Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Verify that a change really works before claiming completion. Use when the user wants confidence that a feature, fix, or refactor actually works — turns vague "it should work" claims into concrete evidence. Pairs with @oath-verifier agent.
.claude/skills/evolution-foundation-dev-verify/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -19% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -19% | 0% |
| case-04 | ✓→✗ | ▼ Worse | -80% | 0% |
| case-05 | ✓→✗ | ▼ Worse | -90% | 0% |
| case-17 | ✓→✗ | ▼ Worse | -45% | 0% |
Derived from oh-my-claudecode (MIT, Yeachan Heo). Adapted for the EvoNexus Engineering Layer.
Use this skill when the user wants confidence that a feature, fix, or refactor actually works.
Turn vague "it should work" claims into concrete evidence.
For deeper verification with formal acceptance criteria mapping, delegate to @oath-verifier.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 4,202 | 3,554 | -15% | 1 | 1 | 0% | 195 | 456 | +134% | 0 | 0 | — |
case-02 | fail→fail | 9,897 | 3,393 | -66% | 1 | 1 | 0% | 1,638 | 621 | -62% | 0 | 0 | — |
case-03 | fail→fail | 18,071 | 4,762 | -74% | 1 | 1 | 0% | 3,115 | 545 | -83% | 0 | 0 | — |
case-04 | pass→fail | 14,703 | 3,220 | -78% | 1 | 1 | 0% | 2,828 | 568 | -80% | 0 | 0 | — |
case-05 | pass→fail | 22,449 | 3,470 | -85% | 1 | 1 | 0% | 4,526 | 432 | -90% | 0 | 0 | — |
case-06 | fail→pass | 10,125 | 6,132 | -39% | 1 | 1 | 0% | 1,593 | 1,286 | -19% | 0 | 0 | — |
case-07 | pass→pass | 5,137 | 4,308 | -16% | 1 | 1 | 0% | 834 | 977 | +17% | 0 | 0 | — |
case-08 | pass→pass | 9,423 | 5,840 | -38% | 1 | 1 | 0% | 1,513 | 1,124 | -26% | 0 | 0 | — |
case-09 | pass→pass | 7,945 | 2,646 | -67% | 1 | 1 | 0% | 1,260 | 703 | -44% | 0 | 0 | — |
case-10 | pass→pass | 12,539 | 6,778 | -46% | 1 | 1 | 0% | 1,981 | 1,339 | -32% | 0 | 0 | — |
case-11 | fail→fail | 15,848 | 10,697 | -33% | 1 | 1 | 0% | 2,552 | 1,824 | -29% | 0 | 0 | — |
case-12 | pass→pass | 9,148 | 5,245 | -43% | 1 | 1 | 0% | 1,471 | 1,119 | -24% | 0 | 0 | — |
case-13 | fail→pass | 7,666 | 4,451 | -42% | 1 | 1 | 0% | 1,156 | 942 | -19% | 0 | 0 | — |
case-14 | pass→pass | 11,901 | 6,144 | -48% | 1 | 1 | 0% | 1,825 | 1,311 | -28% | 0 | 0 | — |
case-15 | pass→pass | 9,789 | 6,030 | -38% | 1 | 1 | 0% | 1,502 | 1,262 | -16% | 0 | 0 | — |
case-16 | pass→pass | 10,072 | 5,449 | -46% | 1 | 1 | 0% | 1,640 | 1,123 | -32% | 0 | 0 | — |
case-17 | pass→fail | 25,278 | 5,105 | -80% | 1 | 1 | 0% | 2,109 | 1,152 | -45% | 0 | 0 | — |
case-18 | fail→fail | 9,234 | 3,117 | -66% | 1 | 1 | 0% | 1,413 | 525 | -63% | 0 | 0 | — |
case-19 | pass→pass | 6,769 | 3,525 | -48% | 1 | 1 | 0% | 1,070 | 798 | -25% | 0 | 0 | — |
case-20 | pass→pass | 8,195 | 3,368 | -59% | 1 | 1 | 0% | 1,234 | 802 | -35% | 0 | 0 | — |
case-21 | pass→pass | 5,755 | 2,633 | -54% | 1 | 1 | 0% | 851 | 718 | -16% | 0 | 0 | — |
case-22 | pass→pass | 7,569 | 4,832 | -36% | 1 | 1 | 0% | 1,198 | 1,111 | -7% | 0 | 0 | — |
case-23 | pass→pass | 11,425 | 6,740 | -41% | 1 | 1 | 0% | 2,016 | 1,395 | -31% | 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. 23 cases were attempted, and 20 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 -4 percentage points is the difference between those two pass rates over the 20 comparable cases. 3 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.