Skip to content

/ardd-audit

Tier: extension

Challenge artifact decisions — simplicity, failure modes, robustness, semantics — and write the findings checklist to .project/audit.md. Takes no proposal input — vet new ideas with /ardd-research instead.

Usage

/ardd-audit             # audit all artifacts
/ardd-audit <name>      # audit one artifact

The only legal argument is the name of an existing artifact. A proposal, an idea, a "what if we…" is redirected to /ardd-research (whose proposal-vetting mode applies these same lenses to a hypothetical change) — this skill audits only decisions already recorded. Unlike /ardd-status (consistency and completeness), it asks whether the decisions themselves are good.

Reads

  • .project/artifacts/*.md (scoped or all) and the feature register
  • The existing .project/audit.md, if any — open items are reported and you choose: keep working the checklist, refresh one artifact's section, or regenerate the full report. Artifacts updated since the last audit are flagged as potentially stale findings.

The lenses

Every artifact is worked through seven critical lenses: simplicity (what could be removed?), failure modes (what breaks off the happy path?), standardness (are conventions being reinvented?), robustness/fragility (what couples what?), DRYness (what goes out of sync first?), semantics (do names mean what they say?), and proportionality (over-engineered for scope? under-specified where mistakes are expensive?).

Writes

  • .project/audit.md — its single writer. Findings are a working checklist, classified:
  • [S] Suggestion — a concrete change, always with a runnable /ardd-refine <artifact> <tight directive> (or /ardd-backlog) command
  • [Q] Question — needs your input; the trade-off stated
  • [R] Risk — a fragility worth acknowledging even without a change

Resolution workflow

You work the checklist directly in the file: [ ] open, [x] resolved, [-] rejected/deferred (optionally with a note). The next /ardd-audit run reads those marks and reports accordingly. Findings are deliberately not softened — you can reject a wrong finding; you can't act on one never raised.

Behavior notes

  • Adoption: an install predating the v1.0.0 rename may still have the legacy critique.md file; the skill renames it to audit.md and continues.
  • The report is one sentence — the file is the deliverable.
  • /ardd-research — vets proposed changes with the same lenses
  • /ardd-refine — where accepted suggestions get applied
  • guides/checking.md — the four checking skills compared