Claude MD Optimizer
Verkleinert überdimensionierte CLAUDE.md / AGENTS.md / copilot-instructions mittels progressive disclosure
Getestet · Funktioniert
Was es kann
Detects und slims down bloated agent instruction files (CLAUDE.md, AGENTS.md, copilot-instructions.md) by extracting reference content verbatim into the format's native sub-doc mechanism und replacing it with a rich inline synopsis plus link. Wird bei requests to optimize/reduce/shrink an over-long agent instructions oder context file ausgelöst.
Testbericht
Führte es against the repo's own 270-line CLAUDE.md fixture aus: the skill produced a 138-line main file (under its 200 target) plus 171 lines of verbatim .claude/rules sub-docs, und all eight probed detail strings survived, from the compose password to the deploy poll interval to the AWS/SendGrid key placeholders. A no-skill baseline pass hit 66 lines but only by summarizing, silently dropping those same specifics (POSTGRES_PASSWORD, wait-on health URL, AWS_SECRET_ACCESS_KEY, SendGrid key). The skill's zero-information-loss claim ist the real, measurable edge, und it ships a genuine eval harness with fixtures und assertions rather than just prose.
Getestet am: 2026-07-20 · Claude Code 2.x (agent harness)
Installation
git clone https://github.com/wrsmith108/claude-md-optimizer mkdir -p ~/.claude/skills cd claude-md-optimizer && cp -r . ~/.claude/skills/claude-md-optimizer
Befehle & Beispiel-Prompts
/claude-md-optimizerVerkleinert überdimensionierte CLAUDE.md / AGENTS.md / copilot-instructions mittels progressive disclosure
Skills reagieren auf normale Anfragen — keine Slash-Befehle nötig. Nach der Installation aktivieren Prompts wie diese den Skill (auf Englisch):
Our CLAUDE.md file is way too long, shrink it without losing infoSplit this bloated AGENTS.md into a synopsis plus linked sub-docsOptimize our copilot-instructions.md using progressive disclosure