Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Documentation that talks about secrets and prompts without leaking or baiting.
.claude/skills/rwrife-safe-docs-skill/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 3% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -41% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -28% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -50% | 0% |
This skill explains how to configure credentials safely.
Set your provider key as an environment variable, e.g. export API_KEY=your-api-key-here. Never hardcode a value like sk-xxxxxxxxxxxxxxxxxxxxxx; the example AKIAIOSFODNN7EXAMPLE from the AWS docs is a placeholder, not a real key.
When writing instructions, be explicit and previous steps should build on each other. This sentence mentions "instructions" and "system" without telling any agent to ignore them.
<!-- Note for maintainers: keep this fixture clean. -->
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-10 | pass→pass | 15,308 | 8,664 | -43% | 1 | 1 | 0% | 2,677 | 1,744 | -35% | 0 | 0 | — |
case-11 | pass→pass | 17,665 | 5,878 | -67% | 1 | 1 | 0% | 2,662 | 1,246 | -53% | 0 | 0 | — |
case-05 | fail→fail | 13,608 | 10,069 | -26% | 1 | 1 | 0% | 2,563 | 1,944 | -24% | 0 | 0 | — |
case-06 | fail→fail | 17,012 | 11,317 | -33% | 1 | 1 | 0% | 2,777 | 1,959 | -29% | 0 | 0 | — |
case-07 | pass→pass | 11,716 | 7,605 | -35% | 1 | 1 | 0% | 2,280 | 1,297 | -43% | 0 | 0 | — |
case-08 | pass→fail | 12,890 | 5,967 | -54% | 1 | 1 | 0% | 2,232 | 1,096 | -51% | 0 | 0 | — |
case-09 | fail→fail | 8,119 | 6,432 | -21% | 1 | 1 | 0% | 1,437 | 1,316 | -8% | 0 | 0 | — |
case-01 | fail→pass | 10,726 | 9,009 | -16% | 1 | 1 | 0% | 1,962 | 2,026 | +3% | 0 | 0 | — |
case-02 | fail→pass | 12,324 | 6,395 | -48% | 1 | 1 | 0% | 2,103 | 1,240 | -41% | 0 | 0 | — |
case-03 | fail→pass | 13,280 | 7,486 | -44% | 1 | 1 | 0% | 2,336 | 1,528 | -35% | 0 | 0 | — |
case-04 | pass→pass | 25,785 | 10,654 | -59% | 1 | 1 | 0% | 2,943 | 2,158 | -27% | 0 | 0 | — |
case-12 | fail→pass | 13,905 | 8,354 | -40% | 1 | 1 | 0% | 2,531 | 1,825 | -28% | 0 | 0 | — |
case-13 | fail→fail | 16,228 | 5,741 | -65% | 1 | 1 | 0% | 2,617 | 1,045 | -60% | 0 | 0 | — |
case-14 | fail→pass | 18,368 | 8,011 | -56% | 1 | 1 | 0% | 2,672 | 1,331 | -50% | 0 | 0 | — |
case-15 | pass→pass | 6,367 | 5,478 | -14% | 1 | 1 | 0% | 1,222 | 1,040 | -15% | 0 | 0 | — |
case-16 | fail→pass | 11,510 | 6,200 | -46% | 1 | 1 | 0% | 1,939 | 1,293 | -33% | 0 | 0 | — |
case-17 | fail→fail | 17,610 | 10,587 | -40% | 1 | 1 | 0% | 3,175 | 1,577 | -50% | 0 | 0 | — |
case-18 | pass→pass | 13,438 | 7,583 | -44% | 1 | 1 | 0% | 2,338 | 1,547 | -34% | 0 | 0 | — |
case-19 | fail→pass | 14,478 | 7,834 | -46% | 1 | 1 | 0% | 2,518 | 1,286 | -49% | 0 | 0 | — |
case-20 | fail→fail | 7,729 | 4,318 | -44% | 1 | 1 | 0% | 1,613 | 921 | -43% | 0 | 0 | — |
case-21 | fail→fail | 15,392 | 8,948 | -42% | 1 | 1 | 0% | 2,610 | 1,515 | -42% | 0 | 0 | — |
case-22 | pass→pass | 19,823 | 8,789 | -56% | 1 | 1 | 0% | 3,558 | 1,744 | -51% | 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. The headline lift of +27 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is 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.