to find out what shape it is actually in. |
 | Legacy test work | Use it to find the missing seams first, before writing tests against untestable code. |
 Where it is confusable with siblings:
 - For designing one module you have already chosen, use [codebase-design](https://aihero.dev/skills-codebase-design) — that is the bench, this is the survey that finds what to put on it.
 - For a whole effort too big to hold in one session, use [wayfinder](https://aihero.dev/skills-wayfinder).
 - For "this specific thing is broken," use [diagnosing-bugs](https://aihero.dev/skills-diagnosing-bugs). It hands back here when the real finding is that there is no good seam to lock the bug down.
 ## Prerequisites
 None to run it. It reads `CONTEXT.md` and any ADRs in `docs/adr/` if they exist, and speaks in your domain's own nouns when they do — a candidate reads as "deepen the Order intake module," not "refactor the FooBarHandler."
 It writes in two places. The report goes to `<tmpdir>/architecture-review-<timestamp>.html`, outside the repo. During the grilling loop it will add or sharpen terms in `CONTEXT.md`, creating that file if it does not exist, and offer to record a rejected candidate as an ADR so a future run does not re-suggest it.
 ## Depth, and the report that hunts for it
 The skill turns on one idea: **depth**. A deep module puts a lot of behaviour behind a small, stable interface. A shallow one leaks its implementation through an interface nearly as wide as the code beneath it. The report is a hunt for shallowness — pure functions extracted only for testability while the real bugs live in how they are called (no **locality**), modules leaking across their **seams**, a concept you cannot understand without opening five files — and a proposal for the deepening that fixes it.
 Each candidate is a card: the files involved, the friction, a plain-English solution, the benefit stated in terms of **locality** and **leverage**, a before/after diagram, and a strength badge.
 | Badge | What it means for you |
 | --- | --- |
 | `Strong` | The deletion test passes clearly and the friction is real. Take these seriously. |
 | `Worth exploring` | Plausible deepening, but the payoff depends on where the code is going next. |
 | `Speculative` | Surfaced for completeness. Most of these are safe to ignore. |
 The report ends with a **Top recommendation** — the one it would tackle first — and then the skill stops and asks which candidate you want to explore. Nothing has been decided at that point, and no code has moved.
 ## What happens after you pick one
 Picking a candidate starts a [grilling](https://aihero.dev/skills-grilling) session over it: constraints, what sits behind the seam, which tests survive, what the deepened interface should look like. The output of that session is a decision, not a diff. From there the normal flow applies — take the decision into [to-spec](https://aihero.dev/skills-to-spec), then [to-tickets](https://aihero.dev/skills-to-tickets), then [implement](https://aihero.dev/skills-implement).
 ## Common questions
 **It grilled me for an hour about one idea instead of showing me options. Can I turn that off?**
 Yes — say so when you invoke it ("don't grill me, just show the report"). This is the loudest complaint the skill has. One user put it bluntly: they liked it as "a convenient way to get a thorough analysis of improvements," and after the grilling loop was added found it "borderline unusable," reporting sessions where it proposed a single solution and then asked "10's or 100's of questions." The design intent is that the report comes first and the grill only starts on a candidate you chose, but weaker [models](https://www.aihero.dev/ai-coding-dictionary/model) skip straight to interviewing you about the first idea they had. Reports in that thread vary sharply by model, and it is an open issue — the skill does not yet have a documented no-grill mode.
 **The report opened as unstyled raw HTML with no diagrams. What happened?**
 The report loads Tailwind and Mermaid from CDNs, so it needs network access when you open it, and it breaks silently when something blocks those scripts. The filed case was a security hook demanding SRI hashes: the agent added them, the CDN served different bytes to the browser than to the `curl` used to compute the hash, and the browser blocked the script. Offline and locked-down environments hit the same wall. The agent cannot see this, because it never renders the page. The workaround is to ask for inline CSS and hand-built SVG diagrams instead of the CDN scaffold. This is an open issue and a real rough edge.
 **It gave me twelve candidates. Do I work through them in the same session or start a new one?**
 One candidate per session. Working through several in one conversation fills the [context window](https://www.aihero.dev/ai-coding-dictionary/context-window) with the report, the grilling, the domain-model edits and the code changes all at once. The report only lives in a temp file, so carry the candidate itself rather than the file: pick one, grill it, take the decision into `/to-spec`, and turn the rest into [tickets](https://www.aihero.dev/ai-coding-dictionary/ticket) you can pick up independently later. Put the chosen improvement into a spec rather than going straight to implementation. This is a recurring question with no documented workflow in the skill itself.
 **How should I prompt it?**
 With the next thing you are building in mind. Where a big build is coming up, point it at the spec and ask "how can we make this change easy?" An unprompted run scans for hot spots on its own, which is fine for routine upkeep, but naming a direction is what makes the report actionable.
 **Does it work on a large legacy codebase?**
 Partly. It is strong on big existing codebases lacking consistent structure, and it is the recommended upkeep mechanism after any one-time structural setup. The honest counterweight: users with genuinely out-of-control projects report it "helped a little but still doesn't seem to cut it," and one developer with an eight-year legacy codebase reported the model going in circles where the same skill produces a clean graph on a tidy repo. There is no dedicated `/refactor` skill for that case yet. If the codebase has no shared vocabulary at all, [grill-with-docs](https://aihero.dev/skills-grill-with-docs) to establish one first tends to make this skill's output much better.
 **How is this different from `/codebase-design`?**
 `/codebase-design` is a reference, not a session driver. It supplies the vocabulary — module, interface, depth, seam, adapter, leverage, locality — and this skill borrows it. Pointing a fresh agent at `/codebase-design` as the thing to "do" is a known failure: with no process of its own to follow, the agent invents one, re-explores code and runs for a very long time before asking you anything. Drive with this skill; consume that one.
 **Will it ever tell me the codebase is fine?**
 Rarely, and you should know that going in. The skill is built to output findings, so the framing pushes it toward producing candidates rather than concluding that nothing is wrong. The strength badges are the defence — a report where everything is `Speculative` is the skill telling you it found nothing, in the only way it knows how.
 **Does it work in Codex or another harness?**
 Partially. The exploration step names Claude Code's `Agent` tool with `subagent_type=Explore` directly, so a [harness](https://www.aihero.dev/ai-coding-dictionary/harness) without that tool may skip the parallel exploration rather than substitute its own. The skill still runs; the scan is just less thorough. A harness-neutral rewrite has been proposed but is not merged.
 **How do I actually implement deep modules in TypeScript?**
 There is no good answer shipped with the skill. The recurring request is for a `TYPESCRIPT.md` giving concrete file and module layouts for the principles, and it does not exist. The skill will tell you where a deepening belongs and what should sit behind the seam; translating that into a package or directory structure is currently on you.
 ## It's working if
 - The candidates name your domain's concepts, not invented class names — "the Order intake module," not "the FooBarHandler."
 - The candidates cluster in files you have edited recently, not in dormant corners of the repo.
 - No code changed during the run. The only new file is the HTML report in your temp directory.
 - It stops after the report and asks which candidate you want, rather than continuing on its own.
 - Each card explains the payoff as locality or leverage, and says which tests get simpler — not just "this is cleaner."
 - Rejecting a candidate for a durable reason gets you an offer to record an ADR, so the next run does not re-suggest it.
 ## Where it fits
 `improve-codebase-architecture` is **periodic maintenance** — run it every few days, outside any chain, to queue up work rather than to do it. Its neighbours are [codebase-design](https://aihero.dev/skills-codebase-design), which owns the depth-and-seam vocabulary every candidate is written in, [grilling](https://aihero.dev/skills-grilling), which walks the decision tree once you have chosen a candidate, and [domain-modeling](https://aihero.dev/skills-domain-modeling), which keeps `CONTEXT.md` and the ADRs current as the decision settles. What it produces is an idea, which re-enters the main build flow at [grill-with-docs](https://aihero.dev/skills-grill-with-docs) or [to-spec](https://aihero.dev/skills-to-spec). For which skill fits a situation, [ask-matt](https://aihero.dev/skills-ask-matt) is the router over the whole set.
@@ -1,67 +0,0 @@
 ## What it does
 `prototype` writes **throwaway code that answers a question** — does this state model feel right, or what should this screen look like. The question comes first and decides the shape of everything that follows; a prototype that answers the wrong question is pure waste, however good it looks.
 Throwaway is a constraint on how the code is *written*, not a promise to destroy it. No tests, no error handling beyond what makes it run, no abstractions, no persistence — because none of that helps you learn the one thing you're trying to learn. What survives is the answer, folded into the real code, and the prototype itself, parked on a branch out of main as the evidence the answer came from.
 ## When to reach for it
 Type `/prototype`, or the [agent](https://www.aihero.dev/ai-coding-dictionary/agent) reaches for it automatically when a task fits.
 Reach for it the moment you hit a question you can't settle by talking — a state machine whose edge cases you can't hold in your head, a screen you can't picture until you see three versions side by side. [Grilling](https://www.aihero.dev/ai-coding-dictionary/grilling) sessions balloon on exactly these questions: the agent rephrases, you guess, and the scope grows to fill the uncertainty. Stop grilling, build the throwaway version, look at it, then answer in one line. If instead something already built is misbehaving and you want to know why, use [diagnosing-bugs](https://aihero.dev/skills-diagnosing-bugs) — prototyping explores what to build, not why the built thing is broken.
 You will also arrive here without choosing to. [wayfinder](https://aihero.dev/skills-wayfinder) files `prototype` decision [tickets](https://www.aihero.dev/ai-coding-dictionary/ticket) on its map, and working one is this skill.
 ## Two branches
 The question picks the branch, and the branches produce very different artifacts:
 - **"Does this logic / state model feel right?"** — a **single shareable HTML file**. One self-contained page, no build and no server, that someone opens by double-clicking. It carries a labelled state panel that re-renders after every click, free-play buttons for poking at the model in any order, and tabbed **guided walkthroughs** — one scenario per tab, each with the ordered buttons to press underneath it. Everything is labelled in domain language, so you can hand it to a designer, a PM or a domain expert and let them feel the model themselves. The logic behind the page is a small pure module — a reducer, a machine, a set of functions — kept clean of the DOM so the validated version lifts straight into the real code.
 - **"What should this look like?"** — several **radically different** UI variations on one route, switchable from a floating bottom bar and a `?variant=` URL param. Variants must disagree about structure, not colour; three tweaked card grids is wallpaper, not a prototype. They render inside a real page wherever possible, against real data and real density, because a variant judged in a vacuum always looks fine.
 Both keep state in memory, start with no thinking required, and show you the full state after every step. The moment you find yourself hardening one — adding a test, wiring the real database, generalising for a case you might want later — you have stopped prototyping.
 ## The prototype is a primary source
 A finished prototype leaves two things, and they go to different places.
 The **answer** — the verdict plus the question it settled — is captured durably: a commit message, an ADR, the implementation issue. That is what the main branch keeps, folded into the real code.
 The **prototype** is the runnable evidence the answer came from, and it is not deleted. It doesn't belong in main either — there is nothing there to maintain and it rots fast — so it is committed to a throwaway `prototype/<name>` branch out of main, never merged, with a [context pointer](https://www.aihero.dev/ai-coding-dictionary/context-pointer) to that branch left on the implementation issue. Main stays clean; the exploration stays findable and re-runnable by whoever picks the work up next.
 ## Common questions
 **Wait — isn't the prototype supposed to be deleted?**
 Not any more. It used to be: build it, keep the answer, bin the code. The sharpest objection to that was never about speed — it was *who picks up the work next [session](https://www.aihero.dev/ai-coding-dictionary/session), and what do they have to work from?* A prose summary of a prototype loses the thing that made it convincing. So the prototype is now treated as a [primary source](https://www.aihero.dev/ai-coding-dictionary/primary-source): it lands on a `prototype/<name>` branch out of main and the implementation issue points at it. What changed is where the code lives, not the discipline — it still never merges into main.
 **It used to build a terminal app. Where did that go?**
 The logic branch now emits a single shareable HTML file instead. A terminal app can only be driven by someone with the repo cloned and a runtime installed, which rules out exactly the people whose opinion the prototype needs — the designer, the PM, the domain expert who knows what the state model is supposed to mean. One self-contained file that opens by double-click and survives being emailed can be driven by anyone. The pure logic module underneath is unchanged, and is still the part that lifts into the real code.
 **An agent told me to `/prototype` when I should have been implementing.**
 Known, and it is a naming problem. `prototype` is a generic, appealing word that reads to a flow-unaware agent as "the obvious next step" once tickets exist, so it gets recommended by name even where the design was fully settled in conversation. If you already know what to build, the next step is `/implement`, per ticket. Reach for a prototype only when a specific design question is genuinely unresolved and talking won't resolve it.
 **Should I prototype the whole application before building any of its production features — say, to demo it to prospects?**
 That is a different artifact wearing this skill's name. A prototype here is scoped to one question, and "what is the whole app?" isn't one. A full-app prototype has no natural stopping point, so it becomes the production app by momentum: the cleanup pass never happens, and code written under prototype rules — no tests, no error handling — ends up in front of users. If you need a sales demo, build it deliberately as a demo and be explicit that none of it is production. If you need to settle a design question, cut it down to that question.
 **How do I run it in its own session?**
 A prototype lives in its own directory and generates a lot of [context](https://www.aihero.dev/ai-coding-dictionary/context) you don't want in the thread that asked the question, so run it somewhere else and bring back only the answer. [handoff](https://aihero.dev/skills-handoff) is the bridge in both directions.
 **Isn't this the fastest possible way to burn tokens?**
 It can be, if you prototype questions you could have answered by talking, or let one prototype sprawl across a whole feature. The comparison that matters isn't tokens against zero; it's [tokens](https://www.aihero.dev/ai-coding-dictionary/token) against building the wrong state model and finding out after it has production callers. Keep the question narrow and the run short, and the spend stays proportionate.
 ## It's working if
 - You can say in one sentence what question the prototype exists to answer — and it's written at the top of the demo, not just in your head.
 - Someone who doesn't read code can drive the logic demo. They open the file, press the buttons in a walkthrough tab, and describe what they see in their own words.
 - Someone says "wait, that shouldn't be possible" or "huh, I assumed X". That's a bug in the *idea*, which is the entire point.
 - The UI variants disagree about layout and information hierarchy, not just colour and copy — and the feedback you get is "the header from B with the sidebar from C".
 - It is answered in one sitting. If you're still building it a day later, the question was too big; split it.
 - When it's over, main contains the decision and none of the prototype, and the implementation issue points at the branch that still holds it.
 ## Where it fits
 `prototype` is a **reach-for-it-anytime standalone** — you drop into it to settle one design question, then drop back out — and it is also machinery another skill runs on.
 Its largest consumer is [wayfinder](https://aihero.dev/skills-wayfinder). A wayfinder map is made of **decision tickets**, and `prototype` is one of the four types a ticket can be: the one used when the blocking question is "how should this look" or "how should it behave", which no amount of discussion resolves. Wayfinder raises the fidelity of a foggy discussion by making something concrete to react to, and this skill is how that concrete thing gets built. A prototype ticket is resolved by the answer, and the prototype is linked from the map as an asset.
 The other neighbours are upstream and downstream of that. [grill-me](https://aihero.dev/skills-grill-me) and [grill-with-docs](https://aihero.dev/skills-grill-with-docs) answer grillable questions; the ungrillable ones come here instead, and the one-line answer goes back into the interview. Downstream, a validated state model or UI direction becomes settled input for [to-spec](https://aihero.dev/skills-to-spec), which can inline the decision-rich snippet the prototype produced rather than describing it in prose. For anything else, [ask-matt](https://aihero.dev/skills-ask-matt) routes you over the whole set.
@@ -1,73 +0,0 @@
 ## What it does
 `research` answers a question by reading the sources that own the answer, then leaves a cited Markdown file in the repo. It works only from **[primary sources](https://www.aihero.dev/ai-coding-dictionary/primary-source)** — official docs, source code, specs, first-party APIs — and follows every claim back to the source that owns it, so it will not repeat a blog post's account of an API when the API's own docs are reachable.
 It does not answer you in the conversation. The output is a file, written where the repo already keeps such notes, with a link on each claim. That is the point: a document you can react to, hand to another agent, or throw away, rather than an answer that vanishes when the [session](https://www.aihero.dev/ai-coding-dictionary/session) ends.
 ## When to reach for it
 Type `/research`, or the [agent](https://www.aihero.dev/ai-coding-dictionary/agent) reaches for it automatically when a task turns into reading legwork.
 Reach for it when the next step is *finding something out* from outside the working directory — how a third-party API behaves, what a spec actually says, whether a version claim holds — and you'd rather not stall your own thread doing the reading. What you need decides which skill:
 | What you need | Reach for |
 | --- | --- |
 | An external fact a decision is waiting on | `research` |
 | A decision made *with* you, by interview | [grilling](https://aihero.dev/skills-grilling) |
 | A durable architecture decision, written into `CONTEXT.md` and ADRs | [grill-with-docs](https://aihero.dev/skills-grill-with-docs) |
 | To find out whether an approach works in your codebase | [prototype](https://aihero.dev/skills-prototype) |
 | A plan too big to hold in one session | [wayfinder](https://aihero.dev/skills-wayfinder) |
 The line between `research` and `grill-with-docs` is the **shelf life of what comes back**. Research produces short-lived assets — what this library's auth mechanism does as of this week. An ADR records a decision you keep. If what you are producing is a decision rather than a fact, you are [grilling](https://www.aihero.dev/ai-coding-dictionary/grilling), not researching.
 ## Delegated legwork
 The defining move is that the reading runs as a **background agent**. You keep working; it goes off, follows each claim to its primary source, writes one Markdown file, and reports back. Research is legwork you delegate, not thinking you outsource — you get a document to grill, plan, or design against, and you still make the call.
 The delegation is unguarded, and the background agent can spawn a further background agent of its own. This is the skill's best-documented rough edge.
 Where the file lands is decided by the repo, not by the skill: it matches whatever convention already exists for notes, and if there is none it picks somewhere sensible and tells you where. It writes one file per run.
 ## Common questions
 **It spawned a second research agent — is that meant to happen?**
 No. This is an open bug, [issue #530](https://github.com/mattpocock/skills/issues/530). The skill tells its caller to spin up a background agent but does not restrict the agent type, so the agent it spawns is a `general-purpose` one that holds the `Agent` tool and the same instructions — and fires them again. One reporter measured a single research task costing roughly 450k [tokens](https://www.aihero.dev/ai-coding-dictionary/token) across three overlapping runs, with the duplicate finishing half an hour later entirely out of view. It reproduces outside Claude Code too; the same nesting was confirmed in Codex with GPT-5.6-sol. There is no shipped fix. Users have patched their own installed copy with a line telling an agent that is already a [subagent](https://www.aihero.dev/ai-coding-dictionary/subagent) to do the work itself, which helps but is instruction-level, not structural. Watch your background task list after invoking, and stop the duplicate.
 The opposite failure exists as well: if your own global instructions forbid an agent from re-delegating work, the background agent will politely decline the task and the skill quietly does nothing.
 **Where should the file live — and should I commit it?**
 The skill puts the file where the repo already keeps notes and does not have an opinion beyond that. The community one is fairly settled: ADRs are kept, research files are not. The sharpest version of it, from a Discord thread on exactly this question: "ADRs yes. Everything else archive or delete after done. It otherwise becomes cruft of work and can poison future repo reads if you've drifted away from the spec/research." A research file records what was true on the day it was written, so a stale one is worse than none. On balance these artifacts don't really belong in git, and there is no canonical home for them — people use Obsidian, a separate knowledge repo, or the issue tracker instead.
 **What counts as a "high-trust" primary source, and who decides?**
 The [model](https://www.aihero.dev/ai-coding-dictionary/model) does. The skill names the *kinds* of source that qualify — official docs, source code, specs, first-party APIs — and there is no allowlist, no domain gate, and no verification pass. This was the loudest objection when the skill was first proposed and it has never been answered publicly: "Five research subagents pointed at junk just gives you five confident wrong answers faster. How are you gating what counts as high-trust sources?" The mitigation you actually have is the citation on each claim. Follow two or three of them. If they land on a summary of the thing rather than the thing, the run failed at its one job.
 **Does a later session reuse what an earlier run found?**
 No. Nothing auto-loads a past research file; it is a document sitting in the repo until a human or a skill points at it. This was raised early as the strongest challenge to the design — "the value's the markdown becoming context the agent re-reads later, not the fetch itself. A write-once dead file is just a fancy search" — and the shipped skill does not solve it. In practice the file earns its keep by being fed into the next step deliberately: attach it to a spec, quote it into a grilling session, point a [ticket](https://www.aihero.dev/ai-coding-dictionary/ticket) at it.
 **Why not just ask the agent to go read the docs?**
 You can, and a two-line prompt saying exactly that was the practice this skill replaced. Two things the skill buys over the prompt: it runs in the background so your session keeps its [context](https://www.aihero.dev/ai-coding-dictionary/context) clean, and the primary-source constraint and the cited-file output come out the same way every time rather than however you happened to phrase it. Against a [harness](https://www.aihero.dev/ai-coding-dictionary/harness)'s own deep-research mode, the difference is the artifact and the source discipline, not the search. If a two-line prompt gets you what you need on a small question, use the two-line prompt.
 **When does it stop reading?**
 There is no stopping criterion in the skill, and this shows up as two complaints that look opposite but are the same gap: agents that go far too deep, and agents that cover a topic broadly while missing the one specific detail that mattered. One practitioner put it as "deep-research skills are a bit too deep sometimes. And telling an agent to research usually results in missing crucial details." Scoping is on you. A narrow, answerable question — one API, one behaviour, one version claim — comes back far better than "research X".
 **`/wayfinder` created research tickets — do I resolve those myself?**
 No, it now fires them for you. In the unreleased changes since v1.1, a charting session spawns a `/research` subagent per research ticket and burns them down in parallel, capturing findings on a throwaway `research/<name>` branch with a [context pointer](https://www.aihero.dev/ai-coding-dictionary/context-pointer) from the ticket. Research tickets are the one exception to wayfinder's one-ticket-per-session rule, because they are [AFK](https://www.aihero.dev/ai-coding-dictionary/afk) — nothing waits on you. Two known snags with those branches: the subagent has been seen opening a draft PR from a branch that is never meant to merge ([issue #576](https://github.com/mattpocock/skills/issues/576)), and deleting the branch later breaks the context pointers the tickets hold.
 ## It's working if
 - Your own session keeps going. If you are sitting watching it read, the delegation didn't happen.
 - Exactly one new background task appears. A second one with a near-identical name is the nesting bug.
 - One new Markdown file shows up, in the folder the repo already uses for notes, and the agent tells you the path.
 - Every claim in it carries a link, and following two at random lands you on an official doc, a spec, or the actual source file — not on someone's write-up of it.
 - You can make the decision you were stuck on from the file alone, without going back to the sources yourself.
 ## Where it fits
 A reach-for-it-anytime standalone that feeds the thinking skills rather than sitting in the build chain. Its file is something to take *into* the flow: [grilling](https://aihero.dev/skills-grilling) and [grill-with-docs](https://aihero.dev/skills-grill-with-docs) ask sharper questions when the facts are already on the table, and [to-spec](https://aihero.dev/skills-to-spec) can synthesise against it. [wayfinder](https://aihero.dev/skills-wayfinder) is the one skill that invokes it directly, resolving each research ticket on its map with a `/research` subagent. For the whole map, see [ask-matt](https://aihero.dev/skills-ask-matt).
@@ -1,51 +0,0 @@
 ## What it does
 `resolving-merge-conflicts` works through an in-progress git merge or rebase, hunk by hunk, then runs the project's own checks and finishes the operation with a commit.
 It refuses to treat a conflict as a text problem. Before touching a hunk it traces each side back to its **[primary source](https://www.aihero.dev/ai-coding-dictionary/primary-source)** — the commit message, the PR, the original issue — so it is choosing between two intents rather than between two blocks of text, and it preserves both wherever they are compatible. Where they genuinely are not, it picks the side matching the merge's stated goal and names the trade-off. It invents no new behaviour to paper over a clash, and `--abort` is not an option it has: the merge is always carried to a finished commit.
 ## When to reach for it
 Type `/resolving-merge-conflicts`, or the [agent](https://www.aihero.dev/ai-coding-dictionary/agent) reaches for it automatically when a task fits.
 Reach for it when git has already stopped on conflicts it could not resolve itself. It is scoped to the conflict in front of you, not to anything either side of it:
 | Your situation | Skill |
 | --- | --- |
 | Mid-merge or mid-rebase, conflict markers in the tree | This one |
 | Merge finished, something now misbehaves for reasons you can't see | [diagnosing-bugs](https://aihero.dev/skills-diagnosing-bugs) |
 | Planning how to slice work so branches collide less | Neither — see the parallel-work question below |
 ## Primary sources over `ours` and `theirs`
 The failure mode this exists to kill is resolving by flag: `--ours`, `--theirs`, or hand-deleting whichever block looks less important, so the markers go away and the build compiles. That resolution can be syntactically perfect and still silently drop a change somebody made on purpose.
 You cannot preserve an intent you have not read. So the work starts in the history — commits, PRs, [tickets](https://www.aihero.dev/ai-coding-dictionary/ticket) — and only then moves to the diff. Another step in the loop exists for the same reason: the skill finds the repo's own [automated checks](https://www.aihero.dev/ai-coding-dictionary/automated-check) and runs them before committing, because a merge is the easiest place in git to produce code that satisfies both branches and passes neither's tests.
 ## Common questions
 **Claude Code already resolves conflicts pretty well on its own. Why does this need a skill?**
 The added value is the "find the primary sources" and "run feedback loops" steps, which otherwise have to be prompted by hand every time. An unprompted agent will usually produce a plausible resolution from the diff alone and stop there. The skill's value is the two steps it will not let the agent skip — reading why each side exists, and running the checks afterwards. That is a thin margin over a good [model](https://www.aihero.dev/ai-coding-dictionary/model), and it is meant to be: at least one reader has predicted this is a whole skill that becomes a no-op as models improve.
 **Should I keep parallel agents off the same files to avoid conflicts in the first place?**
 Mostly no. Zoning files off between parallel tasks costs more than it saves, because agents are good enough at merge conflicts that the tradeoff is not as harsh as it looks. The one piece of discipline worth keeping is to do large refactors first. A large rename landing after ten branches have forked off it is the case that stays expensive.
 One caveat from a user report on parallel worktrees: when sibling [sessions](https://www.aihero.dev/ai-coding-dictionary/session) each build a ticket in their own tree, the merge back is best done by the session that wrote the change, because it is the one that already knows the intent. Batching everybody's conflicts onto one agent at the end throws away exactly the [context](https://www.aihero.dev/ai-coding-dictionary/context) step 2 of this skill has to go and reconstruct.
 **Why never `--abort`?**
 Aborting throws away the resolution work and returns you to the same conflict, unchanged, the next time you try. The skill is written for the case where the merge is going to happen. If you have decided it should not happen, that is a decision to make before invoking, not a branch inside the loop.
 ## It's working if
 - The agent quotes commit messages, PRs or issues at you while resolving, not just diff hunks.
 - Every hunk ends up with both sides' behaviour, or with an explicit note naming what was dropped and why.
 - Nothing appears in the result that was on neither branch.
 - Typecheck, tests and format were located and run green *before* the commit, not after you noticed something broken.
 - You end on a clean tree with the operation completed — including every remaining commit in a multi-commit rebase.
 ## Where it fits
 A reach-for-it-anytime standalone with no dependencies on any other skill: it starts when git stalls and ends when the tree is clean and committed. Its only real neighbour is [diagnosing-bugs](https://aihero.dev/skills-diagnosing-bugs), which takes over at the point where a merge resolved cleanly but the merged code misbehaves — a diagnosis problem, not a conflict one. It sits off the main idea-to-ship flow entirely, so [ask-matt](https://aihero.dev/skills-ask-matt) is the map for what runs before and after it.
@@ -1,94 +0,0 @@
 ## What it does
 `setup-matt-pocock-skills` answers three questions about one repo — where issues live, what the triage labels are called, and where the domain docs sit — and records the answers as markdown files under `docs/agents/`.
 Those files are the only thing that varies between repos. The skills themselves are identical everywhere; they read `docs/agents/issue-tracker.md` at run time and do what it says. That is why the set is not tied to GitHub, and why no skill file ever needs editing to point it somewhere else. Invoking it with "link the skills to a custom issue tracker" works with anything you can connect to programmatically, with zero changes to the skills.
 It is a prompt-driven skill, not a deterministic script. It reads your `git remote`, your existing `CLAUDE.md`, your existing `CONTEXT.md`, proposes what it found, and waits for you to confirm before writing anything.
 ## When to reach for it
 You invoke this by typing `/setup-matt-pocock-skills` — the [agent](https://www.aihero.dev/ai-coding-dictionary/agent) won't reach for it on its own. It is deliberately marked non-invokable, so no other skill can fire it for you.
 Reach for it once per repo, before the first use of any other engineering skill. If [triage](https://aihero.dev/skills-triage), [to-spec](https://aihero.dev/skills-to-spec), [to-tickets](https://aihero.dev/skills-to-tickets) or [wayfinder](https://aihero.dev/skills-wayfinder) start guessing where your issues go, or apply labels your tracker doesn't have, they have not been set up here yet. A repo already halfway through a project is a fine place to run it; the skill reads what is already there and no earlier work is wasted.
 ## Prerequisites
 It writes into the repo you run it in:
 | It writes | Where |
 | --- | --- |
 | `issue-tracker.md` | `docs/agents/` |
 | `domain.md` | `docs/agents/` |
 | `triage-labels.md` | `docs/agents/`, only when the `triage` skill is installed |
 | An `## Agent skills` block | whichever of `CLAUDE.md` / `AGENTS.md` already exists |
 All of it is committed markdown. There is no user-level or global mode: the config lives in the repo, so every repo gets its own copy.
 ## The three decisions
 It leads each section with the recommended answer, and skips whatever exploration already settled. Most runs are two confirmations and done.
 | Decision | What it proposes | When it actually asks |
 | --- | --- | --- |
 | **Issue tracker** | the one matching your `git remote` | always — this is the one real choice |
 | **Triage labels** | keep the five canonical names (`needs-triage`, `needs-info`, `ready-for-agent`, `ready-for-human`, `wontfix`) | only if the `triage` skill is installed |
 | **Domain docs** | single-context: one `CONTEXT.md` plus `docs/adr/` at the root | only if it spots monorepo signals, and then it offers a multi-context `CONTEXT-MAP.md` |
 The tracker options:
 | Option | Where issues live | Needs |
 | --- | --- | --- |
 | **GitHub** | the repo's GitHub Issues | the `gh` CLI |
 | **GitLab** | the repo's GitLab Issues | the `glab` CLI |
 | **Local markdown** | files under `.scratch/<feature>/` in this repo | nothing — no remote at all |
 | **Other** | wherever you say | one paragraph from you describing the workflow |
 The first three ship as templates in the skill and work out of the box. Local markdown is a first-class option, not a fallback: a solo project with no remote is fully supported. One caveat is worth repeating: don't use local markdown if you're using GitHub. They are alternatives, not layers.
 "Other" is not a stub either. It is the reason Jira, Linear, Azure DevOps and Beads all work: you describe the workflow, the skill records your prose in `docs/agents/issue-tracker.md`, and the downstream skills follow the prose. The community has already done this — a Jira-over-[MCP](https://www.aihero.dev/ai-coding-dictionary/mcp) variant, a Gitea CLI shaped like `gh`, a hand-built local dashboard.
