agentsclimarketplace

Audit

Skill erenisci/acta/skills/audit

The engineering memory for Claude Code — a CLAUDE.md brain Claude reads before it writes, plus living docs kept in sync with your code: a tiered design doc-base (tokens, components, DDRs), pricing & unit-economics, and region-aware legal.

Install
npx -y skills add erenisci/acta --skill audit

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 19 days oldThe repository was created 19 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Read-only check that the docs still match the code — drift, dead links, stale TBDs, brain/registry; reports findings (--fix: safe mechanical fixes only). Trigger on /acta:audit, "audit the docs", "check doc drift".

SKILL.md

6.9 KB, as published. Nobody here has run it

acta:audit

The trust check. Acta's whole value is a trustworthy engineering memory — this verifies the docs haven't drifted from reality. Read-only by default: it reports; it fixes only mechanical things when asked (--fix), and it never fabricates content or guesses.

Shared resources at ${CLAUDE_PLUGIN_ROOT}/acta/: doc-catalog.md (the contract), disciplines.md, project-types.md, principles.md.

Language

Write findings and any persisted items in the project's documentation language (registry language:, default English); talk to the user in the language they use. See ${CLAUDE_PLUGIN_ROOT}/acta/principles.md.

Pre-condition

Read .claude/acta.md (registry) and ${CLAUDE_PLUGIN_ROOT}/acta/doc-catalog.md. If there is no registry, this project isn't set up with Acta yet → suggest /acta:build (greenfield) or /acta:adopt (existing code). Don't guess a doc set without them.

Checks

Run these read-only checks and collect findings. Rank each 🔴 broken · 🟡 drift/stale · 🔵 hygiene.

  1. Brain & index integrity — CLAUDE.md has the <!-- acta:index:start/end --> block and its links resolve; docs/README.md lists exactly the docs on disk (no missing, no orphan); registry rows match the filesystem.
  2. Broken links — every relative link inside docs/** (and the brain) points to an existing file.
  3. Doc ↔ code drift — compare docs against the code and flag mismatches:
    • stack in arch-overview / registry vs package.json / pyproject.toml / go.mod / etc.
    • endpoints in api.md vs routes; env-var names in env-vars.md vs .env.example (the committed template only — never open a real .env/.env.local; compare names, never values); project-structure tree vs the actual top-level folders; tables in database-design vs schema/migrations.
    • "doc says X, code says Y" → report both sides.
  4. Stale / empty — docs still mostly TBD; updated: far behind recent code changes; scaffolds never filled.
  5. Contradictions — best-effort scan for conflicting statements across docs (e.g. PRD "no backend" vs architecture "REST API"). Flag them; never resolve by guessing.
  6. Conventions — filename casing (root meta UPPERCASE, docs/** lowercase kebab-case); docs living outside the catalog's paths; disciplines/packs in the registry vs what's actually on disk. Do not flag the skill-owned areas docs/design/, docs/business/, docs/legal/ as orphans — they are valid (see the catalog's "Skill-owned areas" note).
  7. Anti-bloat — in-place docs that have grown append-only (e.g. PROGRESS.md accumulating dated log lines instead of a snapshot); oversized docs that should be consolidated per their growth policy.
  8. Design consistency (only if docs/design/ exists) — the code's real styling vs the design doc-base:
    • rogue colors / spacing / fonts / type used in code but not in tokens.md (and its granular docs — color.md, typography.md, spacing-layout.md, motion.md — where present); a drifted styling approach vs design-system.md.
    • components or variants/states in code missing from components.md.
    • DDR integrity (full tier): docs/design/decisions/README.md index matches the files on disk; no DDR points to a superseding DDR that doesn't exist.
    • a broken or missing {{DESIGN_LINKS}} block in the brain (design docs exist but the "Design & brand" bullet is absent or its links don't resolve). Flag them → the fixer is /acta:track (design sync).
  9. Business consistency (only if docs/business/ exists) — a price / plan / tier in the code or config (e.g. a Stripe price, a plans table) that disagrees with pricing.md; a monetization model in code not reflected in business-model.md. Flag the drift → the fixer is /acta:business. (Read-only; never rewrite pricing.)
  10. Legal consistency (only if docs/legal/ exists) — legal-relevant reality in the code not reflected in the briefs: cookies / analytics / trackers with no consent.md entry; a data field or vendor / sub-processor absent from data-processing.md; a market/region served that the briefs' regimes don't cover; open "⚠️ needs lawyer re-review" items. Flag them → the fixer is /acta:track (legal sync). (Never write binding legal text or give legal advice.)

Output

Print a concise report — do not create a bloated audit file:

Acta audit — <project>
🔴 Broken (N)      : <one line each>
🟡 Drift/stale (N) : <one line each>
🔵 Hygiene (N)     : <one line each>
Clean              : <areas with no findings>
→ Fix: run /acta:track to re-sync current-state docs; <specific doc> needs <the manual bit only a human can supply>.

If everything passes, say so plainly. By default the audit writes nothing — the report is printed; no file is created and no file grows.

Not losing findings (opt-in, never bloats)

Because the report is ephemeral, offer two ways to keep the findings — the user chooses; nothing is written without consent:

  1. Just re-run /acta:audit next session — it's cheap and reflects the current reality, so a real issue is never missed (nothing to forget). This is the default.

  2. Record open items into the living memory (only if the user wants persistence): one concise line per unresolved finding into an existing doc — never a separate audit file:

    • blocking / do-next → docs/progress.md Blocked / Next Up
    • deferred quality or debt → docs/project/tech-debt.md (TD-N)
    • a doc's own unknown → that doc's Open Questions

    These are bounded, self-cleaning docs (a snapshot / an open-debt list) linked from the CLAUDE.md brain: acta:track removes each item once it's fixed, so they track the current open set and shrink as you resolve them — never grow without bound. Anti-bloat: one line per item, dedupe, no audit files.

Fixing — only with --fix, only the safe mechanical set

  • Regenerate the CLAUDE.md brain block, docs/README.md index, and .claude/acta.md registry from the current filesystem (idempotent, marker-scoped).
  • Remove or correct broken internal links.
  • Refresh updated: dates only on docs you actually changed.
  • NEVER fabricate content, resolve a contradiction, or fill a TBD — those are always reported for the human or for /acta:track, never invented.

Rules

  • Read-only by default; --fix performs only the mechanical set above.
  • Never fabricate; unknown / contradiction / gap → report, don't invent.
  • Idempotent. Content in the project's documentation language. Operate by ${CLAUDE_PLUGIN_ROOT}/acta/principles.md.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.