claude-mdlisted

How to write and update the repository's CLAUDE.md (commands, architecture map, conventions, Definition of Done, forbidden actions) and its template. Load at kickoff, legacy assessment, when a retro adds a rule, or when CLAUDE.md is missing or its commands no longer match reality.
jaewhi-park/GARAGISTE · ★ 1 · AI & Automation · score 77

Install: claude install-skill jaewhi-park/GARAGISTE

# CLAUDE.md CLAUDE.md is injected into every session and every subagent. It must be short and true. ## Principles - Under 300 lines. Procedures go in skills; facts go here. - List only commands that were actually run and passed (confirmed by team-verifier; at kickoff the implementer's report seeds them and the verifier confirms right after; at assess each command line ends with ` — unverified` until the verifier has run it; /assess step 4 removes the marker). - Add rules only when something actually went wrong (retro output). No generalities. - Always separate quick verification (under one minute) from full verification. - "## Risk paths" is the one judgment-free input to risk, and "## UI paths" the one to the ux lens: a path match, not an opinion. Keep both short and concrete. ## Template ```` # <Project> One-line summary (details: docs/CHARTER.md · legacy: docs/ASSESSMENT.md, docs/REBUILD_PLAN.md) ## Language - <ko | en | ...> (responses, questions and documents; set with /lang; agents mirror the CEO's language when unset) ## Commands - setup: - quick verification (under one minute — relevant tests, lint, typecheck): - run one test file (the PR's Proven section names its command): - full verification: - parity harness (legacy): - build/run: - screenshots (UI stacks; renders each screen in each state to docs/screens/<plan-slug>/<screen>-<state>.png, git-ignored): ## Architecture map - <directory> — responsibility, entry points ## Conventions - naming, error handling,