Finished Audit Skill Skill

Finished Audit Skill

Purpose: Systematically verify that "finished" is actually true. No vague summaries. No optimistic assumptions. Evidence only.

When to Use

Use this skill when:

A developer or agent claims a task, feature, or project is complete
Before merging a PR or closing a milestone
Before claiming production readiness
Whenever a completion claim is made without accompanying proof

Banned Output Phrases

The following phrases indicate a false completion claim. Never output them:

| Banned Phrase | Reason | | -------------------------- | --------------- | | "Done!" | No evidence | | "That's complete." | No evidence | | "Everything is working." | No evidence | | "You're production ready." | No evidence | | "Finished." | No evidence | | "It's ready." | Subjective | | "Should be working." | Unverified | | "Looks good." | Subjective | | "Tests are passing." | No output shown |

Inputs

The task, feature, or project description
The Definition of Done criteria (from definition-of-done-builder or provided)
Any proof artifacts already provided

Procedure

Step 1: Load or generate DoD

If no DoD exists: activate definition-of-done-builder first. If DoD exists: use it as the audit checklist.

Step 2: Verify each criterion

For each criterion in the DoD:

Check if proof artifact exists
If yes: classify as PROVEN (record the artifact)
If artifact is claimed but not shown: classify as CLAIMED (not accepted)
If no artifact: classify as UNKNOWN
If definitively absent: classify as MISSING

| Status | Meaning | Accepted as proof? | | ------- | ------------------------------ | ------------------ | | PROVEN | Evidence exists, verified | Yes | | CLAIMED | Evidence claimed but not shown | No | | UNKNOWN | Not verified either way | No | | MISSING | Confirmed absent or broken | No |

Step 3: Calculate completion

Completion % = (PROVEN count / total criteria count) × 100

Step 4: Gate decision

If ALL criteria = PROVEN:
  → Output: COMPLETION APPROVED

If ANY criterion = UNKNOWN, CLAIMED, or MISSING:
  → Output: NOT COMPLETE
  → List every blocking criterion
  → Specify exact next action per blocker

Output Format

Not Complete

FINISHED AUDIT
═══════════════════════════════════════════════════
Task:        [task description]
Audit Date:  [DD/MM/YYYY]
Completion:  [N]% ([proven]/[total] criteria PROVEN)

STATUS: NOT COMPLETE

BLOCKING CRITERIA
─────────────────
[criterion] — Status: UNKNOWN
   Required: [exact action to verify]

[criterion] — Status: MISSING
   Required: [exact remediation]

[criterion] — Status: CLAIMED (evidence not shown)
   Required: [show exact output/screenshot]

NEXT ACTION
─────────────
[Single most important thing to do right now]
═══════════════════════════════════════════════════

Completion Approved

FINISHED AUDIT
═══════════════════════════════════════════════════
Task:        [task description]
Audit Date:  [DD/MM/YYYY]
Completion:  100% ([N]/[N] criteria PROVEN)

STATUS: COMPLETION APPROVED

PROOF ARTIFACTS
─────────────────
[criterion] — Proof: [artifact description + path/output]
[criterion] — Proof: [artifact]
...

APPROVED FOR: [merge/deploy/close/ship — as appropriate]
═══════════════════════════════════════════════════

Validation Gates

Before accepting a PROVEN status:

[ ] The proof artifact is shown (not just claimed)
[ ] The artifact is recent (matches the current state)
[ ] The artifact is specific (not a generic screenshot)
[ ] The artifact covers the specific criterion (not adjacent evidence)

Failure Modes

| Failure | Recovery | | ------------------------------------------ | -------------------------------------------- | | No DoD exists | Activate definition-of-done-builder first | | Proof is a screenshot of wrong state | Request updated screenshot | | Proof is from a different environment | Request production environment proof | | Developer argues criterion is "not needed" | Escalate to PM Agent — DoD is non-negotiable |

Eval Examples

Good Example

Input: "I've shipped the auth feature, it's done."

Audit output: NOT COMPLETE — 6/9 criteria PROVEN. 3 criteria UNKNOWN (no production URL proof, no rate limiting proof, JWT secret check missing).

Bad Example (rejected)

Input: "The feature is complete." Output: "Great! Looks like it's done." — REJECTED (no audit performed)

Agent Skills: Finished Audit Skill

Install this agent skill to your local

Skill Files