context-auditlisted

# Context audit Measures a repo against the context-engineering standard and reports what to fix. **Report-only by default** — apply fixes only when the user explicitly asks ("audit and fix", "apply the fixes"). ## Workflow Copy this checklist and tick items off: ``` Audit progress: - [ ] 1. Inventory context files - [ ] 2. Load the checklist - [ ] 3. Evaluate every check - [ ] 4. Emit the report - [ ] 5. Fixes (ONLY if requested) ``` **1. Inventory.** Find the repo's context surface: root CLAUDE.md, AGENTS.md, any per-tool adapters (CODEX.md, GEMINI.md), nested/per-app CLAUDE.md or AGENTS.md, `.claude/skills/` and `skills/`, `docs/` tree. Note line counts. If the Context-Engineering repo is available, run its `scripts/context-lint.mjs <repo-path>` first — it settles the mechanical checks (budgets, adapters, read orders, links, command drift) so the audit can spend judgment on the rest. **2. Load** [references/checklist.md](references/checklist.md) — the full check table with pass conditions and severities. **3. Evaluate every check.** Read the actual files; never guess from names. For "commands verified", cross-check claimed commands against lockfiles and script definitions. For "dead docs", grep for references to each doc from entrypoints, indexes, and skills. **4. Emit the report** in exactly this format: ```markdown ## Context audit: <repo> Score: N/10 | Severity | File | Finding | Fix | |---|---|---|---| | high | ... | ... | ... | ``` Most severe first. Scori