 ## Common questions
 **Do I have to use GitHub?**
 No. GitHub, GitLab and local markdown under `.scratch/` all ship as ready-made templates, and anything else works through the "other" path. This is the most-repeated question in the record, in roughly these words: *"hard locked to github"*, *"can I use GitLab / Jira"*, *"what about Azure DevOps"*. The answer every time is that the tracker is a setup answer, not a skill property.
 **Do I need to re-run it after updating the skills?**
 Asked directly after v1.1, Matt said yes. The skill's own closing message is softer — it tells you re-running is only needed to switch trackers or start over. Both are defensible and the reason for the gap is real: the seed templates change between versions, so a `docs/agents/issue-tracker.md` written by an older release can go stale against the skills now reading it. If a downstream skill starts doing something the docs describe differently, re-running is the cheap fix.
 **It wrote to `CLAUDE.md`, but I'm on Codex.**
 Known gap, still open. The file-selection rule is "edit `CLAUDE.md` if it exists, else `AGENTS.md`" — it checks which file exists, not which [harness](https://www.aihero.dev/ai-coding-dictionary/harness) is running. A repo with a `CLAUDE.md` left over from Claude Code will get its `## Agent skills` block somewhere Codex never reads. Two workarounds are in circulation: move the block to `AGENTS.md` by hand, or keep `AGENTS.md` canonical and make `CLAUDE.md` a one-line pointer at it. If neither file exists, the skill asks you which to create rather than picking, which has confused people who expected it to just decide.
 **It didn't create my triage labels.**
 It doesn't. `docs/agents/triage-labels.md` is a *mapping* — it tells `/triage` which strings in your tracker correspond to the five canonical roles. It does not run `gh label create`. On a fresh GitHub repo the labels genuinely do not exist yet, and this has been filed as a bug more than once. Two follow-ons:
 - If your tracker already uses the canonical names, the mapping is an identity table and there is nothing to configure. That is the intended common case, not a missing step.
 - [wayfinder](https://aihero.dev/skills-wayfinder)'s `wayfinder:map` and `wayfinder:<type>` labels are not created here either, and `gh issue create --label <missing>` fails outright rather than creating the label. Create them by hand before the first wayfinder run on a GitHub repo.
 **Can I configure the other skills' behaviour here — [grilling](https://www.aihero.dev/ai-coding-dictionary/grilling) cadence, question format, tone?**
 No. It configures three things: tracker, labels, doc layout. There have been direct requests to make it the home for per-user preferences, and the standing answer is that skills stay opinionated: *"Config is death."* Preferences belong in your `CLAUDE.md` as plain instructions, which every skill already reads.
 **Can I keep the config in `~/.claude` instead of committing it to every repo?**
 Not today. There is an open request for exactly this from someone running the skills across many repos, and no user-level mode exists. Every repo carries its own `docs/agents/`.
 **Isn't it strange to have a skill that configures the other skills?**
 One long-standing complaint says yes, in these words: *"having a skill to set up the other skill does not feel right to me — that means the LLM is configuring its own skills."* The trade is real and acknowledged: the alternative to a setup step is duplicating tracker instructions into every skill that touches issues. The output is inspectable, editable markdown, which is the mitigation — you can read every file it wrote and change it by hand, and day-to-day tweaks are exactly that, not another run.
 ## It's working if
 - `docs/agents/issue-tracker.md` and `docs/agents/domain.md` exist, plus `triage-labels.md` if `triage` is installed.
 - An `## Agent skills` section appears in the instruction file your harness actually reads, with a one-line summary pointing at each of those files.
 - The tracker it proposed matches the remote you really use, and the label strings match labels that really exist in your tracker.
 - Afterwards, `/to-tickets` publishes without asking you where issues live, and `/triage` applies labels rather than inventing them.
 - Nothing in the skill files themselves changed. If setup edited a `SKILL.md`, something went wrong.
 ## Where it fits
 `setup-matt-pocock-skills` is the **run-once setup** for the engineering flow, the precondition everything else assumes rather than a step in the chain. Its neighbours are its readers: [triage](https://aihero.dev/skills-triage), which applies the label vocabulary written here; [to-spec](https://aihero.dev/skills-to-spec) and [to-tickets](https://aihero.dev/skills-to-tickets), which publish into the tracker named here; and [wayfinder](https://aihero.dev/skills-wayfinder), which reads the "Wayfinding operations" section of the same tracker file to know how maps and child [tickets](https://www.aihero.dev/ai-coding-dictionary/ticket) are stored. The domain-doc layout it records is the one [domain-modeling](https://aihero.dev/skills-domain-modeling) fills in later — it creates `CONTEXT.md` and ADRs lazily, when a term or decision actually gets resolved, so an empty repo after setup is the expected state. For which skill to reach for next, [ask-matt](https://aihero.dev/skills-ask-matt) routes the whole set.
@@ -1,94 +0,0 @@
 ## What it does
 `tdd` builds a feature or fixes a bug test-first: one failing test, then just enough code to pass it, then the next behaviour. It carries the standards that make that loop produce tests worth keeping — what a good test is, where tests go, what mocks are for, and the three anti-patterns that quietly ruin a suite.
 It writes no test at a seam you have not agreed to first. Before any test exists, it names the public boundaries it intends to test at and stops for your confirmation, because testing effort is finite and this is where you spend it on the critical paths instead of on every edge case. The other thing to know is that `tdd` is a **reference**, not a driver. It holds the rules of the loop, and something else (you, or [implement](https://aihero.dev/skills-implement)) runs the [session](https://www.aihero.dev/ai-coding-dictionary/session) that applies them.
 ## When to reach for it
 Type `/tdd`, or the [agent](https://www.aihero.dev/ai-coding-dictionary/agent) reaches for it automatically when a task fits — building a feature or fixing a bug test-first, or when you say "red-green-refactor".
 Reach for it when there is a concrete behaviour to build, with an input and an observable output, and you want tests that survive a refactor.
 | Your situation | Where to go |
 | --- | --- |
 | A behaviour with defined inputs and outputs — business logic, a request/response contract, a transformation, validation | `tdd` |
 | The behaviour isn't pinned down yet | [to-spec](https://aihero.dev/skills-to-spec), which also agrees the test seams before any code is written |
 | The question is really the shape of the interface, not the tests | [codebase-design](https://aihero.dev/skills-codebase-design) |
 | You have a [spec](https://www.aihero.dev/ai-coding-dictionary/spec) or [tickets](https://www.aihero.dev/ai-coding-dictionary/ticket) and want the whole build run for you | [implement](https://aihero.dev/skills-implement), which drives `tdd` per ticket |
 | Config, wiring, glue, type annotations, straight CRUD delegation | Nothing here fits well — see the open gap below |
 That last row is a real hole, not a stylistic preference. The skill decides *where* the seams go; nothing in it decides *whether* a change is worth the loop at all. Run it on a change with no independent source of truth to assert against and you get a test that restates the implementation — the tautological anti-pattern the skill itself warns about, arrived at from the other direction. It is [issue #746](https://github.com/mattpocock/skills/issues/746) and it is open. Until it closes, that judgement is yours or your `CLAUDE.md`'s.
 ## Prerequisites
 [codebase-design](https://aihero.dev/skills-codebase-design) needs to be installed. `tdd` used to carry its own deep-module and interface-design notes; in v1.0 those were deleted in favour of the shared skill, and `tdd` now leans on it for interface-design vocabulary. Nothing else — the skill is [stateless](https://www.aihero.dev/ai-coding-dictionary/stateless) and writes no files of its own.
 ## The loop, and the seam it runs at
 Three words carry this skill.
 **Red-green.** Write the failing test, then only enough code to pass it. No anticipating the test after next. There is no refactor phase: it was dropped in June 2026 because agents essentially never performed it, and because review and implementation work better as separate sessions. Refactoring belongs to [code-review](https://aihero.dev/skills-code-review).
 **Vertical slice.** One seam, one test, one minimal implementation, then repeat — the first cycle being a **tracer bullet** that proves a single path end to end. The opposite is horizontal slicing: all the tests first, then all the code. Bulk tests verify *imagined* behaviour, they check the shape of things rather than what a user does, and they commit you to a test structure before you understand the implementation.
 **Pre-agreed seam.** A seam is the public boundary you observe behaviour at without reaching inside. The rule is absolute: no test at an unconfirmed seam. In the full chain the seams are agreed earlier, during [to-spec](https://aihero.dev/skills-to-spec) — "`/tdd` is told to only work at pre-agreed test seams, `/code-review` checks that only agreed-upon test seams were used." Invoked on its own, `tdd` asks you directly.
 The three anti-patterns it is written to prevent:
 | Anti-pattern | The tell |
 | --- | --- |
 | Implementation-coupled | The test breaks when you rename an internal function, though behaviour did not change. Mocked internal collaborators, asserted call counts, database queries used to verify instead of the interface. |
 | Tautological | The expected value is computed the way the code computes it, so the test passes by construction. Expected values have to come from somewhere else — a known-good literal, a worked example, the spec. |
 | Horizontal slicing | A batch of tests landed before any implementation. |
 Mocks are for system boundaries only — external APIs, time, randomness, sometimes the filesystem or the database. Not your own modules.
 ## Common questions
 **Why doesn't it refactor? The description says "red-green-refactor".**
 Because the refactor step was removed and the description was not. The removal was deliberate: agents essentially never did it, and keeping implementation and review in separate sessions works better. Whether the result still counts as TDD by the book matters less than whether the loop produces better code. The mismatch between the trigger phrase and the body is filed as [issue #589](https://github.com/mattpocock/skills/issues/589) and is still open, so "red-green-refactor" continues to work as a phrase that fires the skill. What you get is red → green, and refactoring in [code-review](https://aihero.dev/skills-code-review).
 **It asked me to choose a test seam and I had no idea which to pick.**
 This is the most-reported friction with the skill ([issue #607](https://github.com/mattpocock/skills/issues/607)). The prompt lists candidate seams by name only, with nothing about what each one catches or misses, so you are choosing between labels. There is no fix shipped yet. The practical workaround is to ask the agent for the trade-offs before answering — what does the component-level seam miss that the integration seam catches, and how much slower is it. It is also why the chain agrees seams up front in `to-spec`, where you have the whole feature in view rather than one prompt.
 **It wrote the implementation before the test, even though the skill says red first.**
 It happens. One user pushed the [model](https://www.aihero.dev/ai-coding-dictionary/model) on it and got an unusually honest answer: "I knew the skill said 'one test at a time, watch it fail for the right reason' — I read it. I just defaulted to my normal habit." The skill is written to live with this. No instruction makes an agent comply 100% of the time, and forcing the point harder restricts the agent's creativity for little gain — the loop is worth running even when it is not followed strictly, because the results are still better overall. If strict adherence matters for a particular slice, watch the run rather than trusting the skill to enforce it.
 **Should it write browser or end-to-end tests first?**
 Usually not, and the skill will not stop it. A user reported the agent writing a Playwright test first, then burning a long loop re-running it and concluding the *test* was broken for a feature that did not exist yet. Configure this in your `CLAUDE.md`. Browser tests are slow enough that the red-green feedback loop stops paying for itself; declare in your repo's `CLAUDE.md` that they are written after the behaviour works.
 **Does `/tdd` replace `/implement`, or the course's `/do-work`?**
 No. `/tdd` documents the methodology; `/implement` is a very simple work→feedback→commit loop and is the direct stand-in for `/do-work`. The course's single `/do-work` step is now split across `/implement`, `/tdd` and `/code-review`. If you are asking which one to run against a ticket, the answer is almost always `/implement`.
 **Where did the deep-modules and interface-design guidance go?**
 Into [codebase-design](https://aihero.dev/skills-codebase-design) in v1.0, generalised so several skills share one vocabulary. `refactoring.md` left at the same time; refactoring is now [code-review](https://aihero.dev/skills-code-review)'s job, and that skill carries the Fowler smell baseline.
 **Does it know about my other tickets?**
 No. Run against one ticket, it will happily propose work that belongs to a sibling ticket, because it has no view of the rest of the issue graph ([issue #129](https://github.com/mattpocock/skills/issues/129)). Matt's position is that this is not `tdd`'s job. Passing the spec alongside the ticket helps; right-sizing the tickets in the first place helps more.
 ## It's working if
 - It stops and names the seams it intends to test at, and waits, before any test file exists.
 - One test appears, goes red, gets just enough code to pass, and only then does the next test appear — not a batch of tests followed by a batch of code.
 - Test names read as capabilities ("user can checkout with valid cart"), not as internals ("checkout calls paymentService.process").
 - Expected values in assertions are literals you can trace to the spec, not values recomputed the way the code computes them.
 - Renaming an internal function breaks nothing in the suite.
 - Mocks appear only at external boundaries — the payment API, the clock — and never around your own modules.
 ## Where it fits
 `tdd` is the engine inside the build step of the main chain, rather than a step of its own:
 ```txt
 grill-with-docs → to-spec → to-tickets → implement → code-review
 ```
 [to-spec](https://aihero.dev/skills-to-spec) agrees the test seams up front, [implement](https://aihero.dev/skills-implement) drives `tdd` per ticket, and [code-review](https://aihero.dev/skills-code-review) checks afterwards that only the agreed seams were used — and owns the refactoring `tdd` no longer does. Its other neighbour is [codebase-design](https://aihero.dev/skills-codebase-design), the shared source of the seam and deep-module vocabulary `tdd` speaks. You can also reach for it on its own, whenever there is a concrete behaviour to build and no full spec in play. When you are unsure which skill fits your situation, [ask-matt](https://aihero.dev/skills-ask-matt) routes you.
@@ -1,81 +0,0 @@
 ## What it does
 `to-spec` turns the conversation you have just had into a **[spec](https://www.aihero.dev/ai-coding-dictionary/spec)**, and publishes it to your issue tracker as a single issue.
 It does not interview you. By the time you reach for it the deciding is already done, so it synthesises what is known — from the thread, from the codebase, from your `CONTEXT.md` and ADRs — rather than opening a fresh round of questions. The spec is a record of decisions already made, not a place where new ones get made.
 ## When to reach for it
 You invoke this by typing `/to-spec` — the [agent](https://www.aihero.dev/ai-coding-dictionary/agent) won't reach for it on its own.
 Reach for it when the build is too big for one agent [session](https://www.aihero.dev/ai-coding-dictionary/session) and has to survive being split across several. That is the whole trigger:
 | Where you are | What to run |
 | --- | --- |
 | You haven't decided anything yet | [grill-with-docs](https://aihero.dev/skills-grill-with-docs) first |
 | Decided, and the work fits one [context window](https://www.aihero.dev/ai-coding-dictionary/context-window) | [implement](https://aihero.dev/skills-implement) — skip the spec |
 | Decided, and the work spans several sessions | `/to-spec`, then [to-tickets](https://aihero.dev/skills-to-tickets) |
 | A [wayfinder](https://aihero.dev/skills-wayfinder) map has cleared | `/to-spec #<map_issue>` |
 ## Prerequisites
 `to-spec` publishes the spec as an issue, so [setup-matt-pocock-skills](https://aihero.dev/skills-setup-matt-pocock-skills) must have configured a tracker and the triage-label vocabulary for this repo first. Either kind works: a real tracker like GitHub, or local markdown files under `.scratch/`, which is supported out of the box.
 ## The spec is a decision record
 The spec exists because context windows end. Everything you settled while [grilling](https://www.aihero.dev/ai-coding-dictionary/grilling) — the shape of the solution, the choices you argued through, what you deliberately refused — is in one conversation that is about to be cleared. The spec is what survives that.
 So it does not validate anything, and it does not decide anything. It captures what was decided, in your project's own vocabulary, so that a fresh session can pick the work up without you re-explaining it. Anything the spec asserts that you never actually said is a defect.
 ## Seams before prose
 Before it writes a word, `to-spec` sketches the **seams** the feature will be tested at, and checks them with you. It prefers seams that already exist to new ones, and takes the highest seam it can — the ideal number across a change is one.
 Those agreed seams then travel. [tdd](https://aihero.dev/skills-tdd) works only at pre-agreed seams, and [code-review](https://aihero.dev/skills-code-review) reviews the diff against the spec, so a seam nobody agreed to shows up as a review finding. The binding is indirect — it runs through this document — which is exactly why the seam conversation is worth taking seriously here rather than deferring it to implementation.
 ## Common questions
 **Where did `/to-prd` go?**
 It is this skill, renamed in v1.1. "Spec" is now the single through-line term, and the old `to-prd` slug is dead — reinstall under the new name. The pair that replaced the old vocabulary is *spec* and *tickets*: the spec is the destination and the decisions that fix it, the [tickets](https://www.aihero.dev/ai-coding-dictionary/ticket) are the execution steps that get there. If you pivot, delete the unfinished tickets and keep the spec.
 **Why does the spec get the `ready-for-agent` label? I don't want an agent implementing off it.**
 The label means "no further triage needed" — the document is complete enough for an agent to work from. It is an input designation, not a work order. But if you run [AFK](https://www.aihero.dev/ai-coding-dictionary/afk) agents that poll for `ready-for-agent`, that distinction isn't visible to them, and they will happily try to build the whole spec in one run instead of picking up the ticket slices. This is the most-reported rough edge on the skill. Until it changes, exclude the parent spec explicitly in your AFK agent's prompt, or strip the label once `/to-tickets` has run.
 **Why not go straight from grilling to `/to-tickets` and skip the spec?**
 Often you should — the spec earns its step only on multi-session work. Where it pays is that the tickets are disposable and the spec isn't: each ticket is sized for one fresh context window and gets deleted or closed, while the spec stays as the one place the reasoning behind them lives. On a single-session change that buys you nothing, and you have paid an extra synthesis step where the [model](https://www.aihero.dev/ai-coding-dictionary/model) can drift. Go grilling → `/implement`.
 **I just finished a wayfinder map. What do I feed it?**
 The main map issue — `/to-spec #<map_issue>`, not the individual decision tickets. [wayfinder](https://aihero.dev/skills-wayfinder) produces decisions rather than deliverables, scattered across a map; `to-spec` is the step that collapses them into one buildable document. Looping the map straight into `/implement` throws that collapse away.
 **Is the spec for me to review, or is it just for the agent?**
 Mostly for the agent, and it reads that way — complete, dense, reference-heavy. The parts worth your eyes are the seams and the out-of-scope section, because those are the two places a wrong decision is cheapest to catch and most expensive to discover later. Reading the whole thing end to end is a real complaint people have, and there is no summary mode: the honest answer is that if the spec surprises you, the grilling was too shallow, not the spec too long.
 **Do I keep the spec frozen once tickets start, or let the agent rewrite it?**
 Nothing keeps it in sync, so in practice it is a snapshot of what you knew at that moment, and it goes stale the first time implementation teaches you something. Treat it as throwaway once the work ships. The artifacts meant to outlive it are your `CONTEXT.md` and your ADRs — if something learned during implementation deserves to last, it belongs there, not in an edited spec.
 **My work is a refactor or a module boundary, not a feature. Does the template fit?**
 Less well, and this is a known limitation. The template leans hard on user stories, which is the wrong shape for architectural work — you end up writing stories nobody asked for around decisions that are really about interfaces and invariants. Lean on the implementation-decisions and testing-decisions sections instead, and let the durable architectural calls land as ADRs via [grill-with-docs](https://aihero.dev/skills-grill-with-docs) rather than trying to make the spec carry them.
 **Will it check the tracker for related work, or cite the ADRs it's respecting?**
 No to both. It reads and respects the ADRs covering the area it touches, but it doesn't link them, and it doesn't search the tracker for overlapping issues before drafting — so a spec can quietly duplicate work someone already filed. Search the tracker yourself first if the area is busy.
 **`/to-tickets` couldn't read my spec — it kept truncating.**
 Very large specs can outgrow what a tracker issue will serve back cleanly, and there is no local copy to fall back on. The fix is context hygiene: don't [clear](https://www.aihero.dev/ai-coding-dictionary/clearing) or [compact](https://www.aihero.dev/ai-coding-dictionary/compaction) between `/to-spec` and `/to-tickets`. Run them in the same window and the spec never has to be re-fetched at all.
 ## It's working if
 - It starts writing rather than asking you a fresh round of questions.
 - It puts the seams to you before it writes, and proposes as few as it can get away with.
 - It comes back in your project's nouns, not generic product-management boilerplate.
 - Every decision in it is one you can remember making. Nothing was invented to fill a section.
 - The out-of-scope section has real things in it — the things you refused are usually the most useful lines on the page.
 ## Where it fits
 `to-spec` is a step in the main build chain, and only on the multi-session branch of it:
 ```txt
 grill-with-docs → to-spec → to-tickets → implement → code-review
 ```
 Its neighbours upstream are [grill-with-docs](https://aihero.dev/skills-grill-with-docs), which does the deciding this skill only records, and [wayfinder](https://aihero.dev/skills-wayfinder), whose finished map merges onto the chain right here. Downstream, [to-tickets](https://aihero.dev/skills-to-tickets) cuts the spec into tracer-bullet tickets for [implement](https://aihero.dev/skills-implement) to build. When you're unsure which skill or flow fits, [ask-matt](https://aihero.dev/skills-ask-matt) routes you.
@@ -1,99 +0,0 @@
 ## What it does
 `to-tickets` takes a plan, a [spec](https://www.aihero.dev/ai-coding-dictionary/spec), or the conversation you are in, and breaks it into a set of **[tickets](https://www.aihero.dev/ai-coding-dictionary/ticket)** on your issue tracker. Each ticket declares its **blocking edges** — the other tickets that have to finish before it can start.
 Every ticket is a **tracer bullet**: a narrow but complete path through every layer of the change — schema, API, UI, tests — that can be demoed on its own the moment it lands. That is the constraint that makes it behave differently from the obvious way to split work, which is to cut one layer at a time and integrate at the end. It also sizes each ticket to fit in a single fresh [context window](https://www.aihero.dev/ai-coding-dictionary/context-window), because the thing that will pick the ticket up is a [session](https://www.aihero.dev/ai-coding-dictionary/session) that has never seen your spec.
 ## When to reach for it
 You invoke this by typing `/to-tickets` — the [agent](https://www.aihero.dev/ai-coding-dictionary/agent) won't reach for it on its own.
 | Where you are | What to run |
 | --- | --- |
 | You have a spec issue and the build spans several sessions | `/to-tickets`, or `/to-tickets #<spec_issue>` |
 | The plan is only in the conversation, never written up | `/to-tickets` reads the thread directly — no spec needed |
 | The whole change fits in one context window | [implement](https://aihero.dev/skills-implement) — skip the tickets |
 | Nothing is decided yet | [grill-with-docs](https://aihero.dev/skills-grill-with-docs), then [to-spec](https://aihero.dev/skills-to-spec) |
 | A [wayfinder](https://aihero.dev/skills-wayfinder) map has cleared | [to-spec](https://aihero.dev/skills-to-spec) first, to collapse the map, then `/to-tickets` |
 Tickets that `to-tickets` produced are agent-ready by construction. Don't run [triage](https://aihero.dev/skills-triage) over them — triage is for work that arrived from someone else.
 ## Prerequisites
 `to-tickets` publishes into a tracker, so [setup-matt-pocock-skills](https://aihero.dev/skills-setup-matt-pocock-skills) must have configured one for this repo, along with the triage-label vocabulary. Either kind works: a real tracker like GitHub or Linear, or local markdown files under `.scratch/`, which is supported out of the box.
 ## Tracer bullets, not layers
 A **horizontal** slice ships one layer of the change. Nothing works until every layer has landed, and each ticket's acceptance criteria have to reach into work that another ticket owns. A **vertical** slice — the tracer bullet — ships one thin path through all the layers at once, so it is verifiable alone and owns everything it grades.
 This is the rule people break most often, and the consequences are well documented. One team ran a 26-ticket stack sliced by layer — corpus, producer, aggregator, selector — and got roughly twenty agent runs per closed ticket, about three quarters of them rework. Their own post-mortem traced every failure class back to the horizontal slicing rather than to the implementations.
 Two things happen before anything is published. `to-tickets` looks for prefactoring — "make the change easy, then make the easy change" — and orders that work first. Then it presents the breakdown as a numbered list and quizzes you on it: is the granularity right, are the blocking edges real, should anything merge or split. Nothing reaches the tracker until you approve, and that quiz is the place to push back.
 ## Blocking edges
 The edges are the point of the artifact. They read two ways depending on the tracker:
 | Tracker | Where the edges live | How you work them |
 | --- | --- | --- |
 | Local markdown | Text in one file per ticket under `.scratch/<feature>/issues/<NN>-<slug>.md`, numbered blockers-first | Top to bottom, by hand |
 | A real tracker (GitHub, Linear) | Native blocking links, or sub-issues where the tracker has them | Any ticket whose blockers are done is on the **frontier** and can be grabbed |
 The edges live in the ticket either way. The medium only decides whether anything can act on them in parallel. `to-tickets` produces the artifact; running it — one session at a time, or a fleet — is your job, not the skill's.
 ## The wide-refactor exception
 One shape breaks the tracer-bullet rule. A **wide refactor** is a single mechanical change — rename a column, retype a shared symbol — whose **blast radius** fans across the whole codebase, so one edit breaks thousands of call sites and no vertical slice can land green.
 `to-tickets` sequences that as **expand–contract** instead:
 - **Expand** — add the new form beside the old, so nothing breaks.
 - **Migrate** — move call sites over in batches sized by blast radius (per package, per directory), one ticket per batch, each blocked by the expand. CI stays green because the old form still exists.
 - **Contract** — delete the old form once no caller remains, in a ticket blocked by every migrate batch.
 Where even the batches can't stay green alone, they share an integration branch and all block a final integrate-and-verify ticket. Green is promised only there.
 ## Common questions
 **It produced twelve tickets for a three-line change.**
 Over-decomposition is the most reported friction on this skill, and it is consistent across practitioners: the [model](https://www.aihero.dev/ai-coding-dictionary/model) defaults to atomic units and loses the grouping that would make them meaningful. The quiz step exists for exactly this — ask it to merge, and it will. The deeper answer is that the tickets have a floor: if the whole change fits in one context window, you don't need this skill at all. Go straight to [implement](https://aihero.dev/skills-implement).
 **The tickets came out one per layer — all the schema in one, all the API in another.**
 This is the failure the vertical-slice rule is written against, and the skill still produces it sometimes. Catch it at the quiz step by asking one question per ticket: what can I demo when this is done? A ticket with no answer is a horizontal slice. Some people add a "demo path" line to each ticket for this reason, and report it nudges the model toward vertical decomposition.
 **On GitHub the tickets weren't created as sub-issues of the spec issue.**
 Known and unfixed. It has been reported across a dozen runs and several models, [most fully in issue #554](https://github.com/mattpocock/skills/issues/554), and it is worse on Codex than on Claude. `gh` has supported this natively since v2.94: `gh issue create --parent <n>`, and `gh issue edit <parent> --add-sub-issue <n>` after the fact. Until the tracker template prefers those, wiring the parent links yourself after a run is the reliable move.
 **"Blocked by" was written into the issue body instead of a real blocking link.**
 Same class of problem, [reported in issue #513](https://github.com/mattpocock/skills/issues/513), where the agent went as far as asserting GitHub has no native blocking relationship at all. It does — `gh issue create --blocked-by 12,15`. Because blockers are published first, their numbers are always available at creation time. The body text is meant to be the fallback for trackers with no native edge, not the default.
 **Where do the local tickets go? The v1.1 notes said a root-level `tickets.md`.**
 They did, and that was a bug — a single shared file also raced when parallel agents wrote to it. Local mode now writes one file per ticket under `.scratch/<feature-slug>/issues/<NN>-<slug>.md`, in dependency order, matching the layout the local tracker template already described. The `NN` prefix is a real ticket ID, so `/implement 03` works instead of retyping a long title.
 **It kept truncating when it tried to read my spec.**
 A very large spec can outgrow what a tracker issue serves back cleanly, and there is no local copy to fall back on — the agent then burns [tool calls](https://www.aihero.dev/ai-coding-dictionary/tool-call) re-fetching chunks and never reaches the end. Don't [clear](https://www.aihero.dev/ai-coding-dictionary/clearing) or [compact](https://www.aihero.dev/ai-coding-dictionary/compaction) between `/to-spec` and `/to-tickets`. Run them in the same context window and the spec never has to be fetched back at all.
 **The acceptance criteria graded nothing — some passed before any work was done.**
 The template asks for criteria and says nothing about whether they can fail, so this happens. Three shapes recur: a criterion already true at the base commit, a criterion that can only be satisfied by work another ticket owns, and one that restates the request rather than deriving from the artifact. Vertical slicing prevents most of it — a slice that delivers behaviour which didn't exist before is red at the base commit by construction — but the check is worth doing by hand. For each criterion, name the observation that would show it false, and confirm it fails at the commit the implementer starts from.
 **The tickets are published. How do I actually run them?**
 The skill stops at the artifact, and there is no auto-dispatch mode. Dispatch is manual: look at the board, count the tickets with no open blockers, and open that many agent sessions. One ticket per fresh context, cleared between them. Be aware that [implement](https://aihero.dev/skills-implement) does not reliably close or check off the ticket when it finishes, on GitHub or in local markdown, so the ticket's state is yours to update.
 ## It's working if
 - Every ticket has an answer to "what can I demo when this is done?" — and the answer is behaviour, not a layer.
 - The list comes back to you numbered, with a "Blocked by" line on each, before anything is published.
 - The ticket at the top has no blockers and can be started immediately.
 - Nothing in a ticket body is a file path or a line number, except a snippet a prototype produced.
 - Each ticket reads like something a fresh session could finish without you in the room.
 - Prefactoring, where it found any, is at the front of the order rather than mixed into feature tickets.
 ## Where it fits
 `to-tickets` is a step in the main build chain:
 ```txt
 grill-with-docs → to-spec → to-tickets → implement → code-review
 ```
 Upstream is [to-spec](https://aihero.dev/skills-to-spec), which hands it a settled spec to slice against — keep both in one unbroken context window. Downstream is [implement](https://aihero.dev/skills-implement), which builds one ticket per fresh session, driving [tdd](https://aihero.dev/skills-tdd) for the tests and closing with [code-review](https://aihero.dev/skills-code-review). When you're unsure which skill or flow fits, [ask-matt](https://aihero.dev/skills-ask-matt) routes you.
@@ -1,99 +0,0 @@
 ## What it does
 `triage` works through the issues on your project's tracker, moving each one through a small state machine of **triage roles** — a category role and a state role — and leaving behind either an agent-ready brief, a specific question for the reporter, or a closed issue with a recorded reason.
 It is only for issues **you didn't create**. Raw bug reports, incoming feature requests, an external pull request that arrived unannounced — work that landed in the tracker from outside, in whatever shape the reporter left it. [Tickets](https://www.aihero.dev/ai-coding-dictionary/ticket) that [to-tickets](https://aihero.dev/skills-to-tickets) produced are already agent-ready by construction, and running `triage` over them is wasted work at best. The rule is flat: `/triage` is only for incoming issues, not for issues you created yourself.
 The second thing that separates it from labelling by hand: it recommends and waits. It tells you its category and state call with reasoning, plus what it found in the codebase, and applies nothing until you direct it.
