Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Verify changes or release readiness for this platform repository. Use when asked to test, validate, run checks, diagnose verification failures, or prove that an implementation works across packages, Docker configuration, HTTP health, or browser behavior.
.claude/skills/qierkang-project-verification/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-17 | ✗→✓ | ▲ Improved | -69% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -36% | 0% |
| case-07 | ✓→✗ | ▼ Worse | 6% | 0% |
| case-20 | ✓→✗ | ▼ Worse | -70% | 0% |
| case-22 | ✓→✗ | ▼ Worse | -83% | 0% |
Verification must match the changed surface and produce real evidence.
package.json files; never assume scripts exist.references/verification-matrix.md.standalone-infra Docker Compose configurations when deployment files change.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 5,042 | 15,565 | +209% | 1 | 1 | 0% | 289 | 615 | +113% | 0 | 0 | — |
case-02 | fail→fail | 8,085 | 4,413 | -45% | 1 | 1 | 0% | 1,296 | 443 | -66% | 0 | 0 | — |
case-03 | fail→fail | 4,308 | 4,222 | -2% | 1 | 1 | 0% | 171 | 427 | +150% | 0 | 0 | — |
case-04 | fail→fail | 6,878 | 8,771 | +28% | 1 | 1 | 0% | 711 | 760 | +7% | 0 | 0 | — |
case-17 | fail→pass | 13,779 | 2,821 | -80% | 1 | 1 | 0% | 2,088 | 650 | -69% | 0 | 0 | — |
case-05 | pass→pass | 5,485 | 2,927 | -47% | 1 | 1 | 0% | 855 | 763 | -11% | 0 | 0 | — |
case-06 | pass→pass | 11,181 | 4,873 | -56% | 1 | 1 | 0% | 1,889 | 1,066 | -44% | 0 | 0 | — |
case-07 | pass→fail | 9,155 | 6,763 | -26% | 1 | 1 | 0% | 1,404 | 1,488 | +6% | 0 | 0 | — |
case-08 | pass→pass | 10,274 | 5,616 | -45% | 1 | 1 | 0% | 1,662 | 1,247 | -25% | 0 | 0 | — |
case-09 | pass→pass | 10,898 | 5,693 | -48% | 1 | 1 | 0% | 1,805 | 1,167 | -35% | 0 | 0 | — |
case-10 | fail→pass | 9,708 | 5,083 | -48% | 1 | 1 | 0% | 1,691 | 1,080 | -36% | 0 | 0 | — |
case-11 | pass→pass | 8,943 | 4,964 | -44% | 1 | 1 | 0% | 1,514 | 1,063 | -30% | 0 | 0 | — |
case-12 | pass→pass | 10,051 | 6,691 | -33% | 1 | 1 | 0% | 1,681 | 1,324 | -21% | 0 | 0 | — |
case-13 | pass→pass | 13,494 | 10,997 | -19% | 1 | 1 | 0% | 2,039 | 1,508 | -26% | 0 | 0 | — |
case-14 | pass→pass | 9,710 | 5,646 | -42% | 1 | 1 | 0% | 1,598 | 1,187 | -26% | 0 | 0 | — |
case-15 | pass→pass | 8,668 | 4,509 | -48% | 1 | 1 | 0% | 1,390 | 981 | -29% | 0 | 0 | — |
case-16 | pass→pass | 10,181 | 4,089 | -60% | 1 | 1 | 0% | 1,532 | 905 | -41% | 0 | 0 | — |
case-18 | pass→pass | 14,860 | 3,002 | -80% | 1 | 1 | 0% | 2,491 | 702 | -72% | 0 | 0 | — |
case-19 | pass→pass | 9,851 | 4,895 | -50% | 1 | 1 | 0% | 1,468 | 933 | -36% | 0 | 0 | — |
case-20 | pass→fail | 9,439 | 5,123 | -46% | 1 | 1 | 0% | 1,534 | 455 | -70% | 0 | 0 | — |
case-21 | pass→pass | 9,150 | 7,965 | -13% | 1 | 1 | 0% | 1,440 | 1,460 | +1% | 0 | 0 | — |
case-22 | pass→fail | 13,489 | 12,581 | -7% | 1 | 1 | 0% | 2,696 | 451 | -83% | 0 | 0 | — |
case-23 | pass→fail | 11,146 | 4,385 | -61% | 1 | 1 | 0% | 2,018 | 444 | -78% | 0 | 0 | — |
case-24 | pass→fail | 6,603 | 3,977 | -40% | 1 | 1 | 0% | 1,277 | 441 | -65% | 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. 24 cases were attempted, and 16 counted toward the lift figure. The other 8 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 16 comparable cases. 5 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.
Other measured skills in the registry, with their headline benchmark lift.