Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Diagnose software failures through reproduction, boundary localization, competing hypotheses, and discriminating experiments. Use when Codex is asked to investigate a bug, flaky test, crash, incorrect result, performance regression, production symptom, or unexplained behavior and should determine the cause before implementing a fix. Produce an evidence-backed causal explanation and verification plan; do not patch unless the user also asks for a fix.
.claude/skills/contextosai-diagnose-root-cause/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 92% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 14% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 46% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 16% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 81% | 0% |
Find the earliest incorrect transition that explains the observed symptom.
versions, timestamps, complete error, and frequency. Redact secrets.
unavailable, use existing logs/tests and label the conclusion accordingly.
primary symptom from secondary errors produced during recovery or cleanup.
invariant and the observed value/state.
contradicting evidence, and one experiment whose outcomes distinguish it from the others.
temporary diagnostic over production mutation. Change one variable at a time.
divergence to the user-visible symptom.
blast radius and regression origin, not to replace causal evidence.
the fix and passes after it. Do not implement unless requested.
root cause without a mechanism.
turn green.
the exact observation needed. Use confidence calibrated to evidence.
and resource exhaustion explicitly.
optimizing code that merely appears hot.
Use references/diagnosis-report.md. Lead with the proven or most likely cause, then the causal chain and decisive evidence. Keep exploration history only when it helps another engineer verify the conclusion.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→fail | 18,468 | 3,922 | -79% | 1 | 1 | 0% | 2,985 | 802 | -73% | 0 | 0 | — |
case-01 | fail→fail | 25,111 | 26,911 | +7% | 1 | 1 | 0% | 4,495 | 5,453 | +21% | 0 | 0 | — |
case-03 | fail→pass | 17,268 | 32,257 | +87% | 1 | 1 | 0% | 3,492 | 6,707 | +92% | 0 | 0 | — |
case-04 | fail→fail | 15,563 | 19,065 | +23% | 1 | 1 | 0% | 3,114 | 4,360 | +40% | 0 | 0 | — |
case-05 | fail→pass | 16,979 | 18,114 | +7% | 1 | 1 | 0% | 3,099 | 3,546 | +14% | 0 | 0 | — |
case-06 | pass→pass | 14,758 | 25,459 | +73% | 1 | 1 | 0% | 2,496 | 5,165 | +107% | 0 | 0 | — |
case-07 | fail→fail | 24,633 | 22,950 | -7% | 1 | 1 | 0% | 4,343 | 4,839 | +11% | 0 | 0 | — |
case-08 | fail→pass | 18,916 | 25,445 | +35% | 1 | 1 | 0% | 3,577 | 5,216 | +46% | 0 | 0 | — |
case-09 | fail→fail | 16,166 | 18,886 | +17% | 1 | 1 | 0% | 2,742 | 3,935 | +44% | 0 | 0 | — |
case-10 | pass→pass | 6,521 | 8,729 | +34% | 1 | 1 | 0% | 1,237 | 2,150 | +74% | 0 | 0 | — |
case-11 | pass→pass | 18,627 | 21,094 | +13% | 1 | 1 | 0% | 3,794 | 5,083 | +34% | 0 | 0 | — |
case-12 | fail→fail | 2,162 | 10,188 | +371% | 1 | 1 | 0% | 369 | 2,404 | +551% | 0 | 0 | — |
case-13 | pass→pass | 17,086 | 28,504 | +67% | 1 | 1 | 0% | 3,051 | 5,655 | +85% | 0 | 0 | — |
case-19 | fail→pass | 17,167 | 17,458 | +2% | 1 | 1 | 0% | 3,264 | 3,791 | +16% | 0 | 0 | — |
case-14 | fail→fail | 17,962 | 3,955 | -78% | 1 | 1 | 0% | 3,218 | 666 | -79% | 0 | 0 | — |
case-15 | pass→pass | 23,485 | 18,726 | -20% | 1 | 1 | 0% | 4,405 | 3,642 | -17% | 0 | 0 | — |
case-16 | pass→pass | 20,832 | 23,119 | +11% | 1 | 1 | 0% | 3,742 | 4,440 | +19% | 0 | 0 | — |
case-17 | pass→fail | 16,779 | 22,771 | +36% | 1 | 1 | 0% | 3,352 | 5,154 | +54% | 0 | 0 | — |
case-18 | fail→pass | 16,308 | 27,108 | +66% | 1 | 1 | 0% | 3,096 | 5,593 | +81% | 0 | 0 | — |
case-20 | fail→fail | 3,826 | 3,395 | -11% | 1 | 1 | 0% | 163 | 656 | +302% | 0 | 0 | — |
case-21 | fail→pass | 18,615 | 22,865 | +23% | 1 | 1 | 0% | 3,448 | 4,989 | +45% | 0 | 0 | — |
case-22 | pass→pass | 13,415 | 19,685 | +47% | 1 | 1 | 0% | 2,521 | 4,192 | +66% | 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 20 counted toward the lift figure. The other 2 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 +23 percentage points is the difference between those two pass rates over the 20 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.