 ## When to reach for it
 You invoke this by typing `/triage` and then describing what you want in plain language — the [agent](https://www.aihero.dev/ai-coding-dictionary/agent) won't reach for it on its own. "Show me anything that needs my attention", "let's look at #42", "move #42 to ready-for-agent".
 | What you have | Where to go |
 | --- | --- |
 | A tracker full of raw reports from other people | `/triage` |
 | A rough idea of your own, nothing written down | [grill-with-docs](https://aihero.dev/skills-grill-with-docs) |
 | A settled conversation to turn into a [spec](https://www.aihero.dev/ai-coding-dictionary/spec) | [to-spec](https://aihero.dev/skills-to-spec) |
 | A spec to split into agent-ready tickets | [to-tickets](https://aihero.dev/skills-to-tickets) |
 | A confirmed bug that needs a root cause, not a label | [diagnosing-bugs](https://aihero.dev/skills-diagnosing-bugs) |
 ## Prerequisites
 `triage` reads and writes your issue tracker, so [setup-matt-pocock-skills](https://aihero.dev/skills-setup-matt-pocock-skills) has to have configured that tracker and its label vocabulary first. The role names below are **canonical**; the label strings in your tracker may differ, and the mapping is what setup provides. If your tracker already uses the canonical names exactly, there is nothing to map and nothing to set up.
 The tracker config also decides whether external pull requests count as a request surface, and who counts as external. That flag defaults to off and is no longer a setup question — flip it in `docs/agents/issue-tracker.md` if you want PRs in scope.
 ## The state machine
 Every triaged item ends up carrying exactly one category role and one state role. Two categories: `bug` (something is broken) and `enhancement` (new feature or improvement). Five states:
 | State | Means |
 | --- | --- |
 | `needs-triage` | You need to evaluate it. Where an unlabelled issue normally lands first. |
 | `needs-info` | Waiting on the reporter. Returns to `needs-triage` when they reply. |
 | `ready-for-agent` | Fully specified, with an agent brief attached. An [AFK](https://www.aihero.dev/ai-coding-dictionary/afk) agent can take it. |
 | `ready-for-human` | The same brief, plus why this can't be delegated — judgment, external access, manual testing. |
 | `wontfix` | Closed, with the reason recorded. |
 That is the whole vocabulary, and the "exactly one state role" invariant is what keeps the queries simple. It is also the most-requested area of the [skill](https://www.aihero.dev/ai-coding-dictionary/skill): users have asked for a sixth state for work that is specified but blocked on another issue, for `deferred` work gated on a future trigger, and for a terminal `implemented` state. None of those has shipped. See the questions below.
 `wontfix` splits three ways, and the difference matters because only one of them writes to the knowledge base:
 | Why you're closing it | What happens |
 | --- | --- |
 | Already implemented | A comment pointing at where it already lives. Nothing is written to `.out-of-scope/` — it's a built feature, not a rejected one, and filing it there would poison the dedup checks. |
 | Rejected bug | Polite explanation, then close. |
 | Rejected enhancement | A file in `.out-of-scope/`, linked from the closing comment, then close. |
 `.out-of-scope/` is one markdown file per rejected **concept**, not per issue, written as a short design document rather than a database row: what was rejected, why, and every issue that has asked for it. `triage` reads the whole directory before it evaluates anything, and matches by concept rather than keyword — "night theme" matches `dark-mode.md`. When it hits a match it surfaces the old decision and asks whether you still feel the same way, instead of re-litigating the request from scratch.
 ## Verify before you brief
 Before any [grilling](https://www.aihero.dev/ai-coding-dictionary/grilling), `triage` checks that the claim actually holds. For a bug, it reproduces it from the reporter's steps. For a PR, it checks the branch out and runs the relevant tests. Then it reports which of three things happened: confirmed, with the code path; failed to reproduce; or not enough detail to try, which is itself the strongest `needs-info` signal there is.
 It runs two more checks against the codebase in the same pass — **redundancy** (is this already implemented, searched by domain concept rather than by the reporter's wording?) and **prior rejection** (does `.out-of-scope/` already say no?). Both are cheap, and both produce a `wontfix` when they hit.
 All of it exists to make one artifact good: the **agent brief**, the structured comment posted when an issue moves to `ready-for-agent`. Once it's posted, the brief is the contract and the original report is only context. Briefs are written to be **durable** rather than precise, because an issue can sit in `ready-for-agent` for weeks while the code moves underneath it. So they name types, signatures and behavioural contracts, and never file paths or line numbers. A confirmed reproduction makes a far stronger brief than a guess does.
 ## A PR is an issue with attached code
 Where the tracker treats external pull requests as a request surface, they run through the same machine — same categories, same states, same transitions. The states just read against the diff: `ready-for-agent` means a brief is attached and an agent should take the next step on the code, `ready-for-human` means it's ready for a person to merge. A brief on a PR describes what's left to do to the existing diff, not how to build the thing from nothing.
 Discovery surfaces only *external* PRs, because a collaborator's in-flight branch is not triage work. That filter is discovery-only — name a PR explicitly and it gets triaged whoever wrote it. One rough edge: the GitHub template's external-PR listing command asks `gh pr list` for an `authorAssociation` field that `gh` does not expose, so the command as written fails outright ([#468](https://github.com/mattpocock/skills/issues/468)).
 ## Common questions
 **I ran `/to-spec` and `/to-tickets`, and now those tickets are sitting there untriaged. Do I run `/triage` over them?**
 No. They are already agent-ready — `to-tickets` applies the `ready-for-agent` label as it publishes, precisely so an AFK runner picks them up without another pass. The user who hit this had run the spec flow, seen `needs-triage` on the output, and found their AFK runner ignoring everything. `triage` is the on-ramp for work that arrives from outside; the spec flow is the lane for work you originate. They meet at `ready-for-agent`, not before.
 **Is `triage` still relevant now that there's a `to-spec` → `to-tickets` → `implement` flow?**
 Only if you have inbound work. `triage` predates that spine and does a different job: it is the lane for reports other people filed. If everything in your tracker came out of your own planning, you will rarely open it. If you maintain anything public, or your team files bugs at you, it is the front door. The main use is open-source repos taking issues from external contributors.
 **The agent tried to apply `ready-for-agent` and `gh` said the label doesn't exist.**
 Known open bug ([#616](https://github.com/mattpocock/skills/issues/616)). `setup-matt-pocock-skills` writes the label vocabulary into `docs/agents/triage-labels.md`, but does not create the labels in your tracker. Create the five state labels and two category labels yourself, once, with `gh label create` or the tracker's UI, and it stops. There is a community fix branch linked from the issue that hasn't been merged.
 **Five states aren't enough — what about blocked, or deferred, or implemented?**
 This is the most-filed gap on the skill, in three shapes. An issue that is fully specified but waiting on another issue to close ([#139](https://github.com/mattpocock/skills/issues/139)) — the reporter's complaint was that `ready-for-agent` is "technically true" there but misleading, so an agent picks it up and hits a wall. Trigger-gated future work that is intended but not actionable yet ([#297](https://github.com/mattpocock/skills/issues/297)). And a terminal state for "implemented, awaiting verification", without which an AFK runner can re-queue finished tickets. Matt has agreed the blocked case is real and is undecided on the name (`blocked` versus `paused`). None of it has shipped. The workaround people use is a repo-local extra label alongside the category, which keeps the canonical state slot occupied by something honest at the cost of the skill not knowing about it. One community derivative goes further, adding `needs-slicing`, `tracking` and effort labels — that works, but it is theirs, not the skill's.
 **How is this different from `/diagnosing-bugs`?**
 The verification step here is deliberately shallow — enough to answer "is this real, and roughly where does it live", not to find a root cause. When a bug won't reproduce from the reporter's steps in a few minutes, the honest move is `needs-info`, or [diagnosing-bugs](https://aihero.dev/skills-diagnosing-bugs) if you want to chase it now. Neither skill's text currently mentions the other; a user found that seam, and it is still open.
 **Can I point it at my whole backlog and let it run?**
 You can ask, but watch what it reads. The "show what needs attention" pass is a cheap listing meant for *selection* — you pick one, and then it gathers full [context](https://www.aihero.dev/ai-coding-dictionary/context) on the one you picked. Run it across twenty issues at once and an agent can quietly fall back to that cheap listing as its evidence base, which returns issue bodies but not comments. A user hit exactly this: three issues already carried a comment saying "already fixed, recommend closing", and all three got fresh agent briefs instead. If you want a bulk pass, say explicitly that comments must be read per issue.
 **Does it work with Linear, or anything other than GitHub Issues?**
 Yes — the tracker is config, not a hard-coded assumption, and people run it against Linear (via the `linear` CLI), GitLab, and plain markdown files under `.scratch/`. A common split is Linear for issues and planning, GitHub for code and PRs: skills that say "issue tracker" map to Linear, skills that say "PR" map to GitHub. On the local-markdown tracker there is an open template bug where the generated file can carry the acceptance criteria twice, once at the top level and once inside the agent brief ([#200](https://github.com/mattpocock/skills/issues/200)).
 ## It's working if
 - Every item it touches ends with exactly one category role and one state role — never zero, never two states in conflict.
 - It gives you a recommendation with reasoning and stops, rather than relabelling and moving on.
 - The bug got reproduced, or the PR got checked out and run, before anything reached `ready-for-agent`.
 - The briefs it writes name types and behaviours, and contain no file paths and no line numbers.
 - A request that was rejected six months ago comes back, and it says so and quotes the old reason instead of triaging it fresh.
 - Every comment it posts opens with `> *This was generated by AI during triage.*`
 ## Where it fits
 `triage` is an **on-ramp**, not a step in the main chain. The main flow runs from an idea you had — grill, spec, tickets, implement, review — and `triage` is the parallel lane for work that arrived instead. It merges at the same place: an issue labelled `ready-for-agent` with a brief on it, which [implement](https://aihero.dev/skills-implement) picks up exactly as it would a ticket from [to-tickets](https://aihero.dev/skills-to-tickets). When a request needs sharpening before it can be briefed, `triage` runs [grilling](https://aihero.dev/skills-grilling) and [domain-modeling](https://aihero.dev/skills-domain-modeling) together, a round of questions at a time, so decisions land in `CONTEXT.md` and the ADRs as they're made. When you're not sure which lane you are in, [ask-matt](https://aihero.dev/skills-ask-matt) routes you.
@@ -1,103 +0,0 @@
 ## What it does
 `wayfinder` takes an effort too big for one agent [session](https://www.aihero.dev/ai-coding-dictionary/session) — an idea whose **destination** you can name but whose route you cannot yet see — and charts it as a shared **map** of **decision tickets** on your issue tracker, then resolves them one at a time until the way is clear.
 It plans, it does not do. Every ticket holds a question whose resolution is a decision, not a slice of a build to execute, and the map is finished when nothing is left to decide before someone goes and builds the thing. That one rule is what separates a wayfinder ticket from an ordinary implementation [ticket](https://www.aihero.dev/ai-coding-dictionary/ticket), and it is the rule agents break most often. When the map clears, wayfinder hands off; it does not carry on into code.
 ## When to reach for it
 You invoke this by typing `/wayfinder` — the [agent](https://www.aihero.dev/ai-coding-dictionary/agent) won't reach for it on its own.
 It is the heaviest, densest flow in the set, so the trigger is narrow: the effort has to be genuinely larger than one agent session can hold, and the route to the destination has to be foggy. The split is a clean one: `/grill-with-docs` for single-session planning, `/wayfinder` for multi-session planning.
 | What you have in front of you | What to run |
 | --- | --- |
 | A well-scoped feature you can settle in one sitting | [grill-me](https://aihero.dev/skills-grill-me), or [grill-with-docs](https://aihero.dev/skills-grill-with-docs) when there is a codebase |
 | A greenfield project, or a build spanning many sessions, with the route still unclear | `/wayfinder` |
 | A thread where the deciding is already done | [to-spec](https://aihero.dev/skills-to-spec) — skip straight past the map |
 | A cleared wayfinder map | [to-spec](https://aihero.dev/skills-to-spec), then [to-tickets](https://aihero.dev/skills-to-tickets) and [implement](https://aihero.dev/skills-implement) |
 | An existing session that has already grown too big | say "hand off to `/wayfinder`" — [handoff](https://aihero.dev/skills-handoff) bridges into a map as well as out of one |
 Greenfield is not a requirement. Wayfinder is used routinely on legacy and half-built codebases, and it is arguably sharper there, because a lot of the fog is "what is already true here" rather than "what should we do".
 ## Prerequisites
 The map and its tickets live on the repo's issue tracker, so wayfinder needs the tracker wiring that [setup-matt-pocock-skills](https://aihero.dev/skills-setup-matt-pocock-skills) lays down. That step writes a "Wayfinding operations" section describing how the map, its child tickets, blocking edges, and frontier queries are expressed for GitHub, GitLab, or local markdown. Wayfinder resolves that doc through the pointer in your `CLAUDE.md` / `AGENTS.md` rather than a fixed path; with no tracker configured at all it falls back to local markdown files.
 The tracker is not decoration. Blocking is what renders the frontier visually in the tracker's own UI, and a tracker without native dependency links — a self-hosted Gitea, say — degrades wayfinder to inferring blockers from the map text, which works but needs closer supervision.
 ## The map, the fog, and the frontier
 The **map** is a single issue labelled `wayfinder:map`; its tickets are its child issues. It is an **index, not a store** — a decision lives in exactly one place, its ticket, and the map only gists it and links. A session loads the map at low resolution and zooms into individual tickets on demand, which is what lets a map keep growing without every session paying for its whole history.
 Four things live on it:
 - **Destination** — what reaching the end of this map looks like. Naming it is the first act of charting, before any ticket exists, because the destination fixes the scope every ticket is measured against.
 - **Decisions so far** — one line per closed ticket, each linking to where the detail actually lives.
 - **Not yet specified** — the **fog of war**. Decisions you can tell are coming but cannot yet phrase sharply. The test for fog versus ticket is whether you can state the question precisely *now*, not whether you can answer it. Resolving a ticket clears the fog ahead of it and graduates whatever is now specifiable into fresh tickets.
 - **Out of scope** — work ruled beyond the destination. Fog only ever gathers *toward* the destination, so out-of-scope work is closed and never graduates.
 The **frontier** is the open, unblocked, unclaimed tickets — the edge of the known. A session claims a ticket by assigning it to itself before doing any work, so the assignee *is* the claim and concurrent sessions skip it. Tickets are referred to by name throughout, never by a bare `#42`; a wall of issue numbers is illegible in narration.
 ## The four decision-ticket types
 Every ticket carries a `wayfinder:<type>` label, and is either **[HITL](https://www.aihero.dev/ai-coding-dictionary/human-in-the-loop)** — worked with a human who speaks for themselves — or **[AFK](https://www.aihero.dev/ai-coding-dictionary/afk)**, driven by the agent alone. A HITL ticket only resolves through the live exchange; an agent that answers its own [grilling](https://www.aihero.dev/ai-coding-dictionary/grilling) questions has broken it.
 | Type | Mode | Reach for it when | Resolved by |
 | --- | --- | --- | --- |
 | `grilling` | HITL | The default. The question can be settled by talking it through. | [grilling](https://aihero.dev/skills-grilling) plus [domain-modeling](https://aihero.dev/skills-domain-modeling), in a fresh session |
 | `prototype` | HITL | "How should this look" or "how should this behave" — a question talking cannot settle. | [prototype](https://aihero.dev/skills-prototype), with the built artifact linked from the ticket as an asset |
 | `research` | AFK | A fact outside the working directory is blocking a decision. | A [research](https://aihero.dev/skills-research) [subagent](https://www.aihero.dev/ai-coding-dictionary/subagent), fired at charting time and burned down in parallel on a `research/<name>` branch |
 | `task` | Either | Nothing to decide, but manual work blocks a decision — provisioning access, signing up for a service, moving data so its shape can be seen. | The agent alone where it can, otherwise a precise checklist for the human |
 `task` is the only type that *does* rather than decides, and it earns its place by unblocking a decision — never by delivering a piece of the destination. This is the type that goes wrong most often in practice: agents interpret it as an implementation step and start writing product code inside the map.
 Research is the only exception to *one ticket per session*.
 ## Common questions
 **How is this different from `/grill-with-docs`? Which should I start with?**
 Session count, not project size. `/grill-with-docs` is single-session planning; wayfinder is multi-session planning. If you can hold the whole thing in one conversation, grilling is the cheaper and better tool, and wayfinder is genuinely slower and denser for that case. The community shorthand that has settled on it: wayfinder only makes sense if the work doesn't fit into a single session. This is by a distance the most-asked wayfinder question, and it keeps being asked because the descriptions do not tell you where your own task sits on that line — you have to judge the session count yourself.
 **When it asks for the "destination", does it mean the end of this session or the end of everything?**
 The whole map — the destination of the entire map, not just the initial session. The question reads ambiguously because wayfinder is by definition a multi-session tool, so a session-scoped answer never makes sense. Typical destinations are a [spec](https://www.aihero.dev/ai-coding-dictionary/spec) to hand off, a decision to lock before planning starts, a proof of concept, or a change made in place like a data migration.
 **The map is cleared. Why do I still need `/to-spec` and `/to-tickets` — didn't wayfinder already write the spec and make the tickets?**
 No. Wayfinder's tickets are decision tickets, and by the time the map closes they are all closed too. What is left is a map full of linked decisions, which is not a build plan. [to-spec](https://aihero.dev/skills-to-spec) collapses those linked decisions into one spec — `/to-spec #<map_issue>` — and [to-tickets](https://aihero.dev/skills-to-tickets) slices that into tracer-bullet implementation tickets. Looping the map straight into [implement](https://aihero.dev/skills-implement) skips the collapse and throws the linked detail away. Go straight to implementation only when the effort turned out genuinely small. People do run the abbreviated pipeline and report it working; the two extra steps buy you an explicit spec artifact that a reviewer or a colleague can read, which matters more the less solo you are.
 **My agent started writing production code in the middle of a wayfinder session.**
 The most-reported failure with this skill, and there is a real hole behind it. Wayfinder's "plan, don't do" default can be overridden in the map's **Notes** — but the Notes are written by the agent, so the constraint and its exemption live in the same file the constrained party owns. One user watched an agent write "this map carries execution" into its own Notes and then read it back in later sessions as its own licence, building on a live server. There is no hard in-skill stop for "I meant the default." Until there is: read the Notes on any map you didn't chart yourself, keep implementation in its own sessions, and treat any `wayfinder:task` that looks like a slice of the build as mis-typed.
 **I charted 27 tickets, and by the time I got to the thirteenth, the rest no longer made sense.**
 A real and repeatedly-reported outcome, verbatim from a field report. Wayfinder's default instinct is to plan comprehensively, and a map whose later tickets rest on assumptions the earlier ones invalidate is exactly the waterfall trap the skill is accused of. Two things push back on it. Scope the map to a bounded destination rather than to the whole product — practitioners consistently report that maps scoped to one defined epic behave better than a sprawling "implement V1", and planning something very big is not the goal in the first place — shipping small increments is. And [prototype](https://www.aihero.dev/ai-coding-dictionary/prototyping) aggressively: the whole reason the route stays current is that uncertainty is flushed out by cheap concrete artifacts before implementation depends on it. Wayfinder is "prototypemaxxing", not "planmaxxing".
 **Can I work several tickets in parallel?**
 The frontier is built to show you what is takeable, and blocking edges are there so parallel work is safe on paper. In practice one-at-a-time is the safer default. Users working two grilling tickets at once get asked in one session a question they just answered in the other, because the sessions share no [context](https://www.aihero.dev/ai-coding-dictionary/context). There is also a known gap on prototype tickets: an agent has been reported building three UI variations, choosing one itself, and closing the ticket — the selection is yours to make, and the skill does not currently say so loudly enough. If you do run in parallel, review the dependency graph yourself first.
 **Do I have to use GitHub Issues?**
 No — any issue tracker works. GitHub is the best-supported path because its native sub-issues and blocking relationships are what make the frontier visible without opening the map; GitLab, Linear, Jira and local markdown all get used. Two honest caveats. A tracker with no native blocking means the dependency graph is inferred from text and needs manual correction. And local markdown puts the artifacts in your repo, which is not recommended: storing this material in the repo tends to lead to accidental persistence. Open-source maintainers hit the opposite problem — public trackers filling with agent-generated planning tickets — and tend to choose local markdown anyway.
 **The grilling is exhausting. Every question is three paragraphs long.**
 This is the sharpest live complaint about wayfinder and it is not resolved. The decomposition one user gave: the verbosity itself causes decision exhaustion, and the length strips out *why* a question is being asked, so you lose the chain from decision to decision as the map gets longer. The verbosity looks like a property of the current set of [models](https://www.aihero.dev/ai-coding-dictionary/model) rather than of the skill, and no fix has landed. Practitioner mitigations in circulation: run a lower [reasoning effort](https://www.aihero.dev/ai-coding-dictionary/effort), and put a plain-language instruction in your global `CLAUDE.md`. Expect to spend real thought here regardless — the amount of thinking wayfinder demands from you is not a defect, it is most of what it is for.
 **A decision I already closed turned out to be wrong. Do I edit the old ticket or make a new one?**
 There is no official guidance, and the agent's instinct is unhelpful: it tends to design around the bad decision rather than challenge it, so you have to steer manually. What does work is telling wayfinder plainly what changed — it updates the map, revises the affected tickets, and comments on already-closed ones. Scope changes mid-map are recoverable. A map you *designed* to change is a scoping smell.
 **Where did `decision-mapping` go?**
 It is this skill, renamed to `wayfinder` in v1.1 and invoked as `/wayfinder`. "Decision map" was jargon and was also inaccurate, since only one of the four ticket types is really a decision by itself. The reframe gave the skill one coherent vocabulary — destination, fog of war, frontier, the map — instead of an invented term layered on top. The unit kept the "decision" word, though: a **decision ticket** is what a wayfinder ticket is called, precisely to stop people reading it as an implementation ticket.
 ## It's working if
 - The destination is written down and agreed before a single ticket exists.
 - Every open ticket reads as a question. Any ticket that reads "build the X" is either mis-typed or belongs downstream of the map.
 - You can look at your tracker and see which tickets are takeable without opening the map — that is the frontier rendering itself through native blocking.
 - A session resolves one ticket, posts the answer as a resolution comment, closes it, and leaves one line on the map's *Decisions so far*. Then it stops.
 - **Not yet specified** shrinks over time. A patch of fog that graduates into a ticket disappears from that section rather than living in both places.
 - When the opening breadth-first grill turns up no fog at all, the skill stops and tells you the effort is small enough to skip the map.
 - The session that finishes the map hands you toward a spec, not a pull request.
 ## Where it fits
 `wayfinder` is a **situational on-ramp**, not the default front door. The grill-led idea → ship chain is still where most work starts; wayfinder is what you climb onto when the idea is too big to hold in one session, and it merges back onto that chain at [to-spec](https://aihero.dev/skills-to-spec), because a cleared map hands off rather than builds.
 Underneath, it is mostly other skills wearing wayfinder's scheduling: [grilling](https://aihero.dev/skills-grilling) and [domain-modeling](https://aihero.dev/skills-domain-modeling) resolve the default ticket type, [prototype](https://aihero.dev/skills-prototype) resolves the tickets that talking cannot, and [research](https://aihero.dev/skills-research) runs as a subagent so its reading never lands in your session. [handoff](https://aihero.dev/skills-handoff) is the bridge in and out — into a map from a conversation that outgrew itself, out of one when a side quest appears mid-session. For anything else, [ask-matt](https://aihero.dev/skills-ask-matt) routes over the whole set.
@@ -1,98 +0,0 @@
 ## What it does
 `wizard` generates an interactive bash script that walks a human, step by step, through a manual procedure — wiring up third-party services, running a one-off migration, moving a project from state A to state B. It opens each URL, says what to click and copy, captures what comes back, and writes it into `.env` files and GitHub Actions secrets.
 The [agent](https://www.aihero.dev/ai-coding-dictionary/agent) writes the script; it never runs it. You do, on your own machine. So a wizard is not a list of instructions you follow — it is a program that drives the procedure and holds the state, and your part is to click, paste, and press Enter.
 ## When to reach for it
 You can type `/wizard`, and the agent can also reach for it on its own. When it hits a step you have to take — a key it can't mint, a dashboard it can't click — it builds you a wizard instead of writing the instructions into the chat, where they scroll away.
 Reach for it when the next thing blocking you is a trip through a dashboard:
 | Situation | What the wizard does |
 | --- | --- |
 | A new dev needs six services configured before the app boots | Opens each dashboard in order, captures the keys, writes them to `.env` and CI |
 | A one-off migration needs switches flipped in a specific order | Sequences the irreversible steps behind confirmation gates |
 | A project has to move from state A to state B once | Walks the transition and reports what it could not do |
 | You are about to write those steps into a README | Writes an executable version instead, which can't rot as quietly |
 Don't reach for it to *decide* what to build; for that, [grill-with-docs](https://aihero.dev/skills-grill-with-docs) and [to-spec](https://aihero.dev/skills-to-spec) are the tools.
 ## Prerequisites
 None to generate one. The wizard it writes runs on bash, and uses `gh` when a stage sets a GitHub secret or variable. If `gh` is missing or unauthenticated, that stage becomes a warning and the closing summary tells you what to set by hand, instead of failing the run.
 ## Stages
 A **stage** is one focused task on one screen. The script clears the terminal between stages, so a stage that overflows the screen loses the part that scrolled away. You author stages in dependency order and set `TOTAL_STAGES`, which drives the progress display.
 Scoping happens before a line is written. The [skill](https://www.aihero.dev/ai-coding-dictionary/skill) reads the repo instead of asking cold: `.env*`, `docker-compose*`, framework config, and every `secrets.*` / `vars.*` reference in `.github/workflows/` — each of those is a value the wizard has to produce. It then shows you the ordered stage list to confirm, and only after that maps each stage to the exact path a human follows ("Dashboard → Developers → API keys → Reveal test key → copy"). Where it doesn't know the current UI, it asks you or checks the docs rather than inventing clicks.
 For each captured value, scoping settles where it lands:
 | Destination | When |
 | --- | --- |
 | `.env` only | Local dev needs it, CI doesn't |
 | GitHub secret | CI reads it, and it's sensitive |
 | GitHub variable | CI reads it, and it's public |
 | Both `.env` and a secret | Local dev and CI both need it |
 | Nowhere | The stage is a pure action — a switch flipped, a plan upgraded |
 ## The template already solves the UX
 The [template](https://github.com/mattpocock/skills/blob/main/skills/wizard/template.sh) ships the whole experience: progress with time remaining, confirmation gates, cross-platform URL opening including WSL, hidden entry for secrets, idempotent `.env` upserts, `gh secret` / `gh variable` writes, and a closing summary of everything it had to skip. Everything above the `STAGES` marker is a fixed library, identical in every wizard and never hand-edited. The consistency is the point. Your job is only to scope the procedure and author its stages.
 The agent that writes a wizard never runs it end to end, because it opens browsers and waits for human input. It verifies statically instead: `bash -n`, `shellcheck` where available, and a trace that every value lands where scoping said it would, with every `set_secret` name matching a real `secrets.*` reference in CI. Set your expectations accordingly — the first run is yours, and that run is the test.
 ## Ephemeral by default
 | What you have | What to do with the script |
 | --- | --- |
 | A one-off migration, a personal setup, a transition you'll never repeat | Save it to a scratch or `scripts/` path, run it, delete it |
 | A setup path the next person on the repo will also need | Commit it and link it from the README, so they run the script instead of re-asking an agent |
 ## Common questions
 **Do my API keys end up in the model's context?**
 No. The agent writes a script; it doesn't run it. You run the script yourself, and it captures the key with hidden terminal entry and writes it straight to `.env` or `gh secret`. The wizard is a CLI, and the model is not connected to it. One caveat: that holds for values the wizard captures at runtime. If you paste a key into the chat while scoping the procedure, it's in the [context](https://www.aihero.dev/ai-coding-dictionary/context) like any other pasted text.
 **Can I go back and fix a value I mistyped?**
 Not mid-run. There is no back button — the stages run forward, and a wrong answer on stage 3 means Ctrl-C and re-run. Re-running is cheap by design: any value already written to `.env` is offered back as a default, so you press Enter through the stages you got right and retype only the wrong one. This came up in the launch week and hasn't been closed since: "loved it! One thing though — is there a way to go back and correct what you've entered?"
 There's a related open bug. Arrow keys in an `ask` prompt insert `^[[D` / `^[[C` instead of moving the cursor, because the prompt uses `read -r` rather than Readline ([issue #741](https://github.com/mattpocock/skills/issues/741)). Backspace works; arrow keys don't. Delete back to the mistake rather than moving the cursor into it.
 **Does it know what I've already set up?**
 Partly, and less than the launch reactions assumed. It reads the repo before it asks — your `.env` files, `docker-compose`, framework config, the `secrets.*` references in CI — so it scopes to values that are genuinely missing rather than starting from zero the way a README does. What it doesn't do is check the third-party service. If a key exists in your `.env` the wizard offers it back and Enter keeps it; if you already created the Stripe account but never saved the key, the wizard still sends you to the dashboard for it.
 **Where does it sit in the workflow — after grilling and the spec?**
 Nowhere in particular. It's a standalone, not a chain step. The common guess is `/grill-with-docs → /to-spec → /wizard`, and that sequence is fine, but the trigger is a manual procedure showing up, which can happen at any point: before you start, mid-build, or long after ship. It also works as a discovery tool — scoping surfaces the hidden prerequisites of a task, like the three API keys you hadn't thought about, before you commit to the work.
 **Does it work outside Claude Code?**
 The artifact does, unconditionally: it's a plain bash script and it doesn't care what [harness](https://www.aihero.dev/ai-coding-dictionary/harness) generated it. The skill itself is model-invoked, so it's listed everywhere — type `/wizard` in Claude Code or `$wizard` in Codex, or just describe the setup you're stuck on. Being model-invoked also keeps it clear of [#693](https://github.com/mattpocock/skills/issues/693), where Claude's desktop and web surfaces drop *user-invoked* skills from the [model](https://www.aihero.dev/ai-coding-dictionary/model)'s listing and report them as not installed.
 **Didn't this used to be user-invoked?**
 It did. It's now model-invoked, so the agent reaches for it unprompted when it hits a step you have to take. Nothing you could do before stopped working — model-invocation *adds* the agent's reach, it never removes yours, so `/wizard` behaves exactly as it did. What changed is the failure mode it retires: the agent hitting a credentials wall mid-build and dumping six numbered steps into the chat for you to follow by hand.
 **It used to be a beta skill — where is it now?**
 Promoted, as of v1.2. It graduated out of the beta channel and now ships in the plugin, so it arrives with the rest of the promoted set rather than needing an individual install. Its behaviour didn't change on graduation.
 ## It's working if
 - You're shown an ordered list of stages, and the values each one produces, and asked to confirm — before any script exists.
 - Every URL is opened before the value from that page is asked for. You're never asked to paste something you haven't been sent to fetch.
 - Secrets are typed blind. Nothing sensitive echoes into your scrollback.
 - Each stage fits one screen. Nothing you still need has scrolled away.
 - Ctrl-C and re-run picks up where you left off, offering the values already saved as defaults.
 - The final screen lists what it wrote, and separately lists what it couldn't do and you have to finish by hand.
 ## Where it fits
 `wizard` is a reach-for-it-anytime standalone, sitting at the line where automation stops and a human has to click. Its nearest neighbour is [setup-matt-pocock-skills](https://aihero.dev/skills-setup-matt-pocock-skills), because both exist to get a repo into a working state — that one configures this skill set, while `wizard` generates a setup path for everything else. It also pairs with [implement](https://aihero.dev/skills-implement): when a build lands a feature that needs credentials or a manual cutover, a wizard is how the human half gets done. When you're unsure which skill fits the moment, [ask-matt](https://aihero.dev/skills-ask-matt) routes you.
