"Re-pitch that last message" pinned the correction to one message. "Re-pitch that" leaves the agent to judge how much of what it said failed to land — and is shorter. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
852 B
mattpocock-skills
| mattpocock-skills |
|---|
| minor |
Add wait-what to in-progress/ — a fire extinguisher for model verbosity. Fire 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. 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's the extinguisher, not the sprinkler. The cure for jargon is a shared language built upfront with /grill-with-docs; this is the in-the-moment corrective for when you don't have one yet.