repo-memory-updater

Skill: repo-memory-updater

When to Use

• After adding a new module, endpoint, or significant capability
• After adding risky code (auth, file I/O, external service)
• After any change to key commands, env vars, or test commands
• When starting a new session and AGENTS.md feels outdated
• After merging a PR that significantly changes structure

Instructions

Step 1 — Inventory what changed

git log --oneline -20                     # recent commits
git diff HEAD~5 --stat                    # files changed in last 5 commits

Look for: new files, renamed modules, new env vars in .env.example, new commands in README.

Step 2 — Check root AGENTS.md

Read AGENTS.md and answer:

• Is the Codebase Map section still accurate? (new files, changed structure)
• Are the Key Commands still correct? (nothing renamed or removed)
• Are the Coding Rules still appropriate? (no new risky patterns uncovered)
• Is the skill-to-situation mapping table still complete?
• Is the Where Deeper Truth Lives table up to date?

Step 3 — Check module AGENTS.md files

For any module that changed significantly:

• Does the module AGENTS.md reflect current invariants?
• Are the env var lists accurate?
• Are the test file references correct?

Step 4 — Update .Codex/state/

Update .Codex/state/agent-state.json:

• last_updated timestamp
• completed_steps if new milestones were reached
• changed_files list if new files were added by recent work

Append to .Codex/state/checkpoint.jsonl with a note like:

{"ts":"<ISO>","step":"repo-memory-update","status":"done","detail":"Updated AGENTS.md for <what changed>"}

Step 5 — Commit the update

git add AGENTS.md agent/AGENTS.md router/AGENTS.md .Codex/state/
git commit -m "docs: sync AGENTS.md with current repo state"

What NOT to Change

• Do not expand AGENTS.md into a full reference manual. Keep it scannable.
• Do not duplicate information already in docs/ — link to it instead.
• Do not add implementation details that belong in module docstrings.