@@ -1,76 +0,0 @@
 ## What it does
 `grill-me` takes a **loose idea** and interviews you until you can commit to it. You do not need a worked-out plan to start — producing one is what the [session](https://www.aihero.dev/ai-coding-dictionary/session) is for. It asks in **rounds**: each round is the whole **frontier** — every question whose prerequisites you have already settled — so you are never asked something that hinges on an answer it hasn't heard yet.
 It is **[stateless](https://www.aihero.dev/ai-coding-dictionary/stateless)**. It writes no files and leaves no workspace behind. The only thing it leaves is a sharper version of the idea, in your own head.
 ## When to reach for it
 You invoke this by typing `/grill-me` — the [agent](https://www.aihero.dev/ai-coding-dictionary/agent) won't reach for it on its own. Start it in a **fresh conversation**, not on top of a plan you already had an agent write.
 Reach for it as soon as you have an idea worth taking seriously — a feature, a product direction, a business call, a piece of writing — and long before you have worked out what it involves. Vagueness is not a reason to wait; it is the thing the session eats. If you can already specify the thing precisely, you don't need to grill it.
 Which of the three grilling skills you want depends on what is in front of you:
 - **Anything, anywhere** — `grill-me`. It needs no repo and writes no files, and the subject doesn't have to be code.
 - **A codebase to align against** — [grill-with-docs](https://aihero.dev/skills-grill-with-docs). The same interview, but [stateful](https://www.aihero.dev/ai-coding-dictionary/stateful): it reads your code and keeps what it learns in `CONTEXT.md` and ADRs.
 - **Too big for one session** — [wayfinder](https://aihero.dev/skills-wayfinder). It charts the effort as a map and runs grilling sessions inside it.
 Leave [plan mode](https://www.aihero.dev/ai-coding-dictionary/agent-mode) off. Plan mode primes the agent to rush toward producing a plan, which is the opposite of staying in inquiry.
 ## It's a conversation, not an interview
 The skill asks the questions, but **you** own the scope. That is the part people miss, and it separates a session that turns an idea into decisions from one that produces confident nonsense.
 The failure mode is **passivity** — answering "agreed, agreed, agreed" for forty questions and coming out with a plan the agent wrote and you nodded at. It feels productive because it was long. Nothing was actually decided, and the result carries a certainty it hasn't earned.
 Being active means steering. Push back on a question pitched beneath the fidelity you need. Say when the scope is drifting. Answer "I don't know" and mean it. This skill is built to aid an engineer, not to replace one: what comes out tracks the quality of your answers, not the number of questions asked.
 The opposite error is real but rarer — staying in the interview so long you never reach code.
 ## Grillable and ungrillable
 Some questions can be answered by talking. Others can't, and no amount of grilling will get you there.
 "One long form or three pages?" and "how should this interaction feel?" are **ungrillable** — they need something to react to. When you hit one, stop grilling. Build the throwaway version with [prototype](https://aihero.dev/skills-prototype), look at it, then come back and answer in one line.
 Talking your way through an ungrillable question is where sessions balloon. The agent keeps rephrasing, you keep guessing, and the scope grows to fill the uncertainty.
 ## It's working if
 - You disagree with something. A session with no pushback from you is a session you didn't need.
 - Questions arrive in a few rounds rather than one long drip, and later rounds clearly build on what you said earlier.
 - You end up somewhere you didn't expect, because a question surfaced a decision you had been making implicitly.
 - At the end you could defend each choice to someone who wasn't there.
 ## Common questions
 **How many questions should I expect, and how do I know when it ends?**
 Count rounds, not questions. Forty-six questions across four rounds is an ordinary session. It ends when the frontier is empty — every branch visited, nothing left silently assumed.
 **It asked me two hundred questions. What went wrong?**
 Usually the scope was too large. Ask the agent to break the work into smaller pieces first, then grill each one. Very long sessions also drift into the **[dumb zone](https://www.aihero.dev/ai-coding-dictionary/smart-zone)**, where the [context window](https://www.aihero.dev/ai-coding-dictionary/context-window) is full enough that the questions get worse.
 **Can I go back to one question at a time?**
 Yes. Add this to your global `CLAUDE.md`:
 ```
 When grilling, ask one question at a time.
 ```
 **What if I genuinely don't know the answer?**
 Say so. "I don't know" is a real answer, and a question you can't answer is usually a sign to prototype rather than to guess.
 **Do I start a fresh session before writing the spec?**
 No. The value of the session is the [context](https://www.aihero.dev/ai-coding-dictionary/context) you just built. Hand the same conversation straight to [to-spec](https://aihero.dev/skills-to-spec).
 **Does the model matter?**
 More than for most skills. Grilling leans on the [model](https://www.aihero.dev/ai-coding-dictionary/model)'s own sense of how systems break, so give it your best one. Implementation mostly follows context and tolerates a cheaper model.
 ## Where it fits
 `grill-me` is a **standalone you can run anywhere, on anything**. Being stateless is what makes it portable: no repo, no workspace, no setup, and no assumption that the idea is even about software. People point it at business decisions, at writing, at what to do next — anything that won't sit still in their head.
 That portability is the whole difference from [grill-with-docs](https://aihero.dev/skills-grill-with-docs), which runs the same interview but reads a codebase to align against and records what it learns as `CONTEXT.md` and ADRs. Both sit on the [grilling](https://aihero.dev/skills-grilling) primitive; `grill-me` is the user-invoked front door that carries nothing with it.
 If what you grilled does turn out to be software, you can hand the same conversation to [to-spec](https://aihero.dev/skills-to-spec) and carry on into the build flow — an option, not the point of the skill. When you're unsure which flow fits, [ask-matt](https://aihero.dev/skills-ask-matt) routes you.
@@ -1,87 +0,0 @@
 ## What it does
 `grilling` is the interview loop that stress-tests a plan, a decision, or an idea before anyone acts on it. It maps the subject as a **design tree** — every decision branches into the decisions that hang off it — and interviews you branch by branch until nothing is left silently assumed.
 It does not ask one question at a time, and it does not ask everything at once. Each **round** asks the whole **frontier**: every decision whose prerequisites are already settled, and nothing else. Two questions never share a round if one depends on the other — a question that hinges on an answer still open belongs to a later round. Your answers settle decisions, the frontier moves outward, and the next round asks what that unblocked. Thirteen questions typically land in about three rounds rather than thirteen.
 ## When to reach for it
 Type `/grilling`, or the [agent](https://www.aihero.dev/ai-coding-dictionary/agent) reaches for it on its own when a task fits. It is the only [skill](https://www.aihero.dev/ai-coding-dictionary/skill) in the grilling family that is model-invoked, which is why you rarely type it: usually a skill you *did* type is running it for you.
 Typing `/grilling` directly gets you the plain interview and nothing else. Where you want something more than that:
 | What you have | Reach for |
 | --- | --- |
 | You aren't working in a working directory | [grill-me](https://aihero.dev/skills-grill-me) — the same [session](https://www.aihero.dev/ai-coding-dictionary/session), under a name the agent will never fire by itself |
 | You are in a working directory | [grill-with-docs](https://aihero.dev/skills-grill-with-docs) — the same session, and it writes `CONTEXT.md` and ADRs as it goes |
 | An effort too big to hold in one session | [wayfinder](https://aihero.dev/skills-wayfinder) — it charts a map and runs grilling inside the decision tickets |
 | A question that talking cannot settle — how something should look or feel | [prototype](https://aihero.dev/skills-prototype) — build the throwaway version, then come back |
 | A skill of your own that needs an interview | Invoke `/grilling` from it, rather than writing another interview |
 ## The round, the frontier, and who decides
 Three ideas carry the whole skill.
 The **design tree** is the model of the subject: decisions with decisions hanging off them. The **frontier** is the set of decisions whose prerequisites are all settled — the only questions that can honestly be asked yet. A **round** is one frontier, asked in full and answered in full.
 Inside a round every question arrives in a fixed shape: numbered and titled behind a `❓`, then the body, then the agent's recommended answer alone on a `➡️` line. That is what makes a round answerable by number — "1 yes, 2 the second option, 3 no, here's why" — instead of by quoting questions back. The format has one known rough edge: the recommendation sometimes argues *against* the question as it was worded, so agreeing with the recommendation means answering "no" to the question. When that happens, answer the recommendation and say so.
 The other half of the design is the split between facts and decisions. Facts are the skill's own job: when a frontier question needs something the [environment](https://www.aihero.dev/ai-coding-dictionary/environment) can settle, it dispatches a [sub-agent](https://www.aihero.dev/ai-coding-dictionary/subagent) to go and find out rather than asking you. It does not block on that — only the questions downstream of a running exploration wait. Decisions are yours, and it must wait for them. An agent running `grilling` that answers its own decisions has broken the skill, not interpreted it liberally. The session ends when the frontier is empty, and it will not act on what you agreed until you confirm you have reached a shared understanding.
 The honest limit: the frontier is the agent's judgement, not a computed graph. It can put two questions in one round and only afterwards discover that one answer should have changed the other. There is no guard against that beyond telling it, which reopens the affected branch in the next round.
 ## What lives here and what lives in the wrappers
 This page covers the mechanism. The things people most often want are documented one level up.
 | Question | Where it is answered |
 | --- | --- |
 | The tree, the frontier, rounds, the question format, facts vs decisions | Here |
 | How long a session should run, what to do with a question you can't answer by talking, how to avoid nodding along | [grill-me](https://aihero.dev/skills-grill-me) |
 | What gets written to `CONTEXT.md`, what becomes an ADR | [grill-with-docs](https://aihero.dev/skills-grill-with-docs) |
 ## Common questions
 **Can I go back to one question at a time?**
 Yes, and a large part of the audience does. Add this to your global `CLAUDE.md`:
 ```
 When grilling, ask one question at a time.
 ```
 The round-based default is genuinely contested. Practitioners who read slowly, who work in a second language, or who use the sequential format as focus scaffolding all report the one-at-a-time rhythm is better for them, and the opt-out is supported rather than tolerated.
 **Where did `/batch-grill-me` go?**
 Into this skill. Round-based questioning shipped briefly as a separate skill, then moved into `grilling` itself, so everything built on the primitive — `grill-me`, `grill-with-docs`, `triage`, `wayfinder` — got it at once. There is no `batch-grill-me` to install, and no separate sequential skill either; the `CLAUDE.md` line above is the way back to one-at-a-time.
 **Asking a whole round at once must lose the questions my earlier answers would have raised. Doesn't it?**
 This is the most common objection to the round design, and the frontier is the answer to it: a round only ever contains questions that do not depend on each other, so no answer in a round can invalidate another question in that round. Answers still reshape everything downstream — the next round is recomputed, not pre-written. What you lose is smaller than "all questions at once" implies, and larger than nothing: see the frontier's limit above.
 **It ran out of questions and started building.**
 A confirmation gate exists precisely for this: the skill is not finished when the frontier empties, it is finished when you say the understanding is shared. Weaker and faster [models](https://www.aihero.dev/ai-coding-dictionary/model) still break it — this is reported most often on lower-effort or non-frontier models, which collapse "interview until shared understanding" into a couple of questions and an outline. If yours does it, the reliable fix is a line in your own `AGENTS.md` or `CLAUDE.md` telling the agent not to implement without permission.
 **It answered its own questions instead of asking me.**
 That is a bug in the run, not the intended behaviour, and it was the reason facts and decisions were separated in the skill's text. It shows up most when another skill runs `grilling` inside a resolve-this-ticket frame, where the surrounding task reads as licence to keep moving. The same constraint is why there is no async mode: people have asked for a variant that reads a GitHub issue and posts one consolidated decision memo, and that is a different skill, because a grilling session that nobody answers has produced the agent's opinion rather than yours.
 **Can I cap the number of questions?**
 No, and a cap is deliberately out of scope. Some plans need three questions and some need fifty; a fixed ceiling either truncates the hard case or feels arbitrary on the easy one. Steering in plain language is the intended control — tell it to wrap up, or stop and accept the plan where it stands. If a session is running very long, the cause is usually that the scope was too big; break the work up and grill the pieces.
 **I installed `grill-me` on its own and nothing happens.**
 `grill-me` is a one-line skill whose whole body is "run a `/grilling` session", so it needs this skill installed too. The same is true of `grill-with-docs`, which additionally needs [domain-modeling](https://aihero.dev/skills-domain-modeling). Installing the whole set avoids the problem; installing selectively means installing the primitives as well.
 **`grill-with-docs` ran, but it never loaded `grilling`.**
 A real and unfixed rough edge, reported across [harnesses](https://www.aihero.dev/ai-coding-dictionary/harness) and models: a skill that names another skill does not reliably cause that skill to load, and `grill-with-docs` names two. The tell is a session that asks everything at once with no recommendations attached — that is the model improvising an interview rather than running this one. Asking the agent directly whether it loaded `grilling` and `domain-modeling` usually recovers it.
 ## It's working if
 - A round arrives as a numbered list, each question with its recommendation on a separate `➡️` line, and you can answer the whole round by number.
 - Nothing in a round needs another question in the same round answered first.
 - Later rounds ask things the first round could not have asked.
 - It goes and looks facts up — reading files, dispatching a sub-agent — rather than asking you something it could have found out.
 - Research running in the background does not stall the round; only the questions that depend on it wait.
 - It stops at the end and asks you to confirm the understanding is shared, instead of starting work.
 - Question count stays high while round count stays low.
 ## Where it fits
 `grilling` is a **primitive**, not a step you schedule: the single source of truth for the interview technique, kept in one place so every skill that needs an interview reaches for it instead of inventing one. [grill-me](https://aihero.dev/skills-grill-me) and [grill-with-docs](https://aihero.dev/skills-grill-with-docs) are its two user-invoked front doors, and `grill-with-docs` is where the main build chain begins, ahead of [to-spec](https://aihero.dev/skills-to-spec). [wayfinder](https://aihero.dev/skills-wayfinder) runs it to resolve decision tickets, [triage](https://aihero.dev/skills-triage) to grill a vague report into a workable one, and [improve-codebase-architecture](https://aihero.dev/skills-improve-codebase-architecture) to walk the tree once you have picked a candidate to deepen. When you are unsure which entry point fits, [ask-matt](https://aihero.dev/skills-ask-matt) routes you.
@@ -1,76 +0,0 @@
 ## What it does
 `handoff` compacts the conversation you are in into a **handoff document** — one markdown file, written to your OS's temporary directory rather than into the workspace, that a fresh [agent](https://www.aihero.dev/ai-coding-dictionary/agent) can read to pick the work up.
 What it buys is **portability**, not compression. That makes the skill narrower than it sounds. You need a file only when the work has to *travel* — to a new [harness](https://www.aihero.dev/ai-coding-dictionary/harness), a new directory, a colleague, or a side task you want to fork off. If nothing is travelling, you do not need a handoff: staying in the [session](https://www.aihero.dev/ai-coding-dictionary/session), `/clear`, a [subagent](https://www.aihero.dev/ai-coding-dictionary/subagent) and `/compact` cover the ordinary end-of-phase case, and `/compact` covers it more often than this skill does.
 ## When to reach for it
 You invoke this by typing `/handoff` — the agent won't reach for it on its own. Pass a note about what the next session is for, and the document is written for it.
 Four situations are the whole trigger:
 | Situation | Why a file |
 | --- | --- |
 | Swapping harness — Claude → Codex | The new harness cannot see the old [context](https://www.aihero.dev/ai-coding-dictionary/context) |
 | Moving to a different directory or repo | A prototype directory is the common case |
 | Sending the work to a colleague | They need something they can read |
 | Forking a side task found mid-phase | You keep working; a second agent takes the fork |
 For anything else — same harness, same directory, you are done [grilling](https://www.aihero.dev/ai-coding-dictionary/grilling) and moving to implementation — `/compact` is the move. [ask-matt](https://aihero.dev/skills-ask-matt) carries the ordered tree over all five options at a phase boundary.
 ## Branching is the use people skip
 The skill's description reads like session resumption: write a summary, end here, resume there. Read that way it looks like a worse `/compact`, so it gets skimmed past. The fork case is the one worth knowing. You **stay in your session** and hand a copy of the accumulated context to a second agent working in parallel.
 That is what the detour through [prototype](https://aihero.dev/skills-prototype) uses. You are deep in a design conversation, you hit a question that only running code will settle, and you do not want to spend the thread you built on finding out. Hand off to a prototype session, get the answer, hand the answer back, and reference it from the original thread. Two crossings, one live conversation, nothing re-explained.
 Three of the five options at a phase boundary preserve different things: `/compact` preserves your intent, `/clear` preserves nothing, `/handoff` preserves the work's ability to move.
 ## What travels, and what doesn't
 The document carries the live thread — what's in flight, why, and what's next — plus a **suggested skills** section naming what the next agent should reach for. Secrets are redacted before it's written.
 What it deliberately does not carry is anything already written down. Specs, plans, ADRs, issues, commits and diffs are referenced by path or URL, never copied. That keeps the file small, and it keeps the settled detail in one place instead of two that drift.
 ## Common questions
 **Handoff or compact?**
 `/compact` unless something is travelling. Staying on the same task is a compact, not a handoff — same harness, same directory, and you need to stay in the loop is where the phase-boundary tree lands most days. `/handoff`'s advantage is not that it summarises better; it's that the result is a file you can carry somewhere `/compact` can't reach.
 **So what's the actual difference between compact, clear and handoff?**
 Three different things being preserved. `/compact` compresses this context and keeps you going in a fresh window — intent survives. `/clear` empties the window and starts from nothing — correct when everything behind you is disposable, and one-way if it isn't. `/handoff` writes a portable file — the work survives the move to somewhere else. Note that all three turn a **[primary source](https://www.aihero.dev/ai-coding-dictionary/primary-source)** (the conversation as it happened) into a **[secondary source](https://www.aihero.dev/ai-coding-dictionary/secondary-source)** (a summary of it). Continuing is the only move that doesn't, which is why it's the first one to rule out.
 **Where did my handoff file go?**
 The temp directory, which is the most-reported friction with the skill: the paths are long, they differ per OS, and on Windows agents sometimes take several attempts to find the right one. Ask for the path back and keep it before you move on. Temp is deliberate: a handoff is a transit document, not an artifact you maintain. It is not a durable one either — see the next question.
 **My handoff vanished between sessions.**
 Some environments clear temp between sessions — Codex is the reported case — and `/private/tmp` goes on reboot. If the next session isn't starting within the hour, or is starting under a different harness, copy the file somewhere durable yourself as soon as it's written. The same applies to anything the document *points at*: a dispatch that references other files in temp is a dispatch the next agent can't follow.
 **How do I actually hand it to the next agent?**
 Open the fresh session and point it at the path: read this file, then continue. Point at the file rather than pasting the summary into a shell command — a summary containing backticks or `$(...)` gets mangled when it's interpolated into `claude "<summary>"`, and the usual failure is silent truncation rather than an error, so the new agent starts with a quietly incomplete brief.
 **Is this the same as `/branch`, `--fork-session`, or the built-in `/handoff`?**
 Analogous, not identical, and `/branch` isn't a shipped skill here — `/handoff` is the canonical name. A fork inherits an exact copy of the context; this skill produces a *targeted* compression aimed at a stated next task, in a file. Where a fork will do — same machine, same harness, same directory — a fork is less work. The file wins the moment the destination is somewhere the fork can't go.
 **When does something belong in `CLAUDE.md` instead?**
 Ask whether it's true next month. `CLAUDE.md` is standing context about the project, loaded into every session whether it's relevant or not. A handoff is about one piece of work in flight and is dead once that work lands. Facts that keep getting re-explained are a `CLAUDE.md` problem; a half-finished task is a handoff.
 **It captures the what, not the why.**
 A fair and repeated criticism. Two things help. Pass the argument — tell it what the next session is for — so the reasoning that bears on *that* is kept rather than flattened. And watch for confident claims the session never actually verified: "X isn't built", "Y is done". The next agent treats the document as a contract and will not re-check it, so a belief written as a fact becomes a false premise for everything that follows. Read the document before you hand it over, and downgrade anything you only assumed.
 **Why is it a skill rather than a slash command?**
 Both work; they suit different situations. As a skill it ships and updates through the same install path as everything else here, which is what makes it shareable — the constraint that the agent won't fire it itself is set by its frontmatter rather than by the mechanism.
 ## It's working if
 - The document is a small fraction of the conversation, and the specs, issues and diffs appear in it as paths and URLs rather than as copied text.
 - You can read it cold, without the original session open, and know what to do next.
 - The fresh agent starts working instead of asking you to re-explain the setup.
 - In the fork case, your original session is still sitting there untouched when you come back to it.
 - The suggested-skills section names the skill you'd have reached for yourself.
 - Nothing in it is a key, a token, or a password.
 ## Where it fits
 `handoff` is a **reach-for-it-anytime standalone** that lives at the seam between sessions rather than inside a build chain — but a narrow one, and the honest map is that you'll use it less often than the other four options at a phase boundary. Its closest neighbour is [prototype](https://aihero.dev/skills-prototype), because a prototype lives in its own directory and the round trip out and back is exactly the crossing this skill is for. When you're at a boundary and unsure whether to continue, clear, hand off, delegate or compact, [ask-matt](https://aihero.dev/skills-ask-matt) carries the tree that orders those five — and routes you over the rest of the set.
@@ -1,97 +0,0 @@
 ## What it does
 `teach` turns the directory you run it in into a standing teaching workspace and teaches you one topic across many [sessions](https://www.aihero.dev/ai-coding-dictionary/session), in short self-contained HTML lessons.
 It does not teach from what the [model](https://www.aihero.dev/ai-coding-dictionary/model) already knows. [Parametric knowledge](https://www.aihero.dev/ai-coding-dictionary/parametric-knowledge) is treated as untrusted: before it teaches, it goes and finds high-trust resources, records them in `RESOURCES.md`, and cites them inside every lesson. The other structural fact is that it is [stateful](https://www.aihero.dev/ai-coding-dictionary/stateful) — the mission, the resources, the lessons and the record of what you have learned all live in the directory as files, so the next session picks up from those files rather than from whatever is left of the last conversation.
 ## When to reach for it
 You invoke this by typing `/teach` — the [agent](https://www.aihero.dev/ai-coding-dictionary/agent) won't reach for it on its own.
 Reach for it when the learning is the project: a language, a framework, a codebase you have just joined, yoga, shaders, a certification. It is not the tool for one explanation in passing.
 | What you want | What to reach for |
 | --- | --- |
 | To learn a topic over weeks, with sessions that accumulate | `teach` |
 | One idea explained inside the session you are already in | Just ask, in that session |
 | The agent's last message re-pitched because it didn't land | [wait-what](https://aihero.dev/skills-wait-what) |
 | To sharpen thinking you already have, rather than acquire new material | [grill-me](https://aihero.dev/skills-grill-me) |
 | A background agent to read [primary sources](https://www.aihero.dev/ai-coding-dictionary/primary-source) and leave you a cited document | [research](https://aihero.dev/skills-research) |
 | To learn something that came up mid-grilling, without derailing the [grilling](https://www.aihero.dev/ai-coding-dictionary/grilling) | [handoff](https://aihero.dev/skills-handoff) out to a teaching workspace, then `teach` there |
 ## Prerequisites
 `teach` builds a directory rather than producing a file, and the skill assumes one mission per workspace — so run it somewhere you are happy to give over to a single topic. Keep it out of the project you are working in: a separate repo is the recommended home, rather than a global `~/.learnings/` folder or the working project itself. A dedicated repo also makes the lessons committable, which is how teams have shared them.
 What accumulates in that directory:
 | Path | What it holds |
 | --- | --- |
 | `MISSION.md` | Why you are learning this. Everything else hangs off it; if it is missing, the first thing `teach` does is interview you until it isn't |
 | `RESOURCES.md` | The vetted sources it teaches from, split into Knowledge and Wisdom (communities) |
 | `lessons/*.html` | The numbered lessons — the primary unit of teaching |
 | `reference/*.html` | Compressed cheat-sheets, algorithms, glossaries: the documents you actually return to |
 | `learning-records/*.md` | ADR-style notes on what you have demonstrably learned, used to decide what to teach next |
 | `assets/*` | Reusable components — a shared stylesheet first — so the lessons look like one course |
 | `NOTES.md` | Your stated teaching preferences |
 Two honest notes on that list. A glossary suits most topics, but the skill ships a `GLOSSARY-FORMAT.md` that `SKILL.md` no longer links to, so you will only get one if you ask ([issue #559](https://github.com/mattpocock/skills/issues/559)). And the workspace is not always created where you expect — see the first question below before you build a long course on top of it.
 ## Storage strength, not fluency
 The word to think with is **storage strength**: long-term retention, as opposed to **fluency**, the in-the-moment recall that feels like mastery while you are reading and is gone a week later. `teach` builds the former through desirable difficulty — retrieval practice, spacing, interleaving. Knowledge comes first, where difficulty is the enemy because it eats the working memory you need in order to understand; then the skill is drilled through a tight feedback loop, where difficulty is the tool.
 Two things steer what you get taught. The **mission** — the concrete real-world reason you want this — grounds every lesson; without it the lessons drift abstract and nothing decides what comes next. From the mission and the learning records, `teach` picks the next lesson inside your **zone of proximal development**: challenging enough to take effort, not so far ahead that it stops being learnable.
 It is also why the skill pushes back rather than obliges. A question that needs **wisdom** — real-world judgement — gets an attempted answer and then a pointer to a community where you can test it. A quiz is a gate, not a formality: one user reported saying "thanks a lot" and being told the drill was still live.
 ## Lessons, references and components
 A **lesson** is one self-contained HTML file, short enough to finish in a sitting, tied to the mission, giving one tangible win. It cites its sources, recommends one primary source to go and read yourself, and links to sibling lessons and reference documents.
 The split worth knowing: lessons are rarely revisited, reference documents are. So the compressed essence of a lesson — the syntax table, the algorithm, the pose sequence, the glossary — belongs in `reference/`, not buried in the lesson that introduced it.
 Lessons are built from **components** in `assets/`: stylesheets, quiz widgets, simulators, diagram helpers. Reuse is the default. The agent reads `assets/` before authoring a lesson and builds from what is there, and anything new that a second lesson could use is written as a component rather than inlined. The shared stylesheet is the first component every workspace earns; it is what stops the output being a pile of one-offs.
 ## Common questions
 **Where does it put the files? Mine ended up in `~/.claude/skills`.**
 A real, open bug ([#377](https://github.com/mattpocock/skills/issues/377)). `SKILL.md` uses `./` for two different roots at once: `./MISSION-FORMAT.md` and its siblings really do sit next to `SKILL.md` in the installed skill, while `./lessons/`, `./reference/`, `./learning-records/` and `./assets/` are meant to be in your directory. An agent that resolves the first kind against the skill's install directory goes on to resolve the second kind there too, and writes your course into the skill folder. Check where the first lesson landed before you build on it, and name the directory explicitly when you start rather than relying on "the current directory" being understood.
 **Do I stay in one session, or start a new one per lesson?**
 All three approaches work — staying in the same session, re-invoking `/teach` in a new session, or opening a new session in the same folder. Each lesson is its own invocation. The folder is the continuity, not the conversation. Common practice is to open a fresh session in the workspace and say `/teach next lesson for <topic>`.
 **How do I know it isn't teaching me something it made up?**
 You don't, on the skill's word alone. You read the primary sources. `teach` is not reliable enough to trust unchecked, and no skill built on an LLM is. The grounding machinery — `RESOURCES.md`, citations in every lesson, one recommended primary source per lesson — exists to make verification cheap, not to remove the need for it. The failure is not hypothetical: one user learning a 2x2 Rubik's cube was given fabricated move sequences that don't solve it. The diagnostic checklist for a case like that is model, harness, effort — and what the source was. Risk is highest in procedural domains with precise notation, and lowest where the output is immediately verifiable, like code you can run.
 **The correct quiz answer is always the first option.**
 Confirmed by several people, on Sonnet, on Opus and on GLM, and still unfixed. `SKILL.md` now requires every answer to be the same number of words, which kills a different tell — the correct answer used to be the only fully-reasoned one — but says nothing about position. One contributor tested an instruction-level fix for position and reported the correct answer still landing in slot A 33 times out of 33 across nine lessons ([#335](https://github.com/mattpocock/skills/issues/335)), which points at a shuffling quiz component in `assets/` as the real fix rather than better wording. Until that ships, treat answer position as meaningless. Your `assets/` directory is yours to change, so asking for a component that shuffles at render time is a legitimate local fix.
 **It assumed I already knew things, and used terms it never defined.**
 The commonest substantive complaint. There is no assessment step: `teach` infers your level from the mission and the learning records, and in session one there are no learning records. One user running it inside a wayfinder pipeline put it plainly — "It never did grilling to establish my starting point so it made lots of assumptions of what I already knew." Another reported lessons leaning on undefined jargon, and a lesson tailored to their hardware that covered what the hardware could do while never saying what it couldn't. Two things help: state your prior knowledge and your gaps in the first message, and correct the level out loud when a lesson misses, because the correction becomes a learning record and steers the next one. An explicit knowledge-assessment step is a standing feature request ([#725](https://github.com/mattpocock/skills/issues/725)), not shipped behaviour.
 **Does it do spaced repetition, and does it know when to stop teaching?**
 No to the first, and not reliably to the second. Spacing and interleaving are principles the lessons are designed against, but nothing schedules a review, and there is no Anki or calendar integration — both are recurring requests. The related gap is exit criteria: as one user put it, `teach` "is good at making the next lesson, but not as good at knowing when to stop and switch to review or real practice." If you want review or drilling instead of new material, ask for it; the skill will not propose the switch on its own.
 **Is it only useful for code?**
 No, and the non-coding use is the larger part of the record: Korean, Japanese formal register, piano, guitar, board game design, OpenSCAD, film plots, Azure and CCNA certifications, university exams, and children of eight and ten getting printable books on escape rooms and fire salamanders. Nothing in the skill is programming-specific — mission, resources, zone of proximal development and drill work the same way in any domain. Within code, the strongest reported use is not learning a language from scratch but getting oriented in an unfamiliar codebase or a new team's stack.
 **Which model should I run it with?**
 There is no canonical answer, and the reported differences are large. Higher [reasoning effort](https://www.aihero.dev/ai-coding-dictionary/effort) has been reported to produce noticeably better lessons than the medium setting. One user ran the same skill through Copilot CLI with Codex and got a single 30-line HTML card where Claude Code produced a full lesson. It runs unmodified in Claude Cowork, subject to whether your organisation allows skills to be added there. If the lessons come out thin, change model, [harness](https://www.aihero.dev/ai-coding-dictionary/harness) or effort before rewriting your prompt.
 ## It's working if
 - The first thing it does in an empty directory is interview you about why you want this, rather than produce a lesson.
 - `RESOURCES.md` fills up before the lessons do, and each lesson names one primary source worth reading yourself.
 - Claims in a lesson carry links out. A lesson with no citations is the skill teaching from memory.
 - A lesson takes one sitting and leaves you able to do one thing you couldn't before.
 - Opening a fresh session in the folder and saying "next lesson" continues the course instead of restarting it.
 - `learning-records/` grows, and lessons stop re-teaching what you have already demonstrated.
 - The lessons look like one course — they link the stylesheet in `assets/` rather than each carrying its own.
 - A question that needs judgement gets you pointed at a forum, subreddit or class, not just an answer.
 ## Where it fits
 `teach` is a **reach-for-it-anytime standalone**. It is not a step in a build chain and shares no artifacts with the engineering flow; it owns its directory and lives there for as long as the topic lasts.
