Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Produce factual development-skills feedback or ingest a report to apply only evidence-backed simplifications.
.claude/skills/hashgraph-online-plugin-feedback/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | -8% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -59% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-24 | ✗→✓ | ▲ Improved | -60% | 0% |
produce writes docs/reports/development-skills-feedback-YYYY-MM-DD.md with the task context, plugin/skill actions, observed outcomes, friction, and reproducible eval ideas. Record events and evidence, not private reasoning.
ingest <report-path> treats the report as a hypothesis. Change the plugin only when an instruction is demonstrably wrong or repeatedly wasteful, and the fix is simpler than the current text.
Prefer deletion or merging. Do not add an exception for one model mistake.
Add an eval only when the Pydantic schema can observe its outcome. Tag the owning paths so normal checks select it only when relevant. Report fixes, rejected suggestions, changed files, and verification. Expect most suggestions to be rejected.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 4,145 | 10,044 | +142% | 1 | 1 | 0% | 512 | 480 | -6% | 0 | 0 | — |
case-02 | fail→fail | 5,241 | 11,301 | +116% | 1 | 1 | 0% | 198 | 563 | +184% | 0 | 0 | — |
case-03 | fail→fail | 9,678 | 16,917 | +75% | 1 | 1 | 0% | 1,172 | 534 | -54% | 0 | 0 | — |
case-04 | fail→fail | 14,962 | 7,724 | -48% | 1 | 1 | 0% | 877 | 988 | +13% | 0 | 0 | — |
case-05 | fail→pass | 5,425 | 2,932 | -46% | 1 | 1 | 0% | 677 | 624 | -8% | 0 | 0 | — |
case-06 | pass→fail | 10,366 | 10,020 | -3% | 1 | 1 | 0% | 1,709 | 977 | -43% | 0 | 0 | — |
case-19 | pass→pass | 13,987 | 8,712 | -38% | 1 | 1 | 0% | 1,474 | 834 | -43% | 0 | 0 | — |
case-07 | pass→pass | 8,705 | 8,921 | +2% | 1 | 1 | 0% | 1,410 | 772 | -45% | 0 | 0 | — |
case-08 | pass→pass | 5,701 | 11,555 | +103% | 1 | 1 | 0% | 942 | 1,339 | +42% | 0 | 0 | — |
case-09 | pass→pass | 14,346 | 8,433 | -41% | 1 | 1 | 0% | 1,222 | 696 | -43% | 0 | 0 | — |
case-10 | pass→pass | 14,449 | 3,607 | -75% | 1 | 1 | 0% | 1,598 | 737 | -54% | 0 | 0 | — |
case-11 | pass→pass | 14,104 | 3,403 | -76% | 1 | 1 | 0% | 1,476 | 723 | -51% | 0 | 0 | — |
case-12 | pass→pass | 13,818 | 7,550 | -45% | 1 | 1 | 0% | 1,379 | 595 | -57% | 0 | 0 | — |
case-13 | fail→pass | 11,365 | 8,713 | -23% | 1 | 1 | 0% | 1,773 | 722 | -59% | 0 | 0 | — |
case-14 | pass→fail | 16,454 | 9,464 | -42% | 1 | 1 | 0% | 1,876 | 1,912 | +2% | 0 | 0 | — |
case-15 | fail→pass | 11,666 | 11,344 | -3% | 1 | 1 | 0% | 2,218 | 1,286 | -42% | 0 | 0 | — |
case-16 | fail→fail | 8,837 | 8,098 | -8% | 1 | 1 | 0% | 1,490 | 750 | -50% | 0 | 0 | — |
case-17 | fail→pass | 16,769 | 9,297 | -45% | 1 | 1 | 0% | 1,806 | 1,404 | -22% | 0 | 0 | — |
case-18 | pass→pass | 13,342 | 2,883 | -78% | 1 | 1 | 0% | 2,095 | 638 | -70% | 0 | 0 | — |
case-20 | pass→fail | 11,344 | 16,713 | +47% | 1 | 1 | 0% | 1,952 | 2,057 | +5% | 0 | 0 | — |
case-21 | pass→pass | 13,034 | 9,169 | -30% | 1 | 1 | 0% | 2,184 | 1,628 | -25% | 0 | 0 | — |
case-22 | pass→pass | 10,724 | 6,866 | -36% | 1 | 1 | 0% | 1,915 | 1,296 | -32% | 0 | 0 | — |
case-23 | pass→pass | 7,094 | 8,877 | +25% | 1 | 1 | 0% | 1,194 | 807 | -32% | 0 | 0 | — |
case-24 | fail→pass | 18,163 | 2,943 | -84% | 1 | 1 | 0% | 1,780 | 704 | -60% | 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 20 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 +8 percentage points is the difference between those two pass rates over the 20 comparable cases. 4 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.