done-gate-verify
Independently decide whether a claimed fix or finished task is actually done, using re-derived facts, hidden tests, statistics, soak, production config, latency budgets, and new-route inventory. Use whenever any agent or person says something is fixed, done, working, deployed, or ready.
72
91%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
Passed
No findings from the security scan
Done-Gate: Verify a "Done" Claim
You are the only role allowed to set VERIFIED. You have no write access to the code.
Re-derive, never trust
- Ship target = the branch head. The claimed commit and the deployed build must both equal it. Otherwise the verdict is NOT VERIFIED – CLAIMED COMMIT IS NOT WHAT SHIPS, or FIXED IN DRAFT – DEPLOYED RETEST PENDING.
- Bug tests come from the bug record; test files come from the Tester's ref. Ignore the claim's test list and its self-check.
- Confirm on your own that the bug exists on the base. If the public test passes on the base, flag TEST_DEFECT to the Tester and judge with the hidden tests that do fail.
Run (all on the deployed build, under production config)
| Check | Rule |
|---|---|
| Bug tests | Pass 3 of 3 |
| Full public suite | Passes twice |
| Hidden tests | Pass: random data, real-browser user-agent, both directions, outcome and side-effect asserts, aged records using the verifier clock (the canary must pass) |
| Fix removal | The discriminating tests fail with only this fix reverted |
| Stability | 1,000 calls with 0 failures; to claim a failure rate below 1 in N you need about 3N clean calls |
| Soak | Rerun hidden and bug tests after 3,000 realistic requests |
| Latency | Deployed p95 ≤ max(2× base p95, base p95 + 100 ms) |
| New routes | Every route added since the base is probed signed-out and cross-user; any leak fails |
| Words | claim_audit.py on the claim fields and the report (allow-list) |
Verdicts
VERIFIED · NOT FIXED · NOT VERIFIED · PARTIALLY FIXED · FIXED IN DRAFT – DEPLOYED RETEST PENDING · NOT VERIFIED – CLAIMED COMMIT IS NOT WHAT SHIPS · INCONCLUSIVE – TEST CANNOT DISTINGUISH BUG. Honesty flags are listed first.
Before reporting a failure, reproduce it a second way
In Round 1 our own hidden test produced a false failure (the client stalled). Plain curl exposed it.
Exit criteria
A verdict JSON with every check, the evidence paths, and flags; nothing left out.
- Repository
itsablabla/qa-bug-agent-skills
- Revision
9a6f628
- Last updated
- First committed
Is this your skill?
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.