 Its one real neighbour is [handoff](https://aihero.dev/skills-handoff), through the composition Matt named as the answer to "what do I do if I'm being grilled about something I don't understand?": don't stop the grilling to learn — `/handoff` to a teaching workspace, learn it there with `/teach`, then go back and pick up where you left off. The nearby alternative is [research](https://aihero.dev/skills-research), for when what you want is a cited document rather than lessons and retention. When you are not sure which skill or flow fits, [ask-matt](https://aihero.dev/skills-ask-matt) routes you over the whole set.
@@ -1,78 +0,0 @@
 ## What it does
 `to-questionnaire` turns a decision you can't settle on your own into a **questionnaire** — a Markdown document you hand to the one person who holds what you're missing, for them to fill in async or for the two of you to work through in a meeting.
 It grills you about the **send**, never the subject. Interviewing you about the topic is pointless here: not knowing the topic is why you're writing to someone else. So it asks the two things you can always answer — who this is going to, and what you need back from them — and aims every question in the document at the **gap** between the two.
 ## When to reach for it
 You invoke this by typing `/to-questionnaire` — the [agent](https://www.aihero.dev/ai-coding-dictionary/agent) won't reach for it on its own.
 Reach for it when a decision is blocked on knowledge that lives in one other person's head: a client, a domain expert, an exec who owns the business rules, a colleague on a team you don't sit with. Which skill you want depends on where the answers actually are:
 | The answers are in… | Reach for |
 | --- | --- |
 | Your own head, unsharpened | [grill-me](https://aihero.dev/skills-grill-me) |
 | The codebase | [grill-with-docs](https://aihero.dev/skills-grill-with-docs) |
 | Someone else's head | `to-questionnaire` |
 | Nobody's head yet — the question needs something to react to | [prototype](https://aihero.dev/skills-prototype) |
 The common case is a [grilling](https://www.aihero.dev/ai-coding-dictionary/grilling) session that stalls: some of what surfaced isn't yours to answer. Run `/to-questionnaire` in that same conversation to take those questions offline, then bring the answers back and carry on.
 ## The send, not the subject
 The interview is two exchanges, and then it stops.
 - **Who is it going to?** Their role, their expertise, their relationship to you. This fixes the tone and how much context the document has to carry — an outside client needs orienting, a teammate does not.
 - **What do you need back?** The concrete decisions or facts you can't resolve alone. This becomes the checklist the finished document is measured against: every item you named gets a question aimed at it.
 Everything after that is drafting. The file lands at `to-questionnaire-<slug>.md` in the current directory. There is no setup, no workspace, and nothing to configure.
 ## The document
 It is framed as a **discovery questionnaire** — you lack the context, the recipient holds it — and that framing drives its shape:
 - A purpose line naming the decision riding on it, and a short context section for a recipient who was never in your head.
 - Questions ordered **most-important-first** and grouped under themed headings, because async means you may only get one pass.
 - One idea per question, never compound, with an answer stub beneath it and a *why this matters* line only where a question could be misread.
 - Explicit permission to answer "I don't know" — a flagged uncertainty is useful; a confident guess that reads like a fact is not.
 - A closing catch-all: anything we didn't ask that we should know?
 Two things it deliberately isn't. It isn't **branching** — the questions are a flat, grouped list, not a tree that skips section D if you answered A. And it isn't **multi-recipient** — one run produces one document for one person.
 ## Common questions
 **Does it read my grilling session and extract the questions from it?**
 Not as a step of its own. The skill has no ingest phase: it asks about the send, then drafts. What makes it work after a grilling session is that you run it in the **same conversation**, so the [session](https://www.aihero.dev/ai-coding-dictionary/session) is already in [context](https://www.aihero.dev/ai-coding-dictionary/context) and the drafting can draw on it. Start it in a fresh session and it knows nothing about the grilling — you'll be re-supplying the topic yourself when you answer "what do you need back?".
 **The missing answers don't all live with the same person. Can it split them by recipient?**
 No. Step one asks for *the* recipient, singular, and the tone and context of the whole document are pitched at them. If three people hold three parts of the answer, run it three times, once per person. Routing questions by discipline or role inside a single document is a request people have made; it isn't what shipped.
 **Are the questions dependent — does it skip sections based on earlier answers?**
 No. The dependent-question design was explored and did not ship. The output is a static document: themed groups, most-important-first, every question live. The objection against it is a fair one — a [model](https://www.aihero.dev/ai-coding-dictionary/model) planning more than two or three questions ahead of a real answer plans badly, and a branching document has to plan all of them ahead of every answer.
 **What if the recipient doesn't know either?**
 The document tells them to say so. "I don't know" and partial answers are asked for explicitly, and a flagged uncertainty is worth more than a guess, because a vague answer and a confidently wrong one look identical once they're back in your context.
 **Does it send it anywhere — Slack, an issue tracker, email?**
 No. It writes a Markdown file in the current directory and tells you the path. Delivery is yours: paste it into a [ticket](https://www.aihero.dev/ai-coding-dictionary/ticket), drop it in a Slack thread, attach it to an email, or open it on a shared screen and work through it live. People have wired up all four by hand.
 **Isn't this just `/grill-me` in batch mode?**
 No, and the distinction is worth holding. `grill-me` already asks in **rounds** — the whole frontier at once, then recomputed from your answers — so the "give me all the questions at once" need is met there. `to-questionnaire` is about a different axis: not how the questions are delivered, but whose head the answers are in. Answering them yourself faster is `grill-me`; getting them out of someone else is this.
 **Couldn't I just ask the agent for this without a skill?**
 Yes, and plenty of people did before it existed — `OPEN_QUESTIONS.md` files, spreadsheets sent to clients, a "needs more info" ticket per unanswered question. The skill buys you two things: the interview never drifts onto the subject, and the document comes out in a shape a non-technical recipient can actually fill in. If you already have a house format that works, the honest answer is that you don't need this.
 ## It's working if
 - It asks about the recipient and about what you need back, then stops asking. A question about the subject itself is the skill off the rails.
 - Every item you named as "what I need back" is traceable to a question in the file.
 - The questions read as aimed at what the *recipient* knows, not as your own open questions copied down verbatim.
 - You could hand the file to someone who wasn't in the conversation and they would know why they got it and by when to reply.
 - The answers that come back are usable input for a new grilling round, rather than a fresh set of questions.
 ## Where it fits
 `to-questionnaire` is a reach-for-it-anytime standalone. It sits at the boundary of your own knowledge, where the next move is another person rather than another skill — most often mid-flow, when planning has stalled on something that isn't yours to decide.
 Its neighbour is [grill-me](https://aihero.dev/skills-grill-me), and the two split on where the answers live: grilling mines you, a questionnaire mines someone else. What comes back is raw material — feed it into another grilling round, or into [grill-with-docs](https://aihero.dev/skills-grill-with-docs) or [to-spec](https://aihero.dev/skills-to-spec) if the work is heading for a build. When you're unsure which skill fits the moment, [ask-matt](https://aihero.dev/skills-ask-matt) routes you.
@@ -1,36 +0,0 @@
 ## What it does
 `wait-what` is what you type when a message didn't land. The [agent](https://www.aihero.dev/ai-coding-dictionary/agent) then re-pitches what it just said. It adds the context you were missing, writes in plain English, and uses the vocabulary from your project's `CONTEXT.md`.
 The skill is three lines long. That is the design, not an unfinished draft. Skills that fight verbosity fail by growing: a four-hundred-line concision skill still leaves the [model](https://www.aihero.dev/ai-coding-dictionary/model) verbose, because the model reads the volume, not the plea. This one carries a single precise leading word and nothing else.
 ## When to reach for it
 You invoke it by typing `/wait-what`. The agent will not reach for it on its own, and it shouldn't. Only you know when you stopped following.
 Use it the second you notice you're skimming. The agent has drifted into jargon it invented, stacked five acronyms, or explained a decision whose premise you never saw. It fixes the conversation you're already in. To stop the jargon arriving at all, use [grill-with-docs](https://aihero.dev/skills-grill-with-docs), which builds the shared language upfront.
 ## The name is the mechanism
 The leading word is **wait**. "Be concise" is an instruction about the agent's output, and the model obeys it by clipping words and losing you further. **Wait** is about *your* state. It says comprehension failed here. An agent that hears "be brief" writes telegrams. An agent that hears "wait, you lost me" backs up and explains.
 That difference is the whole skill. Every popular fix for verbosity names the *output*: `/tldr`, `/no-fluff`, `/talk-normal`. The model over-corrects into a caveman register that is shorter and no clearer. Naming the *listener* asks for both halves at once: fewer words **and** the context you were missing.
 The skill says re-pitch **that**, not "that last message". What lost you is usually bigger than one paragraph, so the agent decides how far back to go.
 ## It plugs into the language you already have
 The body reuses the leading words already in your global `CLAUDE.md` and your project's `CONTEXT.md`. ASD-STE100 Simplified Technical English sets the register. The ubiquitous language supplies the nouns. The skill, `CLAUDE.md` and `CONTEXT.md` reach for the same [tokens](https://www.aihero.dev/ai-coding-dictionary/token), so invoking it is not a new instruction. It is a reminder of one the agent already agreed to.
 If you have no `CONTEXT.md`, the skill still works. You lose only the domain-vocabulary half.
 ## It's working if
 - The re-pitch is **shorter and clearer**, not shorter and blunter.
 - It adds the premise you were missing, instead of only deleting words.
 - Project nouns replace invented ones. The terms in your `CONTEXT.md` come back.
 - You can use it twice in a row, and it does not degrade into terseness.
 ## Where it fits
 You can use `wait-what` at any point, in any conversation, inside any other skill. It repairs one message after the fact. The real cure is a shared language agreed upfront, and that is [grill-with-docs](https://aihero.dev/skills-grill-with-docs): a [grilling](https://www.aihero.dev/ai-coding-dictionary/grilling) session that runs [domain-modeling](https://aihero.dev/skills-domain-modeling) as it goes, so the words you both use land in your `CONTEXT.md`. If you're unsure which skill fits the moment, [ask-matt](https://aihero.dev/skills-ask-matt) routes you.
@@ -1,70 +0,0 @@
 ## What it does
 `writing-for-agents` is the reference you write agent-facing documents against — a skill, an `AGENTS.md` / `CLAUDE.md`, a [spec](https://www.aihero.dev/ai-coding-dictionary/spec), a runtime prompt, a README, any doc an [agent](https://www.aihero.dev/ai-coding-dictionary/agent) reads. The packaging differs; the writing does not: the same levers make each one predictable, so the agent takes the same *process* every run rather than producing the same output.
 Its default move is deletion, not explanation. Ask an agent to write instructions for another agent and it spends most of its words explaining what the [model](https://www.aihero.dev/ai-coding-dictionary/model) already knows — every one of those lines is a **no-op**, paying [context](https://www.aihero.dev/ai-coding-dictionary/context) and changing no behaviour. This reference is the lens that finds them, which is why it earns its keep at least as often on a document you already have as on a blank file.
 It was called `writing-great-skills` until v1.1. The rename tracks what it always was underneath: almost none of it is skill-specific. The skill-only mechanics — frontmatter, the model- versus user-invoked choice, router skills — are disclosed to a linked `SKILL-MECHANICS.md` you read only when the document in front of you is a skill.
 ## When to reach for it
 Type `/writing-for-agents`, or the agent reaches for it on its own when you're creating or editing a skill, or modifying `AGENTS.md` or `CLAUDE.md`.
 Reach for it by hand for everything else an agent reads: your docs, specs and [tickets](https://www.aihero.dev/ai-coding-dictionary/ticket), system and [AFK](https://www.aihero.dev/ai-coding-dictionary/afk) prompts. The test is one question — does an agent read this? — and it does not matter how the document gets in front of it, whether a pointer names it, a human pastes it, or it simply sits in the repo. For working out what a codebase actually contains in the first place, use [grill-with-docs](https://aihero.dev/skills-grill-with-docs) — this reference governs how a document reads, not what it knows.
 ## The two loads
 The idea the whole reference turns on is a pair of budgets every document and pointer spends:
 - **Context load** — the cost of always-loaded material on the agent's window: an `AGENTS.md` line, a skill description, anything sitting in context every [turn](https://www.aihero.dev/ai-coding-dictionary/turn) whether or not it fires.
 - **Cognitive load** — the cost on you: which documents exist, and when to reach for each. You are the index. Not a cost to minimise — it is the price of human agency.
 Once you think in these two loads, most authoring decisions — split or don't, inline or disclose, point or push — become the same trade made in different places.
 ## The levers
 - **[Context pointers](https://www.aihero.dev/ai-coding-dictionary/context-pointer)** — the reference held in context that names out-of-context material and encodes when to reach it. A skill description and an `AGENTS.md` line naming a doc are the same object; the pointer's *wording*, not its target, decides how reliably the agent reaches through it.
 - **Information hierarchy** — the ladder from in-file step, to in-file reference, to disclosed reference behind a pointer. **[Progressive disclosure](https://www.aihero.dev/ai-coding-dictionary/progressive-disclosure)** is the move down that ladder so the top stays legible.
 - **Completion criteria** — the clarity and demand of each step's done-condition, and the **legwork** that demand drives; the defence against **premature completion**.
 - **Leading words** — a compact concept already in the model's pretraining (*tight*, *red*, *tracer bullet*) that the agent thinks with while running the document. It anchors twice: execution in the body, invocation in the pointer.
 - **Pruning** — single source of truth, relevance, and the no-op test applied sentence by sentence, against **duplication**, **sediment** and **sprawl**.
 ## Common questions
 **Where did `/writing-great-skills` go?**
 It is this skill, renamed in v1.1. Practitioners were already pointing it at `AGENTS.md`, docs, specs, tickets and runtime prompts long before the name caught up; structure, leading words and pruning turn out to be the craft of any text an agent reads. There is no alias — reinstall under the new name.
 **"Writing for agents" — so the agent does the writing?**
 The other way round. You are the author; the agent is the reader. That is the whole difficulty of the genre: you are writing for a reader who has already read everything, so explanation is waste and precision is the entire job.
 **Can't I just ask the agent to write it for me?**
 You can, and it will produce something verbose. Left alone the model explains what it already knows, and it will not apply the no-op test or reach for a leading word on its own. Use the reference on the draft — a review pass is where most of its value lands.
 **I asked an agent to trim a document and it cut the functionality.**
 Agents told to "streamline" optimise for length, because length is the thing they can see. The no-op test is behavioural, not aesthetic: delete the line and ask whether the agent's behaviour changed. When a sentence fails, delete the whole sentence rather than trim words from it — and settle a disagreement about it by running the document, not by arguing.
 **How do I know when it's done?**
 When it works, and you can no longer find duplication, sediment or no-ops. There is no automated eval here; the check is a manual run plus the failure-mode vocabulary as a diagnostic. When a document misbehaves, that vocabulary is also the repair kit — name the failure mode first, then fix that.
 **Should this live in `CLAUDE.md` or somewhere else?**
 Ask which load you want to pay. `CLAUDE.md` loads into every [session](https://www.aihero.dev/ai-coding-dictionary/session) unconditionally; material behind a pointer costs only the pointer's own line until it fires. Anything that applies in one context out of ten is paying context load the nine other times.
 **Do I need to rewrite my documents for each new model?**
 Mostly no, and over-fitting to one model is its own trap. Updating for a new model is usually another no-op pass rather than a rewrite.
 **My skill only works on the exact task I built it from.**
 The common route — do the work once, then have the agent write it up as a skill — over-indexes on that one run, and the exemplars come out too specific. Keep the run as evidence, then abstract deliberately: strip what belonged to that repo and those files, and write for the class of task.
 **English isn't my first language. Do I lose the leading-word advantage?**
 No — finding the word that packs the most behaviour into the fewest [tokens](https://www.aihero.dev/ai-coding-dictionary/token) is work the reference does for you. It is one of the things it is for.
 ## It's working if
 - The document gets shorter as it gets better, and you are surprised how little is left.
 - You can point at a leading word and watch it doing work in more than one place.
 - Nothing is stated twice, in any form. Duplication is the most reliable sign a document was never tested.
 - Reference that only one branch needs sits behind a pointer rather than in the main file.
 ## Where it fits
 This is a reach-for-it-anytime standalone reference. It has no neighbour in the chain because it sits underneath the whole set rather than beside any one skill: every skill here was written against it, and the documents the other skills leave behind — a `CONTEXT.md` and its ADRs, a spec, a ticket — are exactly the text it governs once an agent has to read them. When you're unsure which skill or flow fits a task, [ask-matt](https://aihero.dev/skills-ask-matt) routes you over the whole set.
@@ -10,7 +10,7 @@
 ## Language
 **Order**:
-{A one or two sentence description of the term}
+{A concise description of the term}
 _Avoid_: Purchase, transaction
 **Invoice**:
@@ -20,14 +20,31 @@ _Avoid_: Bill, payment request
 **Customer**:
 A person or organization that places orders.
 _Avoid_: Client, buyer, account
 ## Relationships
 - An **Order** produces one or more **Invoices**
 - An **Invoice** belongs to exactly one **Customer**
 ## Example dialogue
 > **Dev:** "When a **Customer** places an **Order**, do we create the **Invoice** immediately?"
 > **Domain expert:** "No — an **Invoice** is only generated once a **Fulfillment** is confirmed."
 ## Flagged ambiguities
 - "account" was used to mean both **Customer** and **User** — resolved: these are distinct concepts.
 ```
 ## Rules
- **Be opinionated.** When multiple words exist for the same concept, pick the best one and list the others under `_Avoid_`.
+- **Be opinionated.** When multiple words exist for the same concept, pick the best one and list the others as aliases to avoid.
- **Keep definitions tight.** One or two sentences max. Define what it IS, not what it does.
+- **Flag conflicts explicitly.** If a term is used ambiguously, call it out in "Flagged ambiguities" with a clear resolution.
 - **Keep definitions tight.** One sentence max. Define what it IS, not what it does.
 - **Show relationships.** Use bold term names and express cardinality where obvious.
 - **Only include terms specific to this project's context.** General programming concepts (timeouts, error types, utility patterns) don't belong even if the project uses them extensively. Before adding a term, ask: is this a concept unique to this context, or a general programming concept? Only the former belongs.
 - **Group terms under subheadings** when natural clusters emerge. If all terms belong to a single cohesive area, a flat list is fine.
 - **Write an example dialogue.** A conversation between a dev and a domain expert that demonstrates how the terms interact naturally and clarifies boundaries between related concepts.
 ## Single vs multi-context repos
@@ -1,13 +1,20 @@
 ---
-name: domain-modeling
+name: domain-model
-description: Build and sharpen a project's domain model. Use when the user wants to pin down domain terminology or a ubiquitous language, record an architectural decision, or when another skill needs to maintain the domain model.
+description: Grilling session that challenges your plan against the existing domain model, sharpens terminology, and updates documentation (CONTEXT.md, ADRs) inline as decisions crystallise. Use when user wants to stress-test a plan against their project's language and documented decisions.
 disable-model-invocation: true
 ---
-# Domain Modeling
+Interview me relentlessly about every aspect of this plan until we reach a shared understanding. Walk down each branch of the design tree, resolving dependencies between decisions one-by-one. For each question, provide your recommended answer.
-Actively build and sharpen the project's domain model as you design. This is the *active* discipline — challenging terms, inventing edge-case scenarios, and writing the glossary and decisions down the moment they crystallise. (Merely *reading* `CONTEXT.md` for vocabulary is not this skill — that's a one-line habit any skill can do. This skill is for when you're changing the model, not just consuming it.)
+Ask the questions one at a time, waiting for feedback on each question before continuing.
-## File structure
+If a question can be answered by exploring the codebase, explore the codebase instead.
 ## Domain awareness
 During codebase exploration, also look for existing documentation:
 ### File structure
 Most repos have a single context:
@@ -61,7 +68,7 @@ When the user states how something works, check whether the code agrees. If you
 When a term is resolved, update `CONTEXT.md` right there. Don't batch these up — capture them as they happen. Use the format in [CONTEXT-FORMAT.md](./CONTEXT-FORMAT.md).
-`CONTEXT.md` should be totally devoid of implementation details. Do not treat `CONTEXT.md` as a spec, a scratch pad, or a repository for implementation decisions. It is a glossary and nothing else.
+Don't couple `CONTEXT.md` to implementation details. Only include terms that are meaningful to domain experts.
 ### Offer ADRs sparingly
@@ -1,16 +0,0 @@
 # Drafts
 Beta. These skills are public on purpose — try them and tell me what breaks. They live outside `skills/`, so they're excluded from the plugin and the top-level README until they graduate, they get no docs pages, and they can change or disappear without warning.
 The plugin won't give you these. Install one directly:
 ```bash
 npx skills@latest add mattpocock/skills --skill=<name>
 ```
 - **[loop-me](./loop-me/SKILL.md)** — Grill yourself into implementable workflow specs over multiple sessions, using the current directory as a stateful workspace. User-invoked.
 - **[writing-beats](./writing-beats/SKILL.md)** — Shape an article as a journey of beats, choose-your-own-adventure style. Pick a starting beat, write only that beat, then pivot to the next, until the article reaches a natural end.
 - **[writing-fragments](./writing-fragments/SKILL.md)** — Grilling session that mines you for fragments — heterogeneous nuggets of writing — and appends them to a single document as raw material for a future article.
 - **[writing-shape](./writing-shape/SKILL.md)** — Take a markdown file of raw material and shape it into an article paragraph by paragraph, arguing format choices at each step.
 - **[claude-handoff](./claude-handoff/SKILL.md)** — Hand the current conversation off to a fresh background agent that picks up the work immediately, seeded with a handoff summary via `claude --bg`. User-invoked.
 - **[setup-ts-deep-modules](./setup-ts-deep-modules/SKILL.md)** — Wire dependency-cruiser into a TypeScript repo so each package is a deep module — implementation hidden in subfolders, reachable only through its entry-point files, tests exercising it through those. User-invoked.
@@ -1,18 +0,0 @@
 ---
 name: claude-handoff
 description: Hand the current conversation off to a fresh background agent that picks up the work immediately.
 argument-hint: "What will the next session be used for?"
 disable-model-invocation: true
 ---
 Write a handoff summary of the current conversation so a fresh agent can continue the work. Instead of saving it, launch a background agent seeded with the summary as its prompt: `claude --bg --name "<descriptive name>" "<handoff summary>"`. It starts in the current working directory and returns immediately; the user manages it with `claude agents`.
 Always pass `-n`/`--name` with a descriptive name (e.g. `--name "Fix login bug"`) — it sets the display name shown in the job list, session picker, and terminal title.
 Include a "suggested skills" section in the summary, which suggests skills that the agent should invoke.
 Do not duplicate content already captured in other artifacts (specs, plans, ADRs, issues, commits, diffs). Reference them by path or URL instead.
 Redact any sensitive information, such as API keys, passwords, or personally identifiable information — the summary becomes the agent's prompt.
 If the user passed arguments, treat them as a description of what the next session will focus on and tailor the summary accordingly.
@@ -1,5 +0,0 @@
 interface:
  display_name: "Claude Handoff"
  short_description: "Hand off to a background agent"
 policy:
  allow_implicit_invocation: false
@@ -1,32 +0,0 @@
 ---
 name: loop-me
 description: Grill me about specs for the workflows I want to build, within this workspace.
 disable-model-invocation: true
 argument-hint: "A workflow to design, or nothing to go find one"
 ---
 Run a stateful `/grilling` session whose only output is **workflow** specs. Use the grilling discipline — relentless, a round of questions at a time, a recommended answer attached to each — aimed at the vocabulary and goal below. Create, edit, and delete specs as the grilling resolves things.
 ## The loop lens
 A **loop** is a recurring pattern in the user's life: their career, their week, their morning, a single repeated activity. Picturing a life as loops within loops reveals how predictable its activities really are — which is what makes them worth **delegating**. Use the lens to find loops worth specifying, and propose ones the user hasn't noticed.
 A **workflow** is the spec of one loop, made real. You run a workflow on a loop — the loop is its running instantiation. Workflows live in `workflows/*.md` and are the source of truth.
 ## Vocabulary
 A shared language, reached for only when a workflow calls for it — never a checklist. **Mandate nothing structural**: a workflow needs no AI, no checkpoint, and no schedule unless the grilling shows it does.
 - **Trigger** — what fires each run: an **event** (a new email, a new issue) or a **schedule** (every morning). Event-triggering is usually the more efficient.
 - **Checkpoint** — a human-in-the-loop point where the user is asked to verify or decide. Some workflows have none and run autonomously; some use no AI at all.
 - **Push right** — defer the checkpoint as far as it will go. Do maximal work before involving the human, so they are asked once, late, with everything prepared.
 - **Brief** — what a checkpoint presents: a tight, decision-ready summary — what was produced, why, and a link down to the asset itself — never the raw output. The user reads a brief, not a draft. Speed of review is imperative.
 ## Definition of done
 A workflow spec is done when an implementer agent could build it without asking a single question. Grill until then; nothing is done while a question remains.
 ## The workspace
 - `workflows/*.md` — one spec per workflow.
 - `NOTES.md` — raw notes on the user's world: the tools they use, the channels they process, and their own terminology for both. When it is empty or thin, interview them about their world before specifying anything. Sharpen fuzzy terms into canonical ones as they surface, and record them here.
@@ -1,5 +0,0 @@
 interface:
  display_name: "Loop Me"
  short_description: "Spec the workflows you want to build"
 policy:
  allow_implicit_invocation: false
@@ -1,102 +0,0 @@
 ---
 name: setup-ts-deep-modules
 description: Wire dependency-cruiser into a TypeScript repo so each package is a deep module — implementation hidden in subfolders, reachable only through its entry-point files. User-invoked.
 disable-model-invocation: true
 ---
 # Setup TS Deep Modules
 Make every package in this repo a **deep module**: a lot of behaviour behind a small interface. A package's public surface is its **entry points** — the files at the package root — and everything in its subfolders is hidden. This skill installs [dependency-cruiser](https://github.com/sverweij/dependency-cruiser) and the rules that make the entry points the only way in, then proves the rules bite.
 For the vocabulary (deep module, interface, seam, depth), run the `/codebase-design` skill — use its language throughout.
 ## The shape this enforces
 ```
 src/packages/
  <name>/
    index.ts        ← an entry point (public). Import this from outside.
    client.ts       ← another entry point. Packages may expose SEVERAL.
    lib/            ← implementation: hidden from outside, free to import each other.
    tests/          ← co-located tests + fixtures (a subfolder, so private).
 ```
 The public surface is the package's **root files** — not one designated `index.ts`. By convention implementation lives in `lib/` and tests in `tests/`, giving every package the same two-folder shape. The rule itself is general, though: *anything* in *any* subfolder is private, so you never extend the config to add a folder.
 Four rules, all `error`:
 1. **Entry-point boundary** — code outside a package (app code or another package) may import only that package's entry points (its root files), never anything in its subfolders.
 2. **Intra-package freedom** — a package's own files import each other freely.
 3. **Tests through the entry points** — files under `<pkg>/tests/` may import any package's entry points and their own `tests/` fixtures, but never any package's subfolder internals (not even their own). Integration tests across packages are fine; deep imports are not.
 4. **No cycles** — no dependency cycles.
 **Entry points, not a barrel.** Because the public surface is *every* root file, a package can expose several small entry points (`index.ts`, `client.ts`, `server.ts`) instead of funnelling everything through one giant `index.ts`. Barrel files that re-export a whole subtree are discouraged — keep entry points small and hide implementation in subfolders.
 Layering (which packages may depend on which) is a *different* concern and is left as a commented stub in the config for this repo to fill in.
 ## Steps
 ### 1. Detect the environment
 - **Package manager** — `pnpm-lock.yaml` → pnpm, `yarn.lock` → yarn, `bun.lockb` → bun, else npm. Use it for every command below (`pnpm`/`yarn`/`npm run`/`bunx`).
 - **Packages root** — if `src/` exists use `src/packages`, else `packages`. Confirm the choice with the user if the repo already has a different obvious convention.
 - **Existing config** — check for a `.dependency-cruiser.*` file. If one exists, do **not** overwrite it: merge the four rules and the options in, and tell the user what you added.
 **Done when:** package manager, packages root, and existing-config status are all known.
 ### 2. Install dependency-cruiser
 Install `dependency-cruiser` as a devDependency with the detected package manager.
 **Done when:** `dependency-cruiser` is in `devDependencies`.
 ### 3. Write the config
 Copy [`dependency-cruiser.config.cjs`](./dependency-cruiser.config.cjs) to the repo root as `.dependency-cruiser.cjs`. Set `PACKAGES_ROOT` to the root detected in step 1. The rules are path-depth based and extension-agnostic, so nothing else needs adapting.
 **Done when:** `.dependency-cruiser.cjs` exists with the correct `PACKAGES_ROOT`, and the four forbidden rules are present.
 ### 4. Wire it into the checks
 - Add a `lint:boundaries` script: `depcruise <packages-root>` (or `depcruise src`).
 - Fold it into the repo's umbrella check command — the one that already runs typecheck (e.g. a `check` / `ci` / `validate` script). Do **not** touch `tsconfig` or add path aliases.
 - If there is no umbrella script, add `lint:boundaries` and tell the user to include it in CI.
 **Done when:** `lint:boundaries` exists and runs as part of the same command as typecheck.
 ### 5. Scaffold the example package
 Create a committed `<packages-root>/example/` as a copy-me template:
 - `index.ts` — an entry point. Export one function that delegates to an internal file (so the package is visibly *deep*, not a pass-through).
 - `lib/impl.ts` — an internal file in a **subfolder**, imported by `index.ts`, not reachable from outside.
 - `tests/example.test.ts` — imports **only** `../index` (an entry point), and asserts against the public function.
 Tell the user this is a starter template to copy or delete.
 **Done when:** the example package exists, exposes its behaviour through a root entry point, and hides `impl` in a subfolder.
 ### 6. Prove the rules bite
 This is the completion criterion for the whole skill — a config that doesn't fail on a violation is worthless.
 1. Run `lint:boundaries`. It must **pass** on the clean example.
 2. Temporarily add a deep import to `tests/example.test.ts` (e.g. `import { thing } from "../lib/impl"`). Run `lint:boundaries` again — it must **fail** with `tests-through-entrypoints`.
 3. Revert the deep import. Run once more — it must **pass**.
 **Done when:** you have observed a pass, then a fail on the deep import, then a pass again. If step 2 does not fail, the rules are not wired correctly — fix before finishing.
 ### 7. Document the convention
 Write a `README.md` **in the packages folder** (`<packages-root>/README.md`) — next to the packages it governs — covering: the `src/packages/<name>/` layout (entry points at the root, `lib/` for implementation, `tests/` for tests), "import only through a package's entry points (its root files)", and how to run `lint:boundaries`. **Discourage barrel files** explicitly — expose several small entry points instead of re-exporting a whole subtree through one index. Keep it to the copy-me snippet plus the four rules in one paragraph each.
 Then add a **context pointer** to it from the repo's agent-instructions file — `CLAUDE.md` if present, else `AGENTS.md` (create `AGENTS.md` if neither exists). One line is enough, e.g. `Packages are deep modules — see [src/packages/README.md](./src/packages/README.md) before adding or importing one.` This is what makes an agent discover the boundary rule instead of tripping over it.
 **Done when:** `<packages-root>/README.md` exists and discourages barrels, and the repo's `CLAUDE.md`/`AGENTS.md` links to it.
 ## Notes
 - The config's `$1` back-references (dependency-cruiser's group matching) are what let a package reach its own internals while outsiders can't — don't flatten them into separate per-package rules.
