Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →An evidence gate that must pass before claiming any task is complete. Use this at the end of every coding task, right before saying "done", "fixed", "implemented", or "should work", and whenever tempted to assume code works without running it. Applies to features, bug fixes, refactors, config changes, and one-off scripts alike. If work is being handed back to a human, this gate comes first.
.claude/skills/adityaarakeri-verify-done/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -39% | 0% |
| case-18 | ✗→✓ | ▲ Improved | -33% | 0% |
| case-20 | ✗→✓ | ▲ Improved | -4% | 0% |
| case-24 | ✗→✓ | ▲ Improved | -48% | 0% |
Claim only what the available evidence demonstrates.
State the command or action, observed result, and residual gap. If a gate fails during verification-only mode, report the task as not verified and stop. If repair is in scope, fix the cause and repeat the gate before claiming completion.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | fail→pass | 3,765 | 4,987 | +32% | 1 | 1 | 0% | 532 | 669 | +26% | 0 | 0 | — |
case-01 | fail→fail | 3,949 | 4,884 | +24% | 1 | 1 | 0% | 410 | 631 | +54% | 0 | 0 | — |
case-02 | fail→fail | 7,164 | 4,866 | -32% | 1 | 1 | 0% | 262 | 582 | +122% | 0 | 0 | — |
case-03 | pass→fail | 15,382 | 4,288 | -72% | 1 | 1 | 0% | 2,250 | 628 | -72% | 0 | 0 | — |
case-04 | pass→pass | 7,241 | 4,409 | -39% | 1 | 1 | 0% | 1,332 | 1,030 | -23% | 0 | 0 | — |
case-05 | pass→pass | 9,656 | 9,597 | -1% | 1 | 1 | 0% | 1,737 | 2,049 | +18% | 0 | 0 | — |
case-06 | pass→pass | 7,005 | 3,969 | -43% | 1 | 1 | 0% | 624 | 640 | +3% | 0 | 0 | — |
case-08 | pass→pass | 7,850 | 2,712 | -65% | 1 | 1 | 0% | 1,345 | 829 | -38% | 0 | 0 | — |
case-09 | pass→pass | 8,382 | 3,388 | -60% | 1 | 1 | 0% | 1,474 | 867 | -41% | 0 | 0 | — |
case-10 | pass→pass | 11,216 | 8,633 | -23% | 1 | 1 | 0% | 1,861 | 1,374 | -26% | 0 | 0 | — |
case-11 | pass→pass | 8,936 | 6,763 | -24% | 1 | 1 | 0% | 1,735 | 1,597 | -8% | 0 | 0 | — |
case-12 | pass→fail | 13,799 | 10,778 | -22% | 1 | 1 | 0% | 2,384 | 537 | -77% | 0 | 0 | — |
case-13 | pass→pass | 13,399 | 6,096 | -55% | 1 | 1 | 0% | 2,244 | 1,259 | -44% | 0 | 0 | — |
case-14 | pass→pass | 10,987 | 7,569 | -31% | 1 | 1 | 0% | 1,860 | 1,525 | -18% | 0 | 0 | — |
case-15 | pass→pass | 9,086 | 8,639 | -5% | 1 | 1 | 0% | 1,567 | 1,825 | +16% | 0 | 0 | — |
case-16 | fail→pass | 11,019 | 5,130 | -53% | 1 | 1 | 0% | 2,049 | 1,249 | -39% | 0 | 0 | — |
case-17 | pass→pass | 6,396 | 4,267 | -33% | 1 | 1 | 0% | 1,170 | 1,023 | -13% | 0 | 0 | — |
case-18 | fail→pass | 9,403 | 4,211 | -55% | 1 | 1 | 0% | 1,547 | 1,030 | -33% | 0 | 0 | — |
case-19 | pass→pass | 7,687 | 4,424 | -42% | 1 | 1 | 0% | 1,343 | 1,075 | -20% | 0 | 0 | — |
case-20 | fail→pass | 10,577 | 8,383 | -21% | 1 | 1 | 0% | 1,858 | 1,779 | -4% | 0 | 0 | — |
case-21 | pass→pass | 9,739 | 5,086 | -48% | 1 | 1 | 0% | 1,614 | 1,126 | -30% | 0 | 0 | — |
case-22 | pass→pass | 7,636 | 4,412 | -42% | 1 | 1 | 0% | 1,297 | 1,065 | -18% | 0 | 0 | — |
case-23 | pass→pass | 6,272 | 4,192 | -33% | 1 | 1 | 0% | 1,021 | 1,032 | +1% | 0 | 0 | — |
case-24 | fail→pass | 13,011 | 4,704 | -64% | 1 | 1 | 0% | 2,097 | 1,094 | -48% | 0 | 0 | — |
case-25 | pass→pass | 9,745 | 3,757 | -61% | 1 | 1 | 0% | 1,703 | 959 | -44% | 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. 25 cases were attempted, and 21 counted toward the lift figure. The other 4 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 +12 percentage points is the difference between those two pass rates over the 21 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/3/2026 | 0% |
| gemini-3.6-flash | verified | 8/3/2026 | +14% |
Other measured skills in the registry, with their headline benchmark lift.