reviewing-agent-instructionslisted

# Reviewing Agent Instructions Manual on-demand review of AI-agent instruction surface in this repo. Produces a markdown report; never auto-fixes. ## When to use - The user asks to audit, review, or check the agent-instruction surface (CLAUDE.md, skills, drift-log, hooks) - The user is preparing for a cleanup pass and wants a fact-finding report first - The user runs this together with `claude-md-improver` (recommended — they're complementary) ## When NOT to use - For project content/research artifacts — those are not part of the agent-instruction surface - For ad-hoc per-file edits (use Edit directly) - As a CI check — this is manual on-demand only ## Workflow ### Step 1: Confirm scope with the user Default scope: - `CLAUDE.md` (root + any nested) - `.claude/skills/*/SKILL.md` + `references/` - `.claude/drift-log/` (open + applied) - `.claude/settings.json` + `.claude/settings.local.json` hooks - `~/.claude/` user-level skills + rules (if present) Ask: "Run with default scope, or narrow to a specific subset?" Default to full scope if user says yes. ### Step 2: Level 1 — Mechanical Run the mechanical checks using however this project runs them (a task runner, a shell script, or by hand). The checks are deterministic; the delivery mechanism varies by project. If a script or task runner is available, run the mechanical checks and write the JSON output to `.claude/audit/YYYY-MM-DD-audit-level1.json`. If no automation is present, perform the equivalent checks manuall