 - Public vs private is decided by **depth**: a package's root files are entry points; anything in a subfolder is private. The conventional subfolders are `lib/` (implementation) and `tests/`, but the rule doesn't hardcode them — any subfolder is private, so a new folder never needs a config change. Adding an entry point is just adding a root file — no barrel.
 - Packages are **flat**: one tier of immediate children under the root. A package's internals may nest as deep as you like; a package may not contain another package.
 - Use `.cjs` (not `.js`) so the config's `module.exports` works even in `"type": "module"` repos.
@@ -1,5 +0,0 @@
 interface:
  display_name: "Setup TS Deep Modules"
  short_description: "Enforce deep TypeScript modules"
 policy:
  allow_implicit_invocation: false
@@ -1,95 +0,0 @@
 // @ts-check
 // Deep-module enforcement for dependency-cruiser.
 //
 // Each package under the packages root is a DEEP MODULE: a lot of behaviour
 // behind a small interface. A package's PUBLIC SURFACE is its ENTRY POINTS —
 // the files at the package root. Implementation lives in SUBFOLDERS and is
 // private — by convention `lib/` for implementation and `tests/` for tests,
 // though any subfolder is private. A package may expose several small entry
 // points (index.ts, client.ts, server.ts, …) — prefer that over one giant
 // barrel index.
 //
 // The only thing you should ever need to edit here is PACKAGES_ROOT.
 /** Where packages live. One immediate child dir per package (flat, no nesting). */
 const PACKAGES_ROOT = "src/packages";
 // --- derived patterns (no need to edit) -------------------------------------
 const R = PACKAGES_ROOT;
 /**
 * A package's private internals: anything nested inside a package subfolder.
 * The package's root files are its entry points and are NOT matched here —
 * they stay importable from outside.
 */
 const PACKAGE_INTERNALS = `^${R}/[^/]+/[^/]+/`;
 /** @type {import('dependency-cruiser').IConfiguration} */
 module.exports = {
  forbidden: [
    {
      name: "entrypoint-boundary-from-app",
      comment:
        "App/root code may import a package's entry points (its root files), but nothing inside its subfolders.",
      severity: "error",
      from: { pathNot: `^${R}/` }, // importer is NOT inside any package
      to: { path: PACKAGE_INTERNALS },
    },
    {
      name: "entrypoint-boundary-across-packages",
      comment:
        "A package's own files import each other freely, but may reach OTHER packages only through their entry points — never their internals.",
      severity: "error",
      // importer is inside a package ($1), but is not a test file
      from: { path: `^${R}/([^/]+)/`, pathNot: `^${R}/[^/]+/tests/` },
      to: {
        path: PACKAGE_INTERNALS,
        pathNot: `^${R}/$1/`, // same package → intra-package freedom
      },
    },
    {
      name: "tests-through-entrypoints",
      comment:
        "A package's tests exercise it through its entry points like everyone else: they may import any package's entry points and their own tests/ fixtures, but never any package's internals — not even their own.",
      severity: "error",
      from: { path: `^${R}/([^/]+)/tests/` }, // a test file, in package $1
      to: {
        path: PACKAGE_INTERNALS,
        pathNot: `^${R}/$1/tests/`, // own tests/ fixtures → allowed
      },
    },
    {
      name: "tests-folder-is-private",
      comment:
        "A package's tests/ folder is reachable only from tests — nothing else may import fixtures.",
      severity: "error",
      from: { pathNot: `^${R}/[^/]+/tests/` }, // importer is not itself a test
      to: { path: `^${R}/[^/]+/tests/` },
    },
    {
      name: "no-circular",
      comment: "No dependency cycles. Scope to `^${R}/` if you want to allow cycles outside packages.",
      severity: "error",
      from: {},
      to: { circular: true },
    },
    // --- Layering (optional, off by default) ----------------------------------
    // Interface-hiding controls HOW you import (through the entry points).
    // Layering controls WHICH packages may depend on which. Add your own rules
    // here, e.g.:
    //
    // {
    //   name: "ui-may-not-depend-on-billing",
    //   severity: "error",
    //   from: { path: `^${R}/ui/` },
    //   to:   { path: `^${R}/billing/` },
    // },
  ],
  options: {
    doNotFollow: { path: "node_modules" },
    tsConfig: { fileName: "tsconfig.json" },
    enhancedResolveOptions: {
      extensions: [".ts", ".tsx", ".js", ".jsx", ".json"],
    },
  },
 };
@@ -1,67 +0,0 @@
 ---
 name: writing-beats
 description: Writing, exploit — assemble raw material into a journey of beats, grounding each term before a beat leans on it.
 disable-model-invocation: true
 ---
 <what-to-do>
 The user has passed (or will pass) a markdown file of raw material. This is **exploit**: the exploring is done, the pile is fixed — commit to a path through it and mine the pile to fill each beat.
 If the user did not say where to save the article, ask once and remember the path.
 Then run a beat-by-beat journey, choose-your-own-adventure style:
 1. **Establish the prerequisites.** Before any beats, settle with the user what the audience already knows walking in — the concepts that are **grounded** from the start. Everything else must be grounded by a beat before a later beat can use it. See [Grounding](#grounding).
 2. Write 2–3 candidate **starting beats**, drawn from the raw material. Each is a different entry point into the article. Each may only lean on grounded concepts; note what new concepts each one grounds. Show the user the beats before writing to the article file. The user picks one. Preview what beats that pick unlocks — as if the user is seeing a little way down the path.
 3. Once the user picks a starting beat, write **only that beat** to the article file. A beat may be one sentence or several paragraphs — whatever that beat naturally is. Stop there.
 4. Re-read the article file from disk. Then offer 2–3 candidate **next beats** — different directions the journey could pivot to from where the article now stands. Each must be reachable from the current grounded set; note what each one grounds.
 5. Loop steps 3–5 until the article reaches a natural end.
 </what-to-do>
 <supporting-info>
 ## Grounding
 Every **concept** has to be **grounded** before a beat can lean on it: the audience either walked in knowing it or met it in an earlier beat. A beat that reaches for an ungrounded concept loses the reader — that is the one move the journey can't make. The unit is the concept, not the word for it: a beat can lean on an idea the reader lacks even with no jargon in sight. Where a concept has a name — a **term** — grounding it means landing the idea and the term together.
 A concept gets grounded one of two ways:
 - **Prerequisite** — grounded before the first beat. The audience brings it. Fixed at the start.
 - **Introduced** — a beat establishes it, and from then on it's grounded for every later beat.
 So each beat does two jobs: it **requires** concepts that are already grounded, and it **grounds** new ones. Keep a running list of what's grounded so far, and update it each time a beat lands.
 This is what shapes the choose-your-own-adventure. A candidate beat is only reachable if everything it requires is already grounded; picking a beat that grounds concept X unlocks every beat that was waiting on X. When you offer next beats, they must all be reachable from the current grounded set — and say what each one grounds, so the user can see which paths it opens.
 The big lever is what you make a prerequisite versus what you ground inside the piece. Demand too much up front and you shut out readers who don't have it; ground too much inside and the early beats drown in definitions. Settle this with the user when you establish prerequisites, and revisit it whenever a tempting beat turns out to require a concept nothing has grounded yet — the fix is either a grounding beat before it, or promoting the concept to a prerequisite.
 ## What is a beat
 A beat is one move in the journey. It does one thing — sets a scene, lands a point, asks a question, drops an aside, twists the angle. Then it stops, leaving the reader at a place where the next beat can pivot.
 A beat is sized by what it needs:
 - A single sentence if that's all the move is ("And then nothing happened for three weeks.").
 - A short paragraph if the move needs setup.
 - Multiple paragraphs if the beat is a self-contained vignette, argument, or example.
 If a "beat" needs five paragraphs and three subheadings, it's not a beat — it's two beats glued together. Split it.
 ## Pulling from the pile
 Pull material from the raw pile to populate each beat. You can paraphrase, split, recombine, or quote. The pile is a quarry.
 ## Ending the journey
 The article ends when the journey is complete — not when the pile is empty. Most piles will have leftover fragments that don't make it in. That is fine; that is the point of having more raw material than you need.
 ## Writing rhythm
 - Append one beat at a time. Never write ahead.
 - Re-read the article file from disk before every write. Preserve user edits absolutely.
 - If the user edits a previous beat substantially, let it change what comes next.
 - If the user says "rewrite that beat" or "go back and try a different beat 3", do it — edit in place, leave the rest alone.
 </supporting-info>
@@ -1,5 +0,0 @@
 interface:
  display_name: "Writing Beats"
  short_description: "Assemble raw material into beats"
 policy:
  allow_implicit_invocation: false
@@ -1,79 +0,0 @@
 ---
 name: writing-fragments
 description: Writing, explore — mine raw fragments, no structure yet.
 disable-model-invocation: true
 ---
 <what-to-do>
 This is pure **explore**: widen the space of what could be written without committing to structure — committing is _exploit_, a separate skill's job. Run a grilling session that produces fragments, interviewing the user relentlessly about whatever they want to write about. Imposing phases, outlines, or article structure is out of scope here.
 As fragments emerge from either side of the conversation, append them to a single markdown file.
 If the user did not pass a path, ask once where to save the document, then remember it for the rest of the session.
 Capture fragments from the very first thing the user says, including the initial prompt.
 On first write, put a single H1 at the top with a working title (it can change later) and nothing else — no metadata, no TOC, no date.
 </what-to-do>
 <supporting-info>
 ## What is a fragment
 A fragment is any piece of text that might survive into the final article. It must be _readable by the author_ — the author can tell what it means — but it does not need to define its terms or be comprehensible to a cold reader. The bar is "is this a piece of good writing?", not "is this a self-contained argument?"
 Fragments are deliberately heterogeneous. Examples of what could be a fragment:
 - A sharp sentence you'd want to deploy somewhere but don't yet know where.
 - A claim with a one-line justification.
 - A vignette: a thing that happened, a code snippet, a scenario, an analogy.
 - A half-thought: "something about how X feels like Y, work this out later."
 - A quote, a piece of dialogue, an overheard line.
 - A list of related observations that hang together by feel.
 - A complaint, a confession, a punchline.
 - A **leading word** — a compact metaphor or coinage the whole piece can hang on (one term that names the idea, the way _tracer bullets_ or _fog of war_ names a whole pattern).
 Of these, the leading word is the most valuable fragment to land. It is load-bearing: name the right one in explore and it shapes the structure, the transitions, and the title later — paying dividends through the entire exploit phase. When the conversation circles a recurring idea, push to coin a word for it.
 The novelist's diary is the model: years of unstructured noticings that later get mined for raw material. Fragments are noticings.
 ## File format
 ```markdown
 # Working title
 A first fragment lives here.
 It can be multiple paragraphs. It can include lists, code, quotes — whatever
 shape the fragment naturally takes.
 ---
 A second fragment.
 ---
 > A quoted line that the user wants to keep around.
 A reaction to it.
 ---
 - A cluster of related observations
 - That hang together by feel
 - And want to be near each other
 ```
 Fragments are separated by a horizontal rule (`\n---\n`). No headings inside the body. No tags. No order beyond the order they were added.
 ## Writing rhythm
 Append silently. Don't ask permission for each fragment. Mention what you added in passing ("adding that"), but don't interrupt the conversation with save dialogs.
 Before every write: re-read the file from disk. The user may have edited, reordered, or deleted fragments between turns — preserve their changes. Never overwrite the file; only append (or, if the user asks, edit a specific fragment in place).
 The user can say "cut the last one", "rewrite that one sharper", "merge those two" at any time. Treat those as first-class instructions.
 </supporting-info>
@@ -1,5 +0,0 @@
 interface:
  display_name: "Writing Fragments"
  short_description: "Mine raw writing fragments"
 policy:
  allow_implicit_invocation: false
@@ -1,79 +0,0 @@
 ---
 name: writing-shape
 description: Writing, exploit — shape raw material into an article, paragraph by paragraph.
 disable-model-invocation: true
 ---
 <what-to-do>
 The user has passed (or will pass) a markdown file of raw material. Treat it as the input pile — anything from a tidy list of fragments to a wall of unstructured prose to a transcript. The format does not matter. Read it end-to-end before doing anything else.
 Then run a shaping session that produces a separate article document. This is **exploit**: the exploring is done, the pile is fixed — commit to a structure and mine the pile to fill it. Do not edit the raw material file — it is read-only to this skill.
 If the user did not say where to save the article, ask once and remember the path.
 </what-to-do>
 <supporting-info>
 ## The loop
 1. **Read the pile.** Read the input file in full. Form a sense of what's in it.
 2. **Establish the prerequisites.** Settle with the user what the reader knows walking in — the concepts that are **grounded** from the start. Everything else must be grounded by a block before a later block can lean on it. See [Grounding](#grounding).
 3. **Draft 2–3 candidate openings.** Each opening should imply a different thesis or angle for the article. Show all of them. Force the user to pick or compose a hybrid. The chosen opening defines what the rest of the article must do.
 4. **Grow paragraph by paragraph.** After the opening lands, ask "given this opening, what does the reader need to hear next?" Pull material from the pile to answer. The next block may only lean on grounded concepts, and grounds new ones as it lands. Argue about the form the next block takes — a paragraph, a list, a table, a callout, a quote, a code block. Each format choice should be deliberate and defensible.
 5. **Append to the article file as you go.** Don't batch. Write each agreed paragraph or block immediately so the user can see the article taking shape.
 6. **Loop step 4 until the article is done.** The user decides when it's done.
 ## Grounding
 Every **concept** has to be **grounded** before a block can lean on it: the reader either walked in knowing it or met it in an earlier block. A block that reaches for an ungrounded concept loses the reader. The unit is the concept, not the word for it — a block can lean on an idea the reader lacks even with no jargon in sight. Where a concept has a name — a **term** — grounding it means landing the idea and the term together.
 A concept gets grounded one of two ways:
 - **Prerequisite** — grounded before the opening. The reader brings it. Fixed at the start.
 - **Introduced** — a block establishes it, and from then on it's grounded for the rest of the article.
 Keep a running list of what's grounded. When you ask "what does the reader need to hear next?", an ungrounded concept the next move needs is itself the answer: ground it first — here or in an earlier block — or you can't make the move. This is the gap-naming of [Pulling from the pile](#pulling-from-the-pile) one level up: there the pile is missing material; here the article is missing a foundation.
 The lever is what you make a prerequisite versus what you ground inside the article. Demand too much up front and you shut readers out; ground too much inside and the opening drowns in definitions. Settle it with the user when you establish prerequisites.
 ## Conversational feel
 This is a grilling session inverted. In ideation, the question was "what are you actually noticing?" Here it's "what is this article actually arguing, and in what order does the reader need to hear it?" Push back. Refuse to let weak transitions slide. If a paragraph doesn't earn its place, cut it.
 Specific moves to keep using:
 - "What does this paragraph do for the reader that the previous one didn't?"
 - "If I cut this, what breaks?"
 - "Is this prose, or should it be a list? Why prose?"
 - "This sentence is doing two jobs — split it or pick one."
 - "The opening promised X. We've drifted to Y. Either re-thread it or change the opening."
 ## Pulling from the pile
 Treat the raw material as a quarry, not a script. Pull a fragment, rework it to fit the surrounding paragraph, and place it. A fragment may be split across multiple paragraphs, merged with another, or paraphrased. The pile's job is to be mined; the article's job is to read as one voice.
 If the pile lacks something the article needs, name the gap explicitly: "We need an example here and the pile doesn't have one — give me one now or we cut this section."
 ## Format arguments to actually have
 When choosing how to render a block, weigh these tradeoffs out loud with the user, not silently:
 - **Prose vs. list.** Prose carries argument; lists carry parallel items. If items aren't truly parallel, prose is better. If they are, a list is faster to scan.
 - **Inline vs. callout.** Tips, warnings, and asides go in callouts (`> [!TIP]`, `> [!NOTE]`) — but only if they'd genuinely derail the main argument inline. Otherwise leave them inline.
 - **Table vs. repeated structure.** If the same shape repeats 3+ times with the same fields, a table. Otherwise prose with bold leads.
 - **Quote vs. paraphrase.** Quote when the original wording is the point. Paraphrase when only the idea matters.
 - **Code block vs. inline code.** Multi-line, runnable, or illustrative → block. Single token or identifier → inline.
 ## Writing rhythm
 Append to the article file as each block is agreed. Re-read the file from disk before every write — the user may have edited between turns. Never overwrite blindly. If the user wants a paragraph rewritten, edit that specific paragraph in place; leave the rest alone.
 ## Out of scope
 - Mining for new fragments that aren't in the pile (handle gaps as in "Pulling from the pile").
 - Editing the raw material file.
 - Publishing, formatting for a specific platform, or adding frontmatter the user didn't ask for.
 </supporting-info>
@@ -1,5 +0,0 @@
 interface:
  display_name: "Writing Shape"
  short_description: "Shape raw material into an article"
 policy:
  allow_implicit_invocation: false
@@ -0,0 +1,14 @@
 ---
 name: edit-article
 description: Edit and improve articles by restructuring sections, improving clarity, and tightening prose. Use when user wants to edit, revise, or improve an article draft.
 ---
 1. First, divide the article into sections based on its headings. Think about the main points you want to make during those sections.
 Consider that information is a directed acyclic graph, and that pieces of information can depend on other pieces of information. Make sure that the order of the sections and their contents respects these dependencies.
 Confirm the sections with the user.
 2. For each section:
 2a. Rewrite the section to improve clarity, coherence, and flow. Use maximum 240 characters per paragraph.
@@ -1,8 +0,0 @@
 # Extras
 Tools I keep around but rarely use. They live outside `skills/`, so the plugin doesn't ship them.
 - **[git-guardrails-claude-code](./git-guardrails-claude-code/SKILL.md)** — Set up Claude Code hooks to block dangerous git commands (push, reset --hard, clean, etc.) before they execute.
 - **[migrate-to-shoehorn](./migrate-to-shoehorn/SKILL.md)** — Migrate test files from `as` type assertions to @total-typescript/shoehorn.
 - **[scaffold-exercises](./scaffold-exercises/SKILL.md)** — Create exercise directory structures with sections, problems, solutions, and explainers.
 - **[setup-pre-commit](./setup-pre-commit/SKILL.md)** — Set up Husky pre-commit hooks with lint-staged, Prettier, type checking, and tests.
@@ -1,3 +0,0 @@
 interface:
  display_name: "Git Guardrails for Claude Code"
  short_description: "Block dangerous git commands"
@@ -1,3 +0,0 @@
 interface:
  display_name: "Migrate to Shoehorn"
  short_description: "Replace test assertions with shoehorn"
@@ -1,3 +0,0 @@
 interface:
  display_name: "Scaffold Exercises"
  short_description: "Scaffold lint-ready course exercises"
@@ -1,3 +0,0 @@
 interface:
  display_name: "Setup Pre-Commit"
  short_description: "Add pre-commit quality checks"
@@ -1,8 +1,6 @@
 # Writing Agent Briefs
-An agent brief is a structured comment posted on a GitHub issue or PR when it moves to `ready-for-agent`. It is the authoritative specification that an AFK agent will work from. The original body and discussion are context — the agent brief is the contract.
+An agent brief is a structured comment posted on a GitHub issue when it moves to `ready-for-agent`. It is the authoritative specification that an AFK agent will work from. The original issue body and discussion are context — the agent brief is the contract.
 The brief states **what the agent should do**, which stretches to both surfaces: for an issue, that's building the change from nothing; for a PR, it's what's left to do *to the existing diff* — finish it, close gaps, address review points. Same principles either way; the PR example below shows the difference.
 ## Principles
@@ -145,43 +143,6 @@ checked for matches.
 - Bug reports (only enhancement rejections go to `.out-of-scope/`)
 ```
 ### Good agent brief (PR)
 For a PR, "Current behavior" describes the state of the diff, and the brief asks the agent to finish or fix it rather than build from scratch.
 ```markdown
 ## Agent Brief
 **Category:** enhancement
 **Summary:** Finish the contributor's `--json` output flag for `triage list`
 **Current behavior:**
 The PR adds a `--json` flag that serializes the issue list to JSON. The happy
 path works and the diff matches the project's command structure. Two gaps
 remain: errors are still printed as human text (not JSON), and the new flag has
 no test coverage.
 **Desired behavior:**
 With `--json`, all output — including errors — is well-formed JSON on stdout,
 and the command's exit codes are unchanged. The existing human-readable output
 is untouched when the flag is absent.
 **Key interfaces:**
 - The command's error path should emit `{ "error": string }` under `--json`
  instead of the plain-text error
 - Reuse the existing serializer the PR already added; don't introduce a second
 **Acceptance criteria:**
 - [ ] `triage list --json` emits valid JSON for both success and error cases
 - [ ] Exit codes match the non-JSON command
 - [ ] A test covers the `--json` success output and one error case
 - [ ] Default (non-JSON) output is byte-for-byte unchanged
 **Out of scope:**
 - Adding `--json` to any other command
 - Changing the JSON shape of the success payload the PR already defined
 ```
 ### Bad agent brief
 ```markdown
@@ -83,11 +83,7 @@ The maintainer may:
 ## When to write to `.out-of-scope/`
-Only when an **enhancement** (not a bug) is *rejected* as `wontfix`. This applies to enhancement PRs exactly as it does to issues — a rejected PR is recorded here so the same request doesn't return as fresh code.
+Only when an **enhancement** (not a bug) is rejected as `wontfix`. The flow:
 Do **not** write here when something is closed as `wontfix` because it's **already implemented**. That's a built feature, not a rejected one; recording it would poison the dedup checks with false rejections. Instead, the closing comment points to where the feature already lives.
 The flow:
 1. Maintainer decides a feature request is out of scope
 2. Check if a matching `.out-of-scope/` file already exists
@@ -0,0 +1,168 @@
 ---
 name: github-triage
 description: Triage GitHub issues through a label-based state machine. Use when user wants to create an issue, triage issues, review incoming bugs or feature requests, prepare issues for an AFK agent, or manage issue workflow.
 ---
 # GitHub Issue Triage
 Triage issues in the current repo using a label-based state machine. Infer the repo from `git remote`. Use `gh` for all GitHub operations.
 ## AI Disclaimer
 Every comment or issue posted to GitHub during triage **must** include the following disclaimer at the top of the comment body, before any other content:
 ```
 > *This was generated by AI during triage.*
 ```
 ## Reference docs
 - [AGENT-BRIEF.md](AGENT-BRIEF.md) — how to write durable agent briefs
 - [OUT-OF-SCOPE.md](OUT-OF-SCOPE.md) — how the `.out-of-scope/` knowledge base works
 ## Labels
 | Label             | Type     | Description                              |
 | ----------------- | -------- | ---------------------------------------- |
 | `bug`             | Category | Something is broken                      |
 | `enhancement`     | Category | New feature or improvement               |
 | `needs-triage`    | State    | Maintainer needs to evaluate this issue  |
 | `needs-info`      | State    | Waiting on reporter for more information |
 | `ready-for-agent` | State    | Fully specified, ready for AFK agent     |
 | `ready-for-human` | State    | Requires human implementation            |
 | `wontfix`         | State    | Will not be actioned                     |
 Every issue should have exactly **one** state label and **one** category label. If an issue has conflicting state labels (e.g. both `needs-triage` and `ready-for-agent`), flag the conflict and ask the maintainer which state is correct before doing anything else. Provide a recommendation.
 ## State Machine
 | Current State  | Can transition to | Who triggers it        | What happens                                                                                                         |
 | -------------- | ----------------- | ---------------------- | -------------------------------------------------------------------------------------------------------------------- |
 | `unlabeled`    | `needs-triage`    | Skill (on first look)  | Issue needs maintainer evaluation. Skill applies label after presenting recommendation.                              |
 | `unlabeled`    | `ready-for-agent` | Maintainer (via skill) | Issue is already well-specified and agent-suitable. Skill writes agent brief comment, applies label.                 |
 | `unlabeled`    | `ready-for-human` | Maintainer (via skill) | Issue requires human implementation. Skill writes a brief comment summarizing the task, applies label.               |
 | `unlabeled`    | `wontfix`         | Maintainer (via skill) | Issue is spam, duplicate, or out of scope. Skill closes with comment (and writes `.out-of-scope/` for enhancements). |
 | `needs-triage` | `needs-info`      | Maintainer (via skill) | Issue is underspecified. Skill posts triage notes capturing progress so far + questions for reporter.                |
 | `needs-triage` | `ready-for-agent` | Maintainer (via skill) | Grilling session complete, agent-suitable. Skill writes agent brief comment, applies label.                          |
 | `needs-triage` | `ready-for-human` | Maintainer (via skill) | Grilling session complete, needs human. Skill writes a brief comment summarizing the task, applies label.            |
 | `needs-triage` | `wontfix`         | Maintainer (via skill) | Maintainer decides not to action. Skill closes with comment (and writes `.out-of-scope/` for enhancements).          |
 | `needs-info`   | `needs-triage`    | Skill (detects reply)  | Reporter has replied. Skill surfaces to maintainer for re-evaluation.                                                |
 An issue can only move along these transitions. The maintainer can override any state directly (see Quick State Override below), but the skill should flag if the transition is unusual.
 ## Invocation
 The maintainer invokes `/github-triage` then describes what they want in natural language. The skill interprets the request and takes the appropriate action.
 Example requests:
 - "Show me anything that needs my attention"
 - "Let's look at #42"
 - "Move #42 to ready-for-agent"
 - "What's ready for agents to pick up?"
 - "Are there any unlabeled issues?"
 ## Workflow: Show What Needs Attention
 When the maintainer asks for an overview, query GitHub and present a summary grouped into three buckets:
 1. **Unlabeled issues** — new, no labels at all. These have never been triaged.
 2. **`needs-triage` issues** — maintainer needs to evaluate or continue evaluating.
 3. **`needs-info` issues with new activity** — the reporter has commented since the last triage notes comment. Check comment timestamps to determine this.
 Display counts per group. Within each group, show issues oldest first (longest-waiting gets attention first). For each issue, show: number, title, age, and a one-line summary of the issue body.
 Let the maintainer pick which issue to dive into.
 ## Workflow: Triage a Specific Issue
 ### Step 1: Gather context
 Before presenting anything to the maintainer:
 - Read the full issue: body, all comments, all labels, who reported it, when
 - If there are prior triage notes comments (from previous sessions), parse them to understand what has already been established
 - Explore the codebase to build context — understand the domain, relevant interfaces, and existing behavior related to the issue
 - Read `.out-of-scope/*.md` files and check if this issue matches or is similar to a previously rejected concept
 ### Step 2: Present a recommendation
 Tell the maintainer:
 - **Category recommendation:** bug or enhancement, with reasoning
 - **State recommendation:** where this issue should go, with reasoning
 - If it matches a prior out-of-scope rejection, surface that: "This is similar to `.out-of-scope/concept-name.md` — we rejected this before because X. Do you still feel the same way?"
 - A brief summary of what you found in the codebase that's relevant
 Then wait for the maintainer's direction. They may:
 - Agree and ask you to apply labels → do it
 - Want to flesh it out → start a /domain-model session
 - Override with a different state → apply their choice
 - Want to discuss → have a conversation
 ### Step 3: Bug reproduction (bugs only)
 If the issue is categorized as a bug, attempt to reproduce it before starting a /domain-model session. This will vary by codebase, but do your best:
 - Read the reporter's reproduction steps (if provided)
 - Explore the codebase to understand the relevant code paths
 - Try to reproduce the bug: run tests, execute commands, or trace the logic to confirm the reported behavior
 - If reproduction succeeds, report what you found to the maintainer — include the specific behavior you observed and where in the code it originates
 - If reproduction fails, report that too — the bug may be environment-specific, already fixed, or the report may be inaccurate
 - If the report lacks enough detail to attempt reproduction, note that — this is a strong signal the issue should move to `needs-info`
 The reproduction attempt informs the /domain-model session and the agent brief. A confirmed reproduction with a known code path makes for a much stronger brief.
 ### Step 4: /domain-model session (if needed)
 If the issue needs to be fleshed out before it's ready for an agent, interview the maintainer to build a complete specification. Use the /domain-model skill.
 ### Step 5: Apply the outcome
 Depending on the outcome:
 - **ready-for-agent** — post an agent brief comment (see [AGENT-BRIEF.md](AGENT-BRIEF.md))
 - **ready-for-human** — post a comment summarizing the task, what was established during triage, and why it needs human implementation. Use the same structure as an agent brief but note the reason it can't be delegated to an agent (e.g. requires judgment calls, external system access, design decisions, or manual testing).
 - **needs-info** — post triage notes with progress so far and questions for the reporter (see Needs Info Output below)
 - **wontfix (bug)** — post a polite comment explaining why, then close the issue
 - **wontfix (enhancement)** — write to `.out-of-scope/`, post a comment linking to it, then close the issue (see [OUT-OF-SCOPE.md](OUT-OF-SCOPE.md))
 - **needs-triage** — apply the label. Optionally leave a comment if there's partial progress to capture.
 ## Workflow: Quick State Override
 When the maintainer explicitly tells you to move an issue to a specific state (e.g. "move #42 to ready-for-agent"), trust their judgment and apply the label directly.
 Still show a confirmation of what you're about to do: which labels will be added/removed, and whether you'll post a comment or close the issue. But skip the /domain-model session entirely.
 If moving to `ready-for-agent` without a /domain-model session, ask the maintainer if they want to write a brief agent brief comment or skip it.
 ## Needs Info Output
 When moving an issue to `needs-info`, post a comment that captures the interview progress and tells the reporter what's needed:
 ```markdown
 ## Triage Notes
 **What we've established so far:**
 - point 1
 - point 2
 **What we still need from you (@reporter):**
 - question 1
 - question 2
 ```
 Include everything resolved during the /domain-model session in "established so far" — this work should not be lost. The questions for the reporter should be specific and actionable, not vague ("please provide more info").
 ## Resuming Previous Sessions
 When triaging an issue that already has triage notes from a previous session:
 1. Read all comments to find prior triage notes
 2. Parse what was already established
 3. Check if the reporter has answered any outstanding questions
 4. Present the maintainer with an updated picture: "Here's where we left off, and here's what the reporter has said since"
 5. Continue the /domain-model session from where it stopped — do not re-ask resolved questions
@@ -0,0 +1,10 @@
 ---
 name: grill-me
 description: Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me".
 ---
 Interview me relentlessly about every aspect of this plan until we reach a shared understanding. Walk down each branch of the design tree, resolving dependencies between decisions one-by-one. For each question, provide your recommended answer.
 Ask the questions one at a time.
 If a question can be answered by exploring the codebase, explore the codebase instead.
@@ -1,6 +1,6 @@
 # Deepening
-How to deepen a cluster of shallow modules safely, given its dependencies. Assumes the vocabulary in [SKILL.md](SKILL.md) — **module**, **interface**, **seam**, **adapter**.
+How to deepen a cluster of shallow modules safely, given its dependencies. Assumes the vocabulary in [LANGUAGE.md](LANGUAGE.md) — **module**, **interface**, **seam**, **adapter**.
 ## Dependency categories
@@ -1,8 +1,8 @@
-# Design It Twice
+# Interface Design
 When the user wants to explore alternative interfaces for a chosen deepening candidate, use this parallel sub-agent pattern. Based on "Design It Twice" (Ousterhout) — your first idea is unlikely to be the best.
-Uses the vocabulary in [SKILL.md](SKILL.md) — **module**, **interface**, **seam**, **adapter**, **leverage**.
+Uses the vocabulary in [LANGUAGE.md](LANGUAGE.md) — **module**, **interface**, **seam**, **adapter**, **leverage**.
