Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use for unattended QQ controlled frontend lockcheck runs, failure classification, bounded autofix decisions, and packaging-gate evidence in this repository. Trigger when the task is to automate full frontend batch tests, run controlled lockcheck loops, classify QQ truth-audit failures, or decide whether an autofix may continue without touching protected core rules.
.claude/skills/ethanyoq-controlled-lockcheck-autofix/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -49% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -52% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -60% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -54% | 0% |
Project-local skill for unattended QQ frontend lockcheck automation.
Read references/workflow.md first. Read references/failure-taxonomy.md when classifying a failed round. Read references/guardrails.md before proposing or applying any fix.
tree20260315 as the reference behavior for core invoice-grab rules.prepare_qq_lockcheck_run.py.postprocess_qq_lockcheck.py.process_log.txt, monitor_status.json, strict_truth_audit.json, and qq_lockcheck_report.json.classification.json, decision.json, and round_summary.md.P0=0 and user_should_review_count<=6.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 18,488 | 4,360 | -76% | 1 | 1 | 0% | 3,092 | 622 | -80% | 0 | 0 | — |
case-02 | fail→fail | 5,151 | 3,316 | -36% | 1 | 1 | 0% | 243 | 566 | +133% | 0 | 0 | — |
case-03 | fail→fail | 4,466 | 3,812 | -15% | 1 | 1 | 0% | 186 | 603 | +224% | 0 | 0 | — |
case-04 | fail→pass | 9,762 | 2,731 | -72% | 1 | 1 | 0% | 1,631 | 827 | -49% | 0 | 0 | — |
case-05 | fail→pass | 12,273 | 3,186 | -74% | 1 | 1 | 0% | 1,954 | 930 | -52% | 0 | 0 | — |
case-06 | pass→pass | 10,279 | 4,838 | -53% | 1 | 1 | 0% | 1,590 | 1,218 | -23% | 0 | 0 | — |
case-07 | pass→pass | 8,928 | 2,055 | -77% | 1 | 1 | 0% | 1,474 | 730 | -50% | 0 | 0 | — |
case-08 | pass→pass | 8,870 | 3,080 | -65% | 1 | 1 | 0% | 1,467 | 923 | -37% | 0 | 0 | — |
case-09 | fail→pass | 9,605 | 1,850 | -81% | 1 | 1 | 0% | 1,650 | 661 | -60% | 0 | 0 | — |
case-10 | fail→pass | 20,484 | 2,730 | -87% | 1 | 1 | 0% | 1,403 | 783 | -44% | 0 | 0 | — |
case-11 | fail→pass | 12,213 | 3,006 | -75% | 1 | 1 | 0% | 1,914 | 880 | -54% | 0 | 0 | — |
case-12 | fail→pass | 33,121 | 5,456 | -84% | 1 | 1 | 0% | 2,832 | 1,376 | -51% | 0 | 0 | — |
case-13 | fail→pass | 7,678 | 1,237 | -84% | 1 | 1 | 0% | 541 | 559 | +3% | 0 | 0 | — |
case-14 | fail→pass | 6,804 | 2,489 | -63% | 1 | 1 | 0% | 1,053 | 841 | -20% | 0 | 0 | — |
case-15 | fail→pass | 11,910 | 3,292 | -72% | 1 | 1 | 0% | 1,953 | 882 | -55% | 0 | 0 | — |
case-16 | fail→pass | 15,466 | 4,366 | -72% | 1 | 1 | 0% | 2,611 | 1,022 | -61% | 0 | 0 | — |
case-17 | pass→pass | 7,342 | 3,732 | -49% | 1 | 1 | 0% | 1,202 | 1,022 | -15% | 0 | 0 | — |
case-18 | fail→pass | 8,392 | 1,712 | -80% | 1 | 1 | 0% | 1,420 | 670 | -53% | 0 | 0 | — |
case-19 | fail→pass | 7,877 | 1,399 | -82% | 1 | 1 | 0% | 1,314 | 572 | -56% | 0 | 0 | — |
case-20 | pass→pass | 3,406 | 2,947 | -13% | 1 | 1 | 0% | 643 | 851 | +32% | 0 | 0 | — |
case-21 | pass→pass | 2,388 | 2,087 | -13% | 1 | 1 | 0% | 423 | 731 | +73% | 0 | 0 | — |
case-22 | pass→pass | 3,987 | 2,920 | -27% | 1 | 1 | 0% | 631 | 867 | +37% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 19 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +55 percentage points is the difference between those two pass rates over the 19 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.