Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Produces a rigorous, forward-looking technical audit with severity ratings and improvement directions.
.claude/skills/tariux-package-audit/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-01 | ✗→✓ | ▲ Improved | -65% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -64% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -70% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -4% | 0% |
System Instruction: Package Audit Mode
You are a technical auditor. Your sole task is to produce a rigorous, forward-looking audit of the provided package or codebase. Do not modify any code. Do not propose concrete code changes; only flag issues, suggest improvement directions, and outline strategic possibilities. Output must be concise, factual, and stripped of all conversational filler.
Examine the package with deep technical scrutiny. Then deliver the following structured report.
Report must be under 800 words. Prioritize signal density. Use bullet points for clarity. No greetings, no closing remarks.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→pass | 24,063 | 19,190 | -20% | 1 | 1 | 0% | 2,454 | 2,680 | +9% | 0 | 0 | — |
case-01 | fail→pass | 45,467 | 36,148 | -20% | 1 | 1 | 0% | 8,289 | 2,886 | -65% | 0 | 0 | — |
case-03 | fail→pass | 39,976 | 13,050 | -67% | 1 | 1 | 0% | 6,847 | 2,488 | -64% | 0 | 0 | — |
case-04 | pass→fail | 9,731 | 11,654 | +20% | 1 | 1 | 0% | 1,773 | 2,237 | +26% | 0 | 0 | — |
case-05 | pass→fail | 18,024 | 12,079 | -33% | 1 | 1 | 0% | 3,555 | 2,117 | -40% | 0 | 0 | — |
case-06 | pass→fail | 24,609 | 10,285 | -58% | 1 | 1 | 0% | 5,217 | 1,956 | -63% | 0 | 0 | — |
case-07 | fail→pass | 43,531 | 13,513 | -69% | 1 | 1 | 0% | 8,061 | 2,439 | -70% | 0 | 0 | — |
case-08 | fail→pass | 16,652 | 12,347 | -26% | 1 | 1 | 0% | 2,381 | 2,276 | -4% | 0 | 0 | — |
case-09 | fail→pass | 32,997 | 30,548 | -7% | 1 | 1 | 0% | 4,520 | 3,197 | -29% | 0 | 0 | — |
case-10 | fail→pass | 9,379 | 22,499 | +140% | 1 | 1 | 0% | 1,146 | 3,552 | +210% | 0 | 0 | — |
case-11 | fail→pass | 24,535 | 24,519 | -0% | 1 | 1 | 0% | 4,272 | 2,498 | -42% | 0 | 0 | — |
case-12 | fail→pass | 45,065 | 15,851 | -65% | 1 | 1 | 0% | 8,241 | 2,818 | -66% | 0 | 0 | — |
case-13 | fail→pass | 20,661 | 36,468 | +77% | 1 | 1 | 0% | 2,667 | 2,486 | -7% | 0 | 0 | — |
case-14 | fail→pass | 21,824 | 13,514 | -38% | 1 | 1 | 0% | 3,371 | 2,445 | -27% | 0 | 0 | — |
case-15 | fail→fail | 33,746 | 41,691 | +24% | 1 | 1 | 0% | 5,846 | 3,123 | -47% | 0 | 0 | — |
case-16 | fail→pass | 19,094 | 19,494 | +2% | 1 | 1 | 0% | 2,413 | 2,728 | +13% | 0 | 0 | — |
case-17 | fail→fail | 24,054 | 16,675 | -31% | 1 | 1 | 0% | 4,514 | 3,103 | -31% | 0 | 0 | — |
case-18 | fail→pass | 23,244 | 54,298 | +134% | 1 | 1 | 0% | 3,168 | 2,434 | -23% | 0 | 0 | — |
case-19 | fail→pass | 46,224 | 14,039 | -70% | 1 | 1 | 0% | 8,239 | 2,543 | -69% | 0 | 0 | — |
case-20 | fail→pass | 31,476 | 10,890 | -65% | 1 | 1 | 0% | 4,554 | 2,132 | -53% | 0 | 0 | — |
case-21 | fail→pass | 24,459 | 16,807 | -31% | 1 | 1 | 0% | 4,715 | 2,888 | -39% | 0 | 0 | — |
case-22 | fail→pass | 17,787 | 12,926 | -27% | 1 | 1 | 0% | 2,908 | 2,544 | -13% | 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 +64 percentage points is the difference between those two pass rates over the 22 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.