 ## Process
@@ -18,7 +18,7 @@ Show this to the user, then immediately proceed to Step 2. The user reads and th
 ### 2. Spawn sub-agents
-Spawn 3+ sub-agents in parallel. Each must produce a **radically different** interface for the deepened module.
+Spawn 3+ sub-agents in parallel using the Agent tool. Each must produce a **radically different** interface for the deepened module.
 Prompt each sub-agent with a separate technical brief (file paths, coupling details, dependency category from [DEEPENING.md](DEEPENING.md), what sits behind the seam). The brief is independent of the user-facing problem-space explanation in Step 1. Give each agent a different design constraint:
@@ -27,7 +27,7 @@ Prompt each sub-agent with a separate technical brief (file paths, coupling deta
 - Agent 3: "Optimise for the most common caller — make the default case trivial."
 - Agent 4 (if applicable): "Design around ports & adapters for cross-seam dependencies."
-Include both [SKILL.md](SKILL.md) vocabulary and CONTEXT.md vocabulary in the brief so each sub-agent names things consistently with the architecture language and the project's domain language.
+Include both [LANGUAGE.md](LANGUAGE.md) vocabulary and CONTEXT.md vocabulary in the brief so each sub-agent names things consistently with the architecture language and the project's domain language.
 Each sub-agent outputs:
@@ -0,0 +1,53 @@
 # Language
 Shared vocabulary for every suggestion this skill makes. Use these terms exactly — don't substitute "component," "service," "API," or "boundary." Consistent language is the whole point.
 ## Terms
 **Module**
 Anything with an interface and an implementation. Deliberately scale-agnostic — applies equally to a function, class, package, or tier-spanning slice.
 _Avoid_: unit, component, service.
 **Interface**
 Everything a caller must know to use the module correctly. Includes the type signature, but also invariants, ordering constraints, error modes, required configuration, and performance characteristics.
 _Avoid_: API, signature (too narrow — those refer only to the type-level surface).
 **Implementation**
 What's inside a module — its body of code. Distinct from **Adapter**: a thing can be a small adapter with a large implementation (a Postgres repo) or a large adapter with a small implementation (an in-memory fake). Reach for "adapter" when the seam is the topic; "implementation" otherwise.
 **Depth**
 Leverage at the interface — the amount of behaviour a caller (or test) can exercise per unit of interface they have to learn. A module is **deep** when a large amount of behaviour sits behind a small interface. A module is **shallow** when the interface is nearly as complex as the implementation.
 **Seam** _(from Michael Feathers)_
 A place where you can alter behaviour without editing in that place. The *location* at which a module's interface lives. Choosing where to put the seam is its own design decision, distinct from what goes behind it.
 _Avoid_: boundary (overloaded with DDD's bounded context).
 **Adapter**
 A concrete thing that satisfies an interface at a seam. Describes *role* (what slot it fills), not substance (what's inside).
 **Leverage**
 What callers get from depth. More capability per unit of interface they have to learn. One implementation pays back across N call sites and M tests.
 **Locality**
 What maintainers get from depth. Change, bugs, knowledge, and verification concentrate at one place rather than spreading across callers. Fix once, fixed everywhere.
 ## Principles
 - **Depth is a property of the interface, not the implementation.** A deep module can be internally composed of small, mockable, swappable parts — they just aren't part of the interface. A module can have **internal seams** (private to its implementation, used by its own tests) as well as the **external seam** at its interface.
 - **The deletion test.** Imagine deleting the module. If complexity vanishes, the module wasn't hiding anything (it was a pass-through). If complexity reappears across N callers, the module was earning its keep.
 - **The interface is the test surface.** Callers and tests cross the same seam. If you want to test *past* the interface, the module is probably the wrong shape.
 - **One adapter means a hypothetical seam. Two adapters means a real one.** Don't introduce a seam unless something actually varies across it.
 ## Relationships
 - A **Module** has exactly one **Interface** (the surface it presents to callers and tests).
 - **Depth** is a property of a **Module**, measured against its **Interface**.
 - A **Seam** is where a **Module**'s **Interface** lives.
 - An **Adapter** sits at a **Seam** and satisfies the **Interface**.
 - **Depth** produces **Leverage** for callers and **Locality** for maintainers.
 ## Rejected framings
 - **Depth as ratio of implementation-lines to interface-lines** (Ousterhout): rewards padding the implementation. We use depth-as-leverage instead.
 - **"Interface" as the TypeScript `interface` keyword or a class's public methods**: too narrow — interface here includes every fact a caller must know.
 - **"Boundary"**: overloaded with DDD's bounded context. Say **seam** or **interface**.
@@ -0,0 +1,76 @@
 ---
 name: improve-codebase-architecture
 description: Find deepening opportunities in a codebase, informed by the domain language in CONTEXT.md and the decisions in docs/adr/. Use when the user wants to improve architecture, find refactoring opportunities, consolidate tightly-coupled modules, or make a codebase more testable and AI-navigable.
 ---
 # Improve Codebase Architecture
 Surface architectural friction and propose **deepening opportunities** — refactors that turn shallow modules into deep ones. The aim is testability and AI-navigability.
 ## Glossary
 Use these terms exactly in every suggestion. Consistent language is the point — don't drift into "component," "service," "API," or "boundary." Full definitions in [LANGUAGE.md](LANGUAGE.md).
 - **Module** — anything with an interface and an implementation (function, class, package, slice).
 - **Interface** — everything a caller must know to use the module: types, invariants, error modes, ordering, config. Not just the type signature.
 - **Implementation** — the code inside.
 - **Depth** — leverage at the interface: a lot of behaviour behind a small interface. **Deep** = high leverage. **Shallow** = interface nearly as complex as the implementation.
 - **Seam** — where an interface lives; a place behaviour can be altered without editing in place. (Use this, not "boundary.")
 - **Adapter** — a concrete thing satisfying an interface at a seam.
 - **Leverage** — what callers get from depth.
 - **Locality** — what maintainers get from depth: change, bugs, knowledge concentrated in one place.
 Key principles (see [LANGUAGE.md](LANGUAGE.md) for the full list):
 - **Deletion test**: imagine deleting the module. If complexity vanishes, it was a pass-through. If complexity reappears across N callers, it was earning its keep.
 - **The interface is the test surface.**
 - **One adapter = hypothetical seam. Two adapters = real seam.**
 This skill is _informed_ by the project's domain model — `CONTEXT.md` and any `docs/adr/`. The domain language gives names to good seams; ADRs record decisions the skill should not re-litigate. See [CONTEXT-FORMAT.md](../domain-model/CONTEXT-FORMAT.md) and [ADR-FORMAT.md](../domain-model/ADR-FORMAT.md).
 ## Process
 ### 1. Explore
 Read existing documentation first:
 - `CONTEXT.md` (or `CONTEXT-MAP.md` + each `CONTEXT.md` in a multi-context repo)
 - Relevant ADRs in `docs/adr/` (and any context-scoped `docs/adr/` directories)
 If any of these files don't exist, proceed silently — don't flag their absence or suggest creating them upfront.
 Then use the Agent tool with `subagent_type=Explore` to walk the codebase. Don't follow rigid heuristics — explore organically and note where you experience friction:
 - Where does understanding one concept require bouncing between many small modules?
 - Where are modules **shallow** — interface nearly as complex as the implementation?
 - Where have pure functions been extracted just for testability, but the real bugs hide in how they're called (no **locality**)?
 - Where do tightly-coupled modules leak across their seams?
 - Which parts of the codebase are untested, or hard to test through their current interface?
 Apply the **deletion test** to anything you suspect is shallow: would deleting it concentrate complexity, or just move it? A "yes, concentrates" is the signal you want.
 ### 2. Present candidates
 Present a numbered list of deepening opportunities. For each candidate:
 - **Files** — which files/modules are involved
 - **Problem** — why the current architecture is causing friction
 - **Solution** — plain English description of what would change
 - **Benefits** — explained in terms of locality and leverage, and also in how tests would improve
 **Use CONTEXT.md vocabulary for the domain, and [LANGUAGE.md](LANGUAGE.md) vocabulary for the architecture.** If `CONTEXT.md` defines "Order," talk about "the Order intake module" — not "the FooBarHandler," and not "the Order service."
 **ADR conflicts**: if a candidate contradicts an existing ADR, only surface it when the friction is real enough to warrant revisiting the ADR. Mark it clearly (e.g. _"contradicts ADR-0007 — but worth reopening because…"_). Don't list every theoretical refactor an ADR forbids.
 Do NOT propose interfaces yet. Ask the user: "Which of these would you like to explore?"
 ### 3. Grilling loop
 Once the user picks a candidate, drop into a grilling conversation. Walk the design tree with them — constraints, dependencies, the shape of the deepened module, what sits behind the seam, what tests survive.
 Side effects happen inline as decisions crystallize:
 - **Naming a deepened module after a concept not in `CONTEXT.md`?** Add the term to `CONTEXT.md` — same discipline as `/domain-model` (see [CONTEXT-FORMAT.md](../domain-model/CONTEXT-FORMAT.md)). Create the file lazily if it doesn't exist.
 - **Sharpening a fuzzy term during the conversation?** Update `CONTEXT.md` right there.
 - **User rejects the candidate with a load-bearing reason?** Offer an ADR, framed as: _"Want me to record this as an ADR so future architecture reviews don't re-suggest it?"_ Only offer when the reason would actually be needed by a future explorer to avoid re-suggesting the same thing — skip ephemeral reasons ("not worth it right now") and self-evident ones. See [ADR-FORMAT.md](../domain-model/ADR-FORMAT.md).
 - **Want to explore alternative interfaces for the deepened module?** See [INTERFACE-DESIGN.md](INTERFACE-DESIGN.md).
@@ -0,0 +1,59 @@
 ---
 name: obsidian-vault
 description: Search, create, and manage notes in the Obsidian vault with wikilinks and index notes. Use when user wants to find, create, or organize notes in Obsidian.
 ---
 # Obsidian Vault
 ## Vault location
 `/mnt/d/Obsidian Vault/AI Research/`
 Mostly flat at root level.
 ## Naming conventions
 - **Index notes**: aggregate related topics (e.g., `Ralph Wiggum Index.md`, `Skills Index.md`, `RAG Index.md`)
 - **Title case** for all note names
 - No folders for organization - use links and index notes instead
 ## Linking
 - Use Obsidian `[[wikilinks]]` syntax: `[[Note Title]]`
 - Notes link to dependencies/related notes at the bottom
 - Index notes are just lists of `[[wikilinks]]`
 ## Workflows
 ### Search for notes
 ```bash
 # Search by filename
 find "/mnt/d/Obsidian Vault/AI Research/" -name "*.md" | grep -i "keyword"
 # Search by content
 grep -rl "keyword" "/mnt/d/Obsidian Vault/AI Research/" --include="*.md"
 ```
 Or use Grep/Glob tools directly on the vault path.
 ### Create a new note
 1. Use **Title Case** for filename
 2. Write content as a unit of learning (per vault rules)
 3. Add `[[wikilinks]]` to related notes at the bottom
 4. If part of a numbered sequence, use the hierarchical numbering scheme
 ### Find related notes
 Search for `[[Note Title]]` across the vault to find backlinks:
 ```bash
 grep -rl "\\[\\[Note Title\\]\\]" "/mnt/d/Obsidian Vault/AI Research/"
 ```
 ### Find index notes
 ```bash
 find "/mnt/d/Obsidian Vault/AI Research/" -name "*Index*"
 ```
@@ -1,21 +0,0 @@
 {
  "name": "mattpocock-skills",
  "version": "1.2.3",
  "private": true,
  "description": "Matt Pocock's agent skills for real engineering",
  "repository": {
    "type": "git",
    "url": "https://github.com/mattpocock/skills"
  },
  "license": "MIT",
  "scripts": {
    "changeset": "changeset",
    "version": "changeset version && node scripts/sync-plugin-version.mjs",
    "check-plugin-version": "node scripts/sync-plugin-version.mjs --check"
  },
  "devDependencies": {
    "@changesets/changelog-github": "^0.7.0",
    "@changesets/cli": "^2.30.0"
  },
  "packageManager": "npm@10.9.4"
 }
@@ -1,22 +0,0 @@
 {
  "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
  "name": "mattpocock-skills",
  "version": "1.2.3",
  "description": "Matt Pocock's agent skills for real engineering — grilling, spec/ticket flows, TDD, code review, domain modelling and more. Plug-and-play, not vibe coding.",
  "author": {
    "name": "Matt Pocock",
    "url": "https://www.aihero.dev"
  },
  "homepage": "https://www.aihero.dev/s/skills-newsletter",
  "repository": "https://github.com/mattpocock/skills",
  "license": "MIT",
  "keywords": [
    "engineering",
    "skills",
    "tdd",
    "code-review",
    "grilling",
    "domain-modeling",
    "productivity"
  ]
 }
@@ -0,0 +1,130 @@
 ---
 name: qa
 description: Interactive QA session where user reports bugs or issues conversationally, and the agent files GitHub issues. Explores the codebase in the background for context and domain language. Use when user wants to report bugs, do QA, file issues conversationally, or mentions "QA session".
 ---
 # QA Session
 Run an interactive QA session. The user describes problems they're encountering. You clarify, explore the codebase for context, and file GitHub issues that are durable, user-focused, and use the project's domain language.
 ## For each issue the user raises
 ### 1. Listen and lightly clarify
 Let the user describe the problem in their own words. Ask **at most 2-3 short clarifying questions** focused on:
 - What they expected vs what actually happened
 - Steps to reproduce (if not obvious)
 - Whether it's consistent or intermittent
 Do NOT over-interview. If the description is clear enough to file, move on.
 ### 2. Explore the codebase in the background
 While talking to the user, kick off an Agent (subagent_type=Explore) in the background to understand the relevant area. The goal is NOT to find a fix — it's to:
 - Learn the domain language used in that area (check UBIQUITOUS_LANGUAGE.md)
 - Understand what the feature is supposed to do
 - Identify the user-facing behavior boundary
 This context helps you write a better issue — but the issue itself should NOT reference specific files, line numbers, or internal implementation details.
 ### 3. Assess scope: single issue or breakdown?
 Before filing, decide whether this is a **single issue** or needs to be **broken down** into multiple issues.
 Break down when:
 - The fix spans multiple independent areas (e.g. "the form validation is wrong AND the success message is missing AND the redirect is broken")
 - There are clearly separable concerns that different people could work on in parallel
 - The user describes something that has multiple distinct failure modes or symptoms
 Keep as a single issue when:
 - It's one behavior that's wrong in one place
 - The symptoms are all caused by the same root behavior
 ### 4. File the GitHub issue(s)
 Create issues with `gh issue create`. Do NOT ask the user to review first — just file and share URLs.
 Issues must be **durable** — they should still make sense after major refactors. Write from the user's perspective.
 #### For a single issue
 Use this template:
 ```
 ## What happened
 [Describe the actual behavior the user experienced, in plain language]
 ## What I expected
 [Describe the expected behavior]
 ## Steps to reproduce
 1. [Concrete, numbered steps a developer can follow]
 2. [Use domain terms from the codebase, not internal module names]
 3. [Include relevant inputs, flags, or configuration]
 ## Additional context
 [Any extra observations from the user or from codebase exploration that help frame the issue — e.g. "this only happens when using the Docker layer, not the filesystem layer" — use domain language but don't cite files]
 ```
 #### For a breakdown (multiple issues)
 Create issues in dependency order (blockers first) so you can reference real issue numbers.
 Use this template for each sub-issue:
 ```
 ## Parent issue
 #<parent-issue-number> (if you created a tracking issue) or "Reported during QA session"
 ## What's wrong
 [Describe this specific behavior problem — just this slice, not the whole report]
 ## What I expected
 [Expected behavior for this specific slice]
 ## Steps to reproduce
 1. [Steps specific to THIS issue]
 ## Blocked by
 - #<issue-number> (if this issue can't be fixed until another is resolved)
 Or "None — can start immediately" if no blockers.
 ## Additional context
 [Any extra observations relevant to this slice]
 ```
 When creating a breakdown:
 - **Prefer many thin issues over few thick ones** — each should be independently fixable and verifiable
 - **Mark blocking relationships honestly** — if issue B genuinely can't be tested until issue A is fixed, say so. If they're independent, mark both as "None — can start immediately"
 - **Create issues in dependency order** so you can reference real issue numbers in "Blocked by"
 - **Maximize parallelism** — the goal is that multiple people (or agents) can grab different issues simultaneously
 #### Rules for all issue bodies
 - **No file paths or line numbers** — these go stale
 - **Use the project's domain language** (check UBIQUITOUS_LANGUAGE.md if it exists)
 - **Describe behaviors, not code** — "the sync service fails to apply the patch" not "applyPatch() throws on line 42"
 - **Reproduction steps are mandatory** — if you can't determine them, ask the user
 - **Keep it concise** — a developer should be able to read the issue in 30 seconds
 After filing, print all issue URLs (with blocking relationships summarized) and ask: "Next issue, or are we done?"
 ### 5. Continue the session
 Keep going until the user says they're done. Each issue is independent — don't batch them.
@@ -0,0 +1,68 @@
 ---
 name: request-refactor-plan
 description: Create a detailed refactor plan with tiny commits via user interview, then file it as a GitHub issue. Use when user wants to plan a refactor, create a refactoring RFC, or break a refactor into safe incremental steps.
 ---
 This skill will be invoked when the user wants to create a refactor request. You should go through the steps below. You may skip steps if you don't consider them necessary.
 1. Ask the user for a long, detailed description of the problem they want to solve and any potential ideas for solutions.
 2. Explore the repo to verify their assertions and understand the current state of the codebase.
 3. Ask whether they have considered other options, and present other options to them.
 4. Interview the user about the implementation. Be extremely detailed and thorough.
 5. Hammer out the exact scope of the implementation. Work out what you plan to change and what you plan not to change.
 6. Look in the codebase to check for test coverage of this area of the codebase. If there is insufficient test coverage, ask the user what their plans for testing are.
 7. Break the implementation into a plan of tiny commits. Remember Martin Fowler's advice to "make each refactoring step as small as possible, so that you can always see the program working."
 8. Create a GitHub issue with the refactor plan. Use the following template for the issue description:
 <refactor-plan-template>
 ## Problem Statement
 The problem that the developer is facing, from the developer's perspective.
 ## Solution
 The solution to the problem, from the developer's perspective.
 ## Commits
 A LONG, detailed implementation plan. Write the plan in plain English, breaking down the implementation into the tiniest commits possible. Each commit should leave the codebase in a working state.
 ## Decision Document
 A list of implementation decisions that were made. This can include:
 - The modules that will be built/modified
 - The interfaces of those modules that will be modified
 - Technical clarifications from the developer
 - Architectural decisions
 - Schema changes
 - API contracts
 - Specific interactions
 Do NOT include specific file paths or code snippets. They may end up being outdated very quickly.
 ## Testing Decisions
 A list of testing decisions that were made. Include:
 - A description of what makes a good test (only test external behavior, not implementation details)
 - Which modules will be tested
 - Prior art for the tests (i.e. similar types of tests in the codebase)
 ## Out of Scope
 A description of the things that are out of scope for this refactor.
 ## Further Notes (optional)
 Any further notes about the refactor.
 </refactor-plan-template>
@@ -0,0 +1,17 @@
 # Research
 Research notes informing the infrastructure plan for this repo.
 - [superpowers.md](./superpowers.md) — `obra/superpowers`: a single skill bundle that ships as Claude Code, Codex App, Cursor, OpenCode, and Gemini plugins simultaneously by colocating manifests at the repo root. Themed grouping happens in a _separate_ curation repo (`obra/superpowers-marketplace`), not as a monorepo.
 - [marketingskills.md](./marketingskills.md) — `coreyhaines31/marketingskills`: minimalist single-plugin marketplace, zero npm deps, all codegen via Node stdlib + bash. Notable patterns: `evals/evals.json` per skill, `VERSIONS.md` auto-update protocol, sync-skills.js GHA that auto-rewrites the marketplace and README on push.
 ## Cross-cutting takeaways
 | Goal                 | superpowers                                                            | marketingskills                           | Implication                                                                                                                                                                   |
 | -------------------- | ---------------------------------------------------------------------- | ----------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
 | `npx skills add`     | not used                                                               | first-class                               | "Just be SKILL.md-spec compliant under `skills/`" — no work                                                                                                                   |
 | Claude Code plugins  | `.claude-plugin/` at root, themed groups via separate marketplace repo | single-plugin marketplace                 | Neither models a multi-plugin monorepo. Matt's plan (multiple plugins under `plugins/<name>/`, one marketplace.json listing them with `source: "./plugins/<name>"`) is novel. |
 | Codex plugins        | `.codex-plugin/plugin.json` (Codex App) + `.codex/INSTALL.md` (CLI)    | implicit via AGENTS.md only               | If Matt wants Codex App marketplace presence, copy superpowers' `.codex-plugin/`. If just CLI, AGENTS.md suffices.                                                            |
 | Docs site            | none                                                                   | none                                      | No prior art — Matt would be the first.                                                                                                                                       |
 | Build tooling        | bash + jq, no package.json scripts                                     | Node stdlib + bash, no deps               | Both prove you don't need a build pipeline. Multi-plugin assembly may change that.                                                                                            |
 | SKILL.md frontmatter | `name`, `description` only                                             | `name`, `description`, `metadata.version` | Keep it minimal.                                                                                                                                                              |
@@ -0,0 +1,152 @@
 # Research: `coreyhaines31/marketingskills`
 A marketing-themed Agent Skills repo by Corey Haines. The infrastructure is genuinely interesting and notably *minimal*: zero npm dependencies, all codegen done with Node stdlib + bash.
 ## Repo structure
 ```
 .claude-plugin/        marketplace.json, plugin.json
 .github/               workflows + sync script + issue/PR templates
 skills/                40 skill dirs (flat, one level deep)
 tools/                 clis/, integrations/, composio/, REGISTRY.md
 AGENTS.md  CLAUDE.md  CONTRIBUTING.md  README.md  VERSIONS.md
 validate-skills.sh  validate-skills-official.sh
 ```
 `CLAUDE.md` is a one-line file: `AGENTS.md` (i.e. it points Claude at the cross-agent file, treating AGENTS.md as the source of truth).
 Each skill directory is uniform:
 ```
 skills/<name>/
 ├── SKILL.md
 ├── evals/evals.json        # eval prompts + assertions
 └── references/*.md         # progressive-disclosure docs
 ```
 No `scripts/` or `assets/`. Skills are flat (no nesting / sub-skills). **No `plugins/` or `dist/` directories** — the repo *is* the plugin, served from root.
 ## Plugin packaging
 Ships as a Claude Code plugin marketplace via two manifests:
 `.claude-plugin/marketplace.json`:
 ```json
 {
  "name": "marketingskills",
  "owner": { "name": "Corey Haines", "url": "https://corey.co" },
  "metadata": { "version": "1.9.0" },
  "plugins": [
    { "name": "marketing-skills",
      "description": "40 marketing skills...",
      "source": "./" }
  ]
 }
 ```
 `.claude-plugin/plugin.json`:
 ```json
 { "name": "marketing-skills", "version": "1.9.0",
  "skills": "./skills", "license": "MIT" }
 ```
 **One marketplace, one plugin, all 40 skills bundled.** No themed sub-grouping into multiple plugins — the README's "categories" exist only as headings, not separate installable units. (Relevant for Matt's multi-plugin goal: this repo doesn't actually demonstrate a multi-plugin marketplace.)
 VERSIONS.md notes the `plugin.json` was added explicitly so Claude Code's loader recognizes the skills directory.
 ## Build tooling
 **No package.json, no TypeScript, no bundler, no dist build.** The only Node code is `.github/scripts/sync-skills.js` — a zero-dependency script (`fs`, `path` only) that:
 1. Walks `skills/*/SKILL.md`
 2. Parses YAML frontmatter with a hand-rolled regex (`/^---\n([\s\S]*?)\n---/`) — naïve key:value splitter, no real YAML lib
 3. Rewrites the `plugins[0].skills` array in `marketplace.json` to a list of `./skills/<name>` paths
 4. Updates the skill count in the plugin description (`/\d+ marketing skills/`)
 5. Replaces the `<!-- SKILLS:START -->...<!-- SKILLS:END -->` block in README.md with a regenerated table (description truncated to 120 chars at a word boundary)
 That's the entire codegen story. `validate-skills.sh` and `validate-skills-official.sh` are bash-only frontmatter linters using `sed`/`grep`.
 ## Installation story
 Six options listed in the README, in priority order:
 1. **vercel-labs/skills CLI** — `npx skills add coreyhaines31/marketingskills [--skill page-cro copywriting] [--list]`. README claims this installs to `.agents/skills/` and symlinks `.claude/skills/` for Claude Code.
 2. **Claude Code plugins** — `/plugin marketplace add coreyhaines31/marketingskills` then `/plugin install marketing-skills`.
 3. Clone-and-copy.
 4. Git submodule.
 5. Fork.
 6. **SkillKit** (`npx skillkit install ...`) for cross-agent install (Cursor/Copilot/etc).
 **Codex support** is implicit: the repo ships `AGENTS.md` (the cross-agent spec file Codex reads) and the README claims compatibility with "Claude Code, OpenAI Codex, Cursor, Windsurf, and any agent that supports the Agent Skills spec." There is **no Codex-specific plugin manifest** — Codex compatibility comes purely from being SKILL.md-spec-compliant + AGENTS.md at root.
 ## Docs site
 **There is no embedded docs site.** No `docs/`, no `site/`, no `website/`, no Docusaurus/Nextra/Astro config, `has_pages: false`. The README claims `homepage: https://marketing-skills.com` but it is not built from this repo.
 The only generated docs artifact is the README skills table (built by `sync-skills.js`).
 ## CI / release
 Two workflows:
 - **`.github/workflows/sync-skills.yml`** — on push to `main` touching `skills/**`, runs `sync-skills.js` and commits the result back via `stefanzweifel/git-auto-commit-action@v7` as a bot user. This is the "build step" — it auto-rewrites `marketplace.json` and `README.md` whenever skills change.
 - **`.github/workflows/validate-skill.yml`** — on push/PR touching `**/SKILL.md`. A `detect-changes` job computes the changed skill dirs via git diff and a jq-built JSON array, then a matrix `validate` job runs `Flash-Brew-Digital/validate-skill@v1` on each. Skips drafts and dependabot.
 No release/publish workflow — versioning is tracked manually in `VERSIONS.md` (per-skill semver + dates) and in `marketplace.json`/`plugin.json`'s top-level `version`.
 ## SKILL.md format
 Frontmatter is minimal:
 ```yaml
 ---
 name: page-cro
 description: When the user wants to optimize... [trigger phrases and "For X, see other-skill" cross-refs]
 metadata:
  version: 1.1.0
 ---
 ```
 Conventions documented in `AGENTS.md`:
 - `name`: 1-64 chars, lowercase a-z + digits + hyphens, must match dir
 - `description`: 1-1024 chars, *must* include trigger phrases AND scope-boundary cross-references to sibling skills
 - `metadata.version` only (no author/license per-skill)
 - SKILL.md kept under 500 lines; details pushed into `references/`
 Distinctive description style: every skill enumerates trigger-phrase-laden quotes ("CRO," "this page isn't converting," "my landing page sucks") plus explicit `For X, see other-skill` boundaries — clearly tuned for the Skill-tool dispatcher's keyword matching.
 ## Distinctive / novel patterns
 1. **Per-skill `evals/evals.json`** — every skill has structured eval prompts with `expected_output` summary + a list of `assertions` strings. Not wired into CI but provides a dataset for offline eval runs.
 2. **Auto-update protocol baked into `AGENTS.md`** — instructs the agent to fetch `VERSIONS.md` from raw.githubusercontent once per session, compare local versions, and surface a non-blocking notification if 2+ skills are stale or any has a major bump. Includes a "say 'update skills'" trigger that runs `git pull`. This is a memory-less, network-fetched update channel.
 3. **`product-marketing-context` as a hub skill** — every other skill is documented to read `.agents/product-marketing-context.md` (with `.claude/` fallback) before doing anything. README diagram shows it as the root of a star topology.
 4. **Claude Code-only escape hatch in AGENTS.md** — explicitly calls out that `` !`command` `` shell-injection syntax is Claude Code-only and **must not** be in the cross-agent SKILL.md files. Suggests local override in `.claude/skills/` if you want it. Clean pattern: keep skills cross-agent, document Claude-Code-only enhancements separately.
 5. **`tools/` registry** — orthogonal to skills, not part of the plugin. 60+ zero-dep Node CLIs (`tools/clis/<vendor>.js`) plus 80+ markdown integration guides plus a Composio mapping. AGENTS.md tells the agent: skills *reference* tools by name, agent reads `tools/REGISTRY.md` and `tools/integrations/<tool>.md` on demand. Effectively a second progressive-disclosure layer beyond `references/`.
 6. **No package.json / no JS deps** — the entire codegen + validation pipeline is `node` (stdlib only) + `bash`. Maximally portable.
 7. **No agents, no hooks, no slash-commands** — README hints at `/page-cro` invocations, but no `commands/` directory exists. These are skill-name invocations the Claude Code plugin loader produces automatically.
 ## Takeaways for Matt's infra goals
 - **Multi-plugin marketplace (goal 2)**: this repo is *not* a model for that — single-plugin marketplace. The marketplace.json schema does support multiple `plugins[]` entries pointing to different `source` paths, so for themed groups Matt would want each plugin to live in its own subdir (e.g. `plugins/architecture/`, `plugins/typescript/`) each with their own `skills/`. coreyhaines31 doesn't demonstrate this.
 - **Codex (goal 3)**: cheap — drop AGENTS.md at root, keep SKILL.md spec-compliant, claim compatibility. No separate manifest needed.
 - **vercel-labs CLI (goal 1)**: zero work — once `skills/<name>/SKILL.md` exists with valid frontmatter, the CLI picks it up. coreyhaines31 added nothing extra.
 - **Docs site (goal 4)**: not modeled here; this repo opts out and links to a separately-hosted marketing site.
 - **Worth copying**: the `sync-skills.js` + GitHub Action pattern (auto-rewrite README table and marketplace skill list on push), the `VERSIONS.md` + auto-update protocol, the `evals/evals.json` per skill, and the AGENTS.md note about keeping `` !`command` `` out of cross-agent SKILL.md.
 ## Key file paths
 (all on `main` branch of `coreyhaines31/marketingskills`)
 - `/.claude-plugin/marketplace.json`, `/.claude-plugin/plugin.json`
 - `/.github/scripts/sync-skills.js`
 - `/.github/workflows/sync-skills.yml`, `/.github/workflows/validate-skill.yml`
 - `/AGENTS.md` (the canonical agent guide; `CLAUDE.md` just points to it)
 - `/VERSIONS.md`
 - `/skills/<name>/SKILL.md` + `/skills/<name>/evals/evals.json` + `/skills/<name>/references/*.md`
 - `/validate-skills.sh`, `/validate-skills-official.sh`
@@ -0,0 +1,110 @@
 # Research: `obra/superpowers`
 A single plugin/skills bundle that ships itself as a Claude Code plugin, a Codex plugin, a Cursor plugin, an OpenCode plugin, and a Gemini extension *simultaneously* by colocating every harness's manifest at the repo root.
