Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Systematic root-cause debugging. Iron Law: no fixes without root cause investigation first. Four phases: Root Cause Investigation, Pattern Analysis (6 known patterns), Hypothesis Testing (3-strike rule), Implementation. Structured debug report output.
.claude/skills/paperclipai-investigate/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 38% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 23% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -7% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -16% | 0% |
name: investigate description: > Systematic root-cause debugging. Iron Law: no fixes without root cause investigation first. Four phases: Root Cause Investigation, Pattern Analysis (6 known patterns), Hypothesis Testing (3-strike rule), Implementation. Structured debug report output. metadata: sources:
repo: garrytan/gstack path: investigate/SKILL.md commit: f4bbfaa5bdfd2d6ce59541c2145432febde57fed attribution: Garry Tan license: MIT usage: referenced
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 20,599 | 24,741 | +20% | 1 | 1 | 0% | 3,302 | 4,572 | +38% | 0 | 0 | — |
case-02 | fail→fail | 22,224 | 6,861 | -69% | 1 | 1 | 0% | 4,365 | 841 | -81% | 0 | 0 | — |
case-08 | fail→fail | 21,040 | 18,013 | -14% | 1 | 1 | 0% | 3,773 | 3,808 | +1% | 0 | 0 | — |
case-03 | fail→fail | 29,234 | 7,073 | -76% | 1 | 1 | 0% | 4,281 | 395 | -91% | 0 | 0 | — |
case-04 | pass→pass | 15,884 | 22,504 | +42% | 1 | 1 | 0% | 3,009 | 4,186 | +39% | 0 | 0 | — |
case-05 | pass→pass | 9,521 | 6,916 | -27% | 1 | 1 | 0% | 1,856 | 1,424 | -23% | 0 | 0 | — |
case-06 | pass→pass | 8,458 | 6,191 | -27% | 1 | 1 | 0% | 1,446 | 1,326 | -8% | 0 | 0 | — |
case-07 | fail→fail | 6,431 | 19,958 | +210% | 1 | 1 | 0% | 1,137 | 3,715 | +227% | 0 | 0 | — |
case-09 | fail→fail | 60,445 | 19,512 | -68% | 1 | 1 | 0% | 1,387 | 3,384 | +144% | 0 | 0 | — |
case-10 | fail→pass | 20,872 | 24,377 | +17% | 1 | 1 | 0% | 3,812 | 4,297 | +13% | 0 | 0 | — |
case-11 | pass→pass | 20,863 | 21,717 | +4% | 1 | 1 | 0% | 3,288 | 3,749 | +14% | 0 | 0 | — |
case-12 | fail→pass | 21,077 | 23,193 | +10% | 1 | 1 | 0% | 3,431 | 4,208 | +23% | 0 | 0 | — |
case-13 | fail→pass | 20,782 | 17,408 | -16% | 1 | 1 | 0% | 3,233 | 3,006 | -7% | 0 | 0 | — |
case-14 | fail→fail | 32,557 | 22,979 | -29% | 1 | 1 | 0% | 4,271 | 4,000 | -6% | 0 | 0 | — |
case-15 | fail→fail | 23,007 | 39,960 | +74% | 1 | 1 | 0% | 4,218 | 6,330 | +50% | 0 | 0 | — |
case-16 | pass→pass | 18,568 | 39,820 | +114% | 1 | 1 | 0% | 3,540 | 3,033 | -14% | 0 | 0 | — |
case-17 | fail→pass | 18,935 | 15,201 | -20% | 1 | 1 | 0% | 3,244 | 2,721 | -16% | 0 | 0 | — |
case-18 | fail→pass | 21,805 | 33,138 | +52% | 1 | 1 | 0% | 3,656 | 4,969 | +36% | 0 | 0 | — |
case-19 | fail→fail | 27,546 | 20,479 | -26% | 1 | 1 | 0% | 3,774 | 3,655 | -3% | 0 | 0 | — |
case-20 | pass→pass | 24,644 | 20,165 | -18% | 1 | 1 | 0% | 4,512 | 3,158 | -30% | 0 | 0 | — |
case-21 | fail→pass | 21,900 | 24,563 | +12% | 1 | 1 | 0% | 3,534 | 3,758 | +6% | 0 | 0 | — |
case-22 | fail→fail | 23,308 | 35,576 | +53% | 1 | 1 | 0% | 3,299 | 5,715 | +73% | 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, and 19 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 +32 percentage points is the difference between those two pass rates over the 19 comparable cases.
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.