 ## Repo structure
 ```
 .claude-plugin/      marketplace.json + plugin.json (Claude Code)
 .codex-plugin/       plugin.json (Codex App)
 .codex/              INSTALL.md (Codex CLI bootstrap docs)
 .cursor-plugin/      plugin.json (Cursor)
 .opencode/           INSTALL.md + plugins/superpowers.js (OpenCode)
 gemini-extension.json (Gemini CLI)
 agents/              code-reviewer.md
 commands/            brainstorm.md, execute-plan.md, write-plan.md
 hooks/               hooks.json, hooks-cursor.json, run-hook.cmd, session-start
 skills/              14 top-level SKILL.md folders (FLAT, not nested)
 docs/                handwritten markdown — README.codex.md, README.opencode.md, plans/, specs/
 scripts/             bump-version.sh, sync-to-codex-plugin.sh
 tests/               brainstorm-server, claude-code, codex-plugin-sync, opencode, skill-triggering, subagent-driven-dev
 assets/              icons, logos
 package.json, AGENTS.md, CLAUDE.md, GEMINI.md, RELEASE-NOTES.md
 ```
 No `dist/`, no `plugins/` subdir, no nested skills. All skills live at `skills/<kebab-name>/SKILL.md` (14 of them). Skills can have sibling reference files or a `references/` subdir.
 ## Plugin packaging — multi-harness, not multi-plugin
 The repo ships **one logical plugin packaged six ways**:
 - **Claude Code**: `.claude-plugin/plugin.json` + `.claude-plugin/marketplace.json` (the marketplace declares a single plugin named `superpowers` with `"source": "./"`, so the repo *is* its own dev marketplace).
 - **Codex App**: `.codex-plugin/plugin.json` — much richer than the Claude one, includes an `interface` block (`displayName`, `shortDescription`, `defaultPrompt`, `brandColor`, `composerIcon`, `logo`, `category: "Coding"`).
 - **Cursor**: `.cursor-plugin/plugin.json` — declares `skills`, `agents`, `commands`, `hooks: ./hooks/hooks-cursor.json`.
 - **OpenCode**: `.opencode/plugins/superpowers.js` — a real JS module that injects bootstrap context via system-prompt transform and parses SKILL.md frontmatter at runtime.
 - **Codex CLI**: install via `git clone` + `ln -s skills ~/.agents/skills/superpowers` (no plugin manifest, uses native skill discovery).
 - **Gemini CLI**: `gemini-extension.json` with `contextFileName: "GEMINI.md"`.
 The **separate marketplace repo** `obra/superpowers-marketplace` is where themed grouping happens. Its `.claude-plugin/marketplace.json` lists 7 plugins by remote git URL: `superpowers`, `superpowers-chrome`, `elements-of-style`, `episodic-memory`, `superpowers-lab`, `superpowers-developing-for-claude-code`, `superpowers-dev`. Each entry is `{name, source: {source: "url", url}, description, version, strict: true}`. So the marketplace is a thin curation layer pointing at independent plugin repos — **no monorepo, no themed sub-plugins inside one repo**.
 ## Build tooling
 `package.json` is minimal — just `name`, `version`, `type: "module"`, `main` pointing at the OpenCode plugin file. **No scripts, no dependencies, no build step, no TypeScript.** Manifests are handwritten and kept in sync by a custom version-bump tool.
 Two interesting scripts:
 1. **`scripts/bump-version.sh`** — driven by `.version-bump.json`, which lists every file containing a version field (`package.json`, `.claude-plugin/plugin.json`, `.cursor-plugin/plugin.json`, `.codex-plugin/plugin.json`, `.claude-plugin/marketplace.json` at `plugins.0.version`, `gemini-extension.json`). Has `--check` (drift detection) and `--audit` (greps repo for stale version strings). Pure bash + jq.
 2. **`scripts/sync-to-codex-plugin.sh`** — ~300-line bash tool that rsyncs the upstream tree into a fork (`prime-radiant-inc/openai-codex-plugins/plugins/superpowers/`), opens a PR via `gh`. Has `--dry-run`, `--bootstrap`, `--local`. Deterministic: same upstream SHA → identical PR diff. This is how the Codex App marketplace gets updated.
 ## Installation story
 **No `npx` installer of its own.** Installation is per-harness:
 - Claude Code: `/plugin marketplace add obra/superpowers-marketplace` then `/plugin install superpowers@superpowers-marketplace`, OR Anthropic's official marketplace `/plugin install superpowers@claude-plugins-official`.
 - Codex App: search & install via in-app plugin UI.
 - Codex CLI: clone + symlink (manual).
 - Cursor: `/add-plugin superpowers`.
 - OpenCode: tell agent to fetch `INSTALL.md`.
 - Gemini: `gemini extensions install https://github.com/obra/superpowers`.
 - Copilot CLI: `copilot plugin marketplace add obra/superpowers-marketplace`.
 Vercel-labs/skills is **not mentioned anywhere**.
 ## Docs site
 **There is no docs site.** No Docusaurus/Nextra/Astro/Starlight, no GitHub Pages (`has_pages: false`). `docs/` is a flat folder of handwritten markdown plus design specs dated YYYY-MM-DD. All discovery happens via the README plus per-harness INSTALL.md files.
 ## CI / release
 `.github/` contains only `FUNDING.yml`, `ISSUE_TEMPLATE/`, and `PULL_REQUEST_TEMPLATE.md` — **no GitHub Actions workflows**. Releases appear to be manual: bump versions with the script, push tag, the marketplace repo points at git URLs (and `superpowers-dev` entry pins `ref: "dev"`).
 ## SKILL.md format
 Frontmatter is intentionally minimal — only `name` and `description`:
 ```yaml
 ---
 name: test-driven-development
 description: Use when implementing any feature or bugfix, before writing implementation code
 ---
 ```
 Body uses heavy XML-style emphasis tags (`<EXTREMELY-IMPORTANT>`, `<SUBAGENT-STOP>`, `<important-reminder>`) and prescriptive uppercase rules. No tags, no allowed-tools, no version field per skill.
 ## Distinctive patterns
 - **Bootstrap via SessionStart hook**: `hooks/hooks.json` registers a `SessionStart` (matchers `startup|clear|compact`) that runs `hooks/run-hook.cmd session-start`, which reads `skills/using-superpowers/SKILL.md` and injects it as additional system context — this is how skills "trigger automatically" without the user opting in.
 - **`using-superpowers` as meta-skill**: forces the agent to invoke the Skill tool *before any reply, including clarifying questions*.
 - **OpenCode runtime SKILL.md parser**: `.opencode/plugins/superpowers.js` reimplements frontmatter extraction inline ("avoid dependency on skills-core for bootstrap").
 - **Tests for skill triggering**: `tests/skill-triggering/`, `tests/codex-plugin-sync/`, `tests/opencode/` — actual integration tests for whether agents invoke the right skill.
 - **Multi-harness CLAUDE.md/AGENTS.md/GEMINI.md** at root, one per agent's convention.
 - **Spec-driven development** visible in `docs/superpowers/specs/`.
 ## Implications for Matt's repo
 - **Multi-plugin themed marketplace**: superpowers' pattern is **separate plugin repos + one curation repo with `.claude-plugin/marketplace.json`** listing them by URL — *not* a monorepo. If Matt wants a monorepo, he'd be inventing a different pattern.
 - **Codex support**: cheapest path is `.codex-plugin/plugin.json` at root (Codex App) plus `.codex/INSTALL.md` for CLI clone+symlink. No build step needed.
 - **Docs site**: superpowers offers no precedent — Matt would be ahead of it.
 - **Build tooling**: superpowers proves you can run a popular skills plugin with **zero npm scripts** and just two bash scripts (version bump + cross-repo sync). The `.version-bump.json` config-driven approach is worth copying.
 - **SKILL.md conventions**: keep frontmatter to `name` + `description`; lean on body prose for behavior.
 ## Key URLs
 - https://github.com/obra/superpowers/blob/main/.claude-plugin/marketplace.json
 - https://github.com/obra/superpowers/blob/main/.codex-plugin/plugin.json
 - https://github.com/obra/superpowers/blob/main/.opencode/plugins/superpowers.js
 - https://github.com/obra/superpowers/blob/main/scripts/bump-version.sh
 - https://github.com/obra/superpowers/blob/main/scripts/sync-to-codex-plugin.sh
 - https://github.com/obra/superpowers/blob/main/hooks/hooks.json
 - https://github.com/obra/superpowers-marketplace/blob/main/.claude-plugin/marketplace.json
@@ -1,57 +0,0 @@
 #!/usr/bin/env bash
 set -euo pipefail
 # NOTE: This is a dev-only script, intended for use by maintainers of this repo.
 # It is not a supported installer. Modifications to it — or requests for
 # modifications — will not be approved.
 #
 # Links every skill in the repository — promoted (`skills/`) as well as
 # unpromoted (`drafts/`, `extras/`) — into the local skill directories used by
 # each agent harness:
 #   - ~/.claude/skills  — Claude Code
 #   - ~/.agents/skills  — Codex and other Agent Skills-compatible harnesses
 # Each entry is a symlink into this repo, so a `git pull` is all that's needed
 # to keep installed skills up to date.
 REPO="$(cd "$(dirname "$0")/.." && pwd)"
 DESTS=("$HOME/.claude/skills" "$HOME/.agents/skills")
 # Collect the repo's skills once, link into every destination.
 names=()
 srcs=()
 while IFS= read -r -d '' skill_md; do
  src="$(dirname "$skill_md")"
  names+=("$(basename "$src")")
  srcs+=("$src")
 done < <(find "$REPO/skills" "$REPO/drafts" "$REPO/extras" -maxdepth 2 -name SKILL.md -not -path '*/node_modules/*' -print0)
 for DEST in "${DESTS[@]}"; do
  # If $DEST is a symlink that resolves into this repo, we'd end up writing the
  # per-skill symlinks back into the repo's own skills/ tree. Detect and bail
  # out instead of polluting the working copy.
  if [ -L "$DEST" ]; then
    resolved="$(readlink -f "$DEST")"
    case "$resolved" in
      "$REPO"|"$REPO"/*)
        echo "error: $DEST is a symlink into this repo ($resolved)." >&2
        echo "Remove it (rm \"$DEST\") and re-run; the script will recreate it as a real dir." >&2
        exit 1
        ;;
    esac
  fi
  mkdir -p "$DEST"
  for i in "${!names[@]}"; do
    name="${names[$i]}"
    src="${srcs[$i]}"
    target="$DEST/$name"
    if [ -e "$target" ] && [ ! -L "$target" ]; then
      rm -rf "$target"
    fi
    ln -sfn "$src" "$target"
    echo "linked $name -> $src ($DEST)"
  done
 done
@@ -1,7 +0,0 @@
 #!/usr/bin/env bash
 set -euo pipefail
 REPO="$(cd "$(dirname "$0")/.." && pwd)"
 cd "$REPO"
 find . -name SKILL.md -not -path '*/node_modules/*' | sed 's|^\./||' | sort
@@ -1,55 +0,0 @@
 #!/usr/bin/env node
 // Copies package.json's version into every plugin manifest:
 //   plugin.json                  — the Agent Plugins 1.0 manifest
 //   .claude-plugin/plugin.json   — the Claude Code manifest
 // Runs as part of `npm run version`, immediately after `changeset version`.
 // With --check it changes nothing and exits 1 if any version differs.
 import { readFileSync, writeFileSync } from "node:fs";
 import { dirname, join, relative } from "node:path";
 import { fileURLToPath } from "node:url";
 const repo = join(dirname(fileURLToPath(import.meta.url)), "..");
 const manifests = [
  join(repo, "plugin.json"),
  join(repo, ".claude-plugin", "plugin.json"),
 ];
 const { version } = JSON.parse(readFileSync(join(repo, "package.json"), "utf8"));
 const check = process.argv.includes("--check");
 let failed = false;
 for (const path of manifests) {
  const name = relative(repo, path);
  const source = readFileSync(path, "utf8");
  const manifest = JSON.parse(source);
  if (manifest.version === version) {
    console.log(`${name} version is ${version} — already in sync`);
    continue;
  }
  if (check) {
    console.error(
      `${name} version is ${manifest.version}, package.json is ${version}. Run \`node scripts/sync-plugin-version.mjs\`.`,
    );
    failed = true;
    continue;
  }
  // Rewrite only the version line, to keep the key order and the formatting.
  const updated = source.replace(
    /("version"\s*:\s*")[^"]*(")/,
    `$1${version}$2`,
  );
  if (JSON.parse(updated).version !== version) {
    console.error(`Could not find a version field to replace in ${path}.`);
    process.exit(1);
  }
  writeFileSync(path, updated);
  console.log(`${name} version ${manifest.version} -> ${version}`);
 }
 process.exit(failed ? 1 : 0);
@@ -1,55 +0,0 @@
 # Phase boundaries
 A **phase** is a chunk of work inside a session — the grilling, the implementation, the QA. The definition is fuzzy on purpose: a phase ends when you think *"ok, we're done with that"*.
 The **phase boundary** is the gap between two phases, and it is the only place this decision belongs. Mid-phase there is no decision to make — continue, or split the work that's left into subagents. Compacting mid-phase makes the agent lose the thread.
 ## The five options
 | Option       | What it does                                                    |
 | ------------ | --------------------------------------------------------------- |
 | **Continue** | Stay in the session. No context switch at all.                    |
 | **`/clear`** | Empty the context window and start from nothing.                  |
 | **`/handoff`** | Write a portable markdown file and seed a session anywhere with it. |
 | **Subagent** | Send the task to its own context window and get a report back.     |
 | **`/compact`** | Compress this context and seed a fresh session with the summary.  |
 ## The tree
 Work top to bottom at the boundary. The first **yes** wins.
 **1. Can you continue in this session?** Two things make the answer yes: the next phase needs this phase as a **primary source**, or you have enough [smart zone](https://www.aihero.dev/ai-coding-dictionary/smart-zone) left (~150k tokens) for the next phase to fit. Grilling → implementation is the standard yes: the implementation wants the reasoning verbatim, not a summary of it. Continue costs nothing and loses nothing, so rule it out before anything else.
 **2. Is the context irrelevant to what comes next?** Is everything in this session — the exploration, the decisions, the dead ends — disposable? If so, **`/clear`**. It is the cheapest move on the board: it takes no time and hands back the whole window. `/clear` also isn't terminal — the old session stays resumable.
 The cost of getting this wrong is one-way. Clear a *relevant* context and you lose the **why** behind what you built, and no amount of reading the diff back gets it returned.
 **3. Do you need to hand off?** `/handoff` is narrow. You need it only when you are:
 - swapping to a **new harness** (Claude → Codex),
 - moving to a **new directory** or repo,
 - sending the work to a **colleague**,
 - or forking a side task you found **mid-phase** without derailing what you're doing.
 That list is the whole clause. What `/handoff` buys is **portability** — a file that travels. If nothing is travelling, you don't need it.
 **4. Can the task be done AFK?** Is it scoped tightly enough to run with you away from the keyboard, no steering? Then send it to a **subagent** and leave this session untouched. Automated review is the standard case: the agent reads the diff and reports, and you aren't needed while it does.
 **5. Otherwise, `/compact`.** Relevant context, same harness, same directory, and you need to stay in the loop — this is where the tree lands, and it lands here often. Pass it an instruction (`/compact we're going to QA this area`) so the summary keeps what the next phase needs.
 `/compact` is the **default, not the first reach**. It sits at the bottom because the four questions above it are all cheaper or more precise. The failure mode when people start here is a fresh session that is confidently wrong about a decision the summary flattened.
 ## Primary and secondary sources
 Every move except **Continue** turns a **primary source** into a **secondary source** — the session as it happened, replaced by a summary of it. The trade is always the same shape:
 | Source                            | Information | Noise | Room to move |
 | --------------------------------- | ----------- | ----- | ------------ |
 | Primary (Continue)                | Full        | Lots  | Little       |
 | Secondary (`/compact`, `/handoff`) | Lossy       | Less  | Lots         |
 This is why question 1 comes first. You only pay the lossiness when staying costs more than it saves.
 ## These are judgement calls
 The questions are not objective — each has taste in it, and the same boundary can go two ways on two days. The value is in asking them **in order**, at the boundary rather than in the middle of the work.
@@ -1,90 +0,0 @@
 ---
 name: ask-matt
 description: Ask which skill or flow fits your situation. A router over the skills in this repo.
 disable-model-invocation: true
 ---
 # Ask Matt
 You don't remember every skill, so ask.
 A **flow** is a path through the skills. Most paths run along one **main flow**, and two **on-ramps** merge onto it. Everything else is standalone, or a vocabulary layer that runs underneath.
 ## The main flow: idea → ship
 The route most work travels. You have an idea and want it built.
 1. **`/grill-with-docs`** — sharpen the idea by interview. Start here whenever you are **working in a working directory**: it's stateful, retaining what it learns in `CONTEXT.md` and ADRs. (No working directory? Use `/grill-me` — see Standalone. Both run the same `/grilling` primitive; `grill-with-docs` is the one that leaves a paper trail, which makes it the better of the two whenever a repo is there to leave it in.)
 2. **Branch — can you settle every question in conversation?** If a question needs a runnable answer (state, business logic, a UI you have to see), detour through a prototype, bridged by **`/handoff`** in both directions (a prototype lives in its own directory, which is exactly what `/handoff` is for — see Phase boundaries):
   - **`/handoff`** out, then open a fresh session against that file,
   - **`/prototype`** to answer the question with throwaway code,
   - **`/handoff`** back what you learned, and reference it from the original idea thread.
 3. **Branch — is this a multi-session build?**
   - **Yes** → **`/to-spec`** (turn the thread into a spec), then **`/to-tickets`** to split it into tracer-bullet tickets, each declaring its **blocking edges**. On a local tracker that's one file per ticket under `.scratch/<feature>/issues/`, worked blockers-first by hand; on a real tracker the edges become native blocking links, so any ticket whose blockers are done can be grabbed — kick off **`/implement`** per ticket, **`/clear`ing context between each one**. Each ticket is self-contained, so the last one's context is disposable.
   - **No** → **`/implement`** right here, in the same context window.
   Either way, **`/implement`** builds each issue by driving **`/tdd`** internally — one red-green slice at a time — then closes out by running **`/code-review`**, a two-axis review (Standards + Spec) of the diff, before committing. Reach for **`/tdd`** on its own when you just want to build a concrete behaviour test-first without a full spec, and **`/code-review`** on its own whenever you want to review a branch or PR against a fixed point.
 ### Context hygiene
 Keep steps 1–3 in **one unbroken context window** — don't compact or clear until after `/to-tickets` — so the grilling, spec, and tickets all build on the same thinking. Each `/implement` then starts fresh, working from the ticket.
 The limit on this is the **[smart zone](https://www.aihero.dev/ai-coding-dictionary/smart-zone)**: the window (~150k tokens on state-of-the-art models) within which the model still reasons sharply. If a session approaches it before `/to-tickets`, don't push on degraded — `/compact` at the nearest phase boundary and carry on (see Phase boundaries).
 ## On-ramps
 A starting situation that generates work, then merges onto the main flow.
 - **Bugs and requests piling up** → **`/triage`**. It moves issues through triage roles and produces agent-ready issues, which **`/implement`** later picks up.
  Triage is only for issues **you didn't create** — bug reports, incoming feature requests, anything that arrives raw. Tickets that `/to-tickets` produced are already agent-ready, so **don't triage them**.
 - **Something's broken** → **`/diagnosing-bugs`**. For the hard ones: the bug that resists a first glance, the intermittent flake, the regression that crept in between two known-good states. It refuses to theorise until it has a **tight feedback loop** — one command that already goes red on *this* bug — then fixes with a regression test. Its post-mortem hands off to **`/improve-codebase-architecture`** when the real finding is that there's no good seam to lock the bug down.
 - **A huge, foggy effort — a greenfield project or a huge feature build, too big for one session** → **`/wayfinder`**, the most cognitively demanding flow here. When the way from here to the destination isn't visible yet, it charts a **shared map** of **decision tickets** on the issue tracker and resolves them one at a time — producing **decisions, not deliverables** — until the fog is pushed back and the way is clear. Where **`/grill-with-docs`** sharpens an idea you can hold in one session, wayfinder is for the idea you can't — and it's slower and denser, so save it for exactly that, never a well-scoped feature.
  When the map clears, **it hands off, it doesn't build**: merge onto the main flow at **`/to-spec`**, which collapses the map's linked decisions into a buildable plan, then `/to-tickets` and `/implement` as usual. Looping the map straight into `/implement` skips that collapse and throws the linked detail away — go straight to `/implement` only when the effort turned out genuinely small.
 ## Codebase health
 Not feature work — upkeep.
 - **`/improve-codebase-architecture`** — run whenever you have a spare moment to keep the codebase good for agents to operate in. It surfaces **deepening opportunities**; picking one _generates an idea_ you can take into the main flow at `/grill-with-docs`. It's the survey that finds the candidates; **`/codebase-design`** (below) is the bench you design the chosen one on.
 ## Vocabulary underneath
 Two model-invoked references that run *beneath* the other skills — each the single source of truth for its vocabulary. Reach for them directly when the **words**, not the process, are the problem; or let the skills above pull them in.
 - **`/domain-modeling`** — sharpen the project's *domain* language: challenge a fuzzy term, resolve an overloaded word ("account" doing three jobs), record a hard-to-reverse decision as an ADR. It's the active discipline `/grill-with-docs` drives to keep `CONTEXT.md` a clean glossary.
 - **`/codebase-design`** — the deep-module vocabulary (module, interface, depth, seam, adapter, leverage, locality) for designing a module's *shape*: a lot of behaviour behind a small interface at a clean seam. `/tdd` and `/improve-codebase-architecture` both speak it.
 ## Phase boundaries
 A **phase** is a chunk of work inside a session — the grilling, the implementation, the QA. At the **boundary** between two of them you have five options, and picking between them is the fuzziest decision in this whole map:
 - **Continue** — stay put. Costs nothing, loses nothing.
 - **`/clear`** — empty the window, when nothing here matters to what's next.
 - **`/handoff`** — write a portable markdown file. Narrow: only for a **new harness**, a **new directory**, a **colleague**, or forking a side task **mid-phase**. What it buys is portability.
 - **Subagent** — send a tightly-scoped task to its own window and get a report back.
 - **`/compact`** — compress this context and seed a fresh session with it. The **default**, at the bottom of the tree rather than the first reach.
 Read [PHASE-BOUNDARIES.md](PHASE-BOUNDARIES.md) for the ordered tree — the five questions, the reasoning behind each branch, and why the primary-source cost makes **Continue** the one to rule out first. Make the decision **at** a boundary; mid-phase, continue or split the rest into subagents.
 ## Standalone
 Off the main flow entirely.
 - **`/grill-me`** — the same relentless interview as `/grill-with-docs`, but **stateless**: it saves nothing locally and builds no `CONTEXT.md`. Reach for it when you are **not working in a working directory** — sharpening a plan, a design, a piece of writing, anything with no repo under it. If you are in a working directory, use `/grill-with-docs` instead: it runs the same interview and leaves a paper trail, so it is strictly the better one.
 - **`/grilling`** — the interview primitive itself: rounds, the frontier, facts are the agent's job and decisions are yours. `/grill-me` and `/grill-with-docs` are the two named ways in, and `/triage`, `/wayfinder` and `/improve-codebase-architecture` all run it internally. Reach for it directly only when you want the interview with no wrapper around it.
 - **`/resolving-merge-conflicts`** — work an in-progress merge or rebase conflict hunk by hunk, resolving by **intent** traced to each side's primary source rather than by picking lines, then finish the operation. It never runs `--abort`. Standalone and off every flow: reach for it when you are already mid-conflict.
 - **`/prototype`** — a small, throwaway program that answers one design question: does this state model feel right, or what should this UI look like. Throwaway is a constraint on how the code is written, not a promise to destroy it: the answer folds into the real code, and the prototype itself is kept as a **primary source** on a `prototype/<name>` branch out of main, pointed at from the implementation issue. It's the detour in step 2 of the main flow, but reach for it any time a design question is hard to settle on paper.
 - **`/research`** — delegate reading legwork to a **background agent**: it investigates a question against **primary sources**, then leaves a cited Markdown file in the repo. Keep working while it reads. The file it produces is something to take *into* the main flow at `/grill-with-docs` — research feeds the thinking, it doesn't replace it.
 - **`/to-questionnaire`** — when the thing blocking you isn't in your head or the codebase but in **someone else's**, this writes them a questionnaire to fill in. It's the inverse of `/grill-me`: instead of interviewing you about the subject, it interviews you about the **send** — who it's going to, what you need back — and aims the questions at the gap. What comes back is material for `/grill-with-docs` or `/to-spec`.
 - **`/wizard`** — for the steps only a **human** can take: provisioning infrastructure, setting up credentials or CI secrets, clicking through an unfamiliar third-party dashboard, running a one-off migration or cutover. It generates an interactive bash script that opens each URL, captures each value, and writes it into `.env` and GitHub secrets — so the procedure stops being something you re-explain to an agent every time. Model-invoked, so the agent reaches for it the moment it hits a wall only you can pass. If the agent could just do it itself, it should; this is for where a human is genuinely in the loop.
 - **`/wait-what`** — the corrective for a message that didn't land. Use it mid-conversation, inside any other skill, and the agent re-pitches what it just said with the context you were missing, in plain English, using the `CONTEXT.md` vocabulary. It works after the fact; `/grill-with-docs` is the upfront cure, because a shared language agreed early is what stops the jargon arriving at all.
 - **`/teach`** — learn a concept over multiple sessions, using the current directory as a stateful workspace.
 - **`/writing-for-agents`** — reference for writing documents agents consume: skills, AGENTS.md, pointed-at docs.
 ## Precondition
 **`/setup-matt-pocock-skills`** — run before your first engineering flow to configure the issue tracker, triage labels, and doc layout the other skills assume. Custom issue trackers also work.
@@ -1,5 +0,0 @@
 interface:
  display_name: "Ask Matt"
  short_description: "Find the right skill or workflow"
 policy:
  allow_implicit_invocation: false
@@ -1,87 +0,0 @@
 ---
 name: code-review
 description: Review the changes since a fixed point (commit, branch, tag, or merge-base) along two axes — Standards (does the code follow this repo's documented coding standards?) and Spec (does the code match what the originating issue/spec asked for?). Runs both reviews in parallel sub-agents and reports them side by side. Use when the user wants to review a branch, a PR, work-in-progress changes, or asks to "review since X".
 ---
 Two-axis review of the diff between `HEAD` and a fixed point the user supplies:
 - **Standards** — does the code conform to this repo's documented coding standards?
 - **Spec** — does the code faithfully implement the originating issue / spec?
 Both axes run as **parallel sub-agents** so they don't pollute each other's context, then this skill aggregates their findings.
 The issue tracker should have been provided to you — run `/setup-matt-pocock-skills` if `docs/agents/issue-tracker.md` is missing.
 ## Process
 ### 1. Pin the fixed point
 Whatever the user said is the fixed point — a commit SHA, branch name, tag, `main`, `HEAD~5`, etc. If they didn't specify one, ask for it.
 Capture the diff command once: `git diff <fixed-point>...HEAD` (three-dot, so the comparison is against the merge-base). Also note the list of commits via `git log <fixed-point>..HEAD --oneline`.
 Before going further, confirm the fixed point resolves (`git rev-parse <fixed-point>`) and the diff is non-empty. A bad ref or empty diff should fail here — not inside two parallel sub-agents.
 ### 2. Identify the spec source
 Look for the originating spec, in this order:
 1. Issue references in the commit messages (`#123`, `Closes #45`, GitLab `!67`, etc.) — fetch via the workflow in `docs/agents/issue-tracker.md`.
 2. A path the user passed as an argument.
 3. A spec file under `docs/`, `specs/`, or `.scratch/` matching the branch name or feature.
 4. If nothing is found, ask the user where the spec is. If they say there isn't one, the **Spec** sub-agent will skip and report "no spec available".
 ### 3. Identify the standards sources
 Anything in the repo that documents how code should be written, such as `CODING_STANDARDS.md` or `CONTRIBUTING.md`.
 On top of whatever the repo documents, the Standards axis always carries the **smell baseline** below — a fixed set of Fowler code smells (_Refactoring_, ch.3) that applies even when a repo documents nothing. Two rules bind it:
 - **The repo overrides.** A documented repo standard always wins; where it endorses something the baseline would flag, suppress the smell.
 - **Always a judgement call.** Each smell is a labelled heuristic ("possible Feature Envy"), never a hard violation — and, like any standard here, skip anything tooling already enforces.
 Each smell reads *what it is* → *how to fix*; match it against the diff:
 - **Mysterious Name** — a function, variable, or type whose name doesn't reveal what it does or holds. → rename it; if no honest name comes, the design's murky.
 - **Duplicated Code** — the same logic shape appears in more than one hunk or file in the change. → extract the shared shape, call it from both.
 - **Feature Envy** — a method that reaches into another object's data more than its own. → move the method onto the data it envies.
 - **Data Clumps** — the same few fields or params keep travelling together (a type wanting to be born). → bundle them into one type, pass that.
 - **Primitive Obsession** — a primitive or string standing in for a domain concept that deserves its own type. → give the concept its own small type.
 - **Repeated Switches** — the same `switch`/`if`-cascade on the same type recurs across the change. → replace with polymorphism, or one map both sites share.
 - **Shotgun Surgery** — one logical change forces scattered edits across many files in the diff. → gather what changes together into one module.
 - **Divergent Change** — one file or module is edited for several unrelated reasons. → split so each module changes for one reason.
 - **Speculative Generality** — abstraction, parameters, or hooks added for needs the spec doesn't have. → delete it; inline back until a real need shows.
 - **Message Chains** — long `a.b().c().d()` navigation the caller shouldn't depend on. → hide the walk behind one method on the first object.
 - **Middle Man** — a class or function that mostly just delegates onward. → cut it, call the real target direct.
 - **Refused Bequest** — a subclass or implementer that ignores or overrides most of what it inherits. → drop the inheritance, use composition.
 ### 4. Spawn both sub-agents in parallel
 **Standards sub-agent prompt** — include:
 - The full diff command and commit list.
 - The list of standards-source files you found in step 3, **plus the smell baseline from step 3** pasted in full — the sub-agent has no other access to it.
 - The brief: "Report — per file/hunk where relevant — (a) every place the diff violates a documented standard: cite the standard (file + the rule); and (b) any baseline smell you spot: name it and quote the hunk. Distinguish hard violations from judgement calls — documented-standard breaches can be hard, but baseline smells are always judgement calls, and a documented repo standard overrides the baseline. Skip anything tooling enforces. Under 400 words."
 **Spec sub-agent prompt** — include:
 - The diff command and commit list.
 - The path or fetched contents of the spec.
 - The brief: "Report: (a) requirements the spec asked for that are missing or partial; (b) behaviour in the diff that wasn't asked for (scope creep); (c) requirements that look implemented but where the implementation looks wrong. Quote the spec line for each finding. Under 400 words."
 If the spec is missing, skip the Spec sub-agent and note this in the final report.
 ### 5. Aggregate
 Present the two reports under `## Standards` and `## Spec` headings, verbatim or lightly cleaned. Do **not** merge or rerank findings — the two axes are deliberately separate (see _Why two axes_).
 End with a one-line summary: total findings per axis, and the worst issue _within each axis_ (if any). Don't pick a single winner across axes — that's the reranking the separation exists to prevent.
 ## Why two axes
 A change can pass one axis and fail the other:
 - Code that follows every standard but implements the wrong thing → **Standards pass, Spec fail.**
 - Code that does exactly what the issue asked but breaks the project's conventions → **Spec pass, Standards fail.**
 Reporting them separately stops one axis from masking the other.
--- a/Show More
+++ b/Show More