Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Systematic debugging with Iron Law methodology. 5-phase investigation from evidence collection to verified fix. Triggers on "investigate", "debug", "root cause".
.claude/skills/investigate/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-18 | ✗→✓ | ▲ Improved | — | — |
| case-07 | ✗→✓ | ▲ Improved | — | — |
| case-10 | ✗→✓ | ▲ Improved | — | — |
| case-11 | ✗→✓ | ▲ Improved | — | — |
| case-21 | ✗→✓ | ▲ Improved | — | — |
Systematic debugging skill that enforces the Iron Law: never fix without a confirmed root cause. Produces a structured debug report with full evidence chain, minimal fix, and regression test.
No fix without confirmed root cause. Every investigation follows a strict evidence chain:
Violation of the Iron Law (skipping to Phase 4 without Phase 3 confirmation) is prohibited.
Phase 1: Root Cause Investigation
Reproduce bug, collect evidence (errors, logs, traces)
Use ccw cli --tool gemini --mode analysis for initial diagnosis
Output: investigation-report.json
|
v
Phase 2: Pattern Analysis
Search codebase for similar patterns (same error, module, antipattern)
Assess scope: isolated vs systemic
Output: pattern-analysis section in report
|
v
Phase 3: Hypothesis Testing
Form max 3 hypotheses from evidence
Test each with minimal read-only probes
3-strike rule: STOP and escalate on 3 consecutive failures
Output: confirmed root cause with evidence chain
|
v
Phase 4: Implementation [GATE: requires Phase 3 confirmed root cause]
Implement minimal fix
Add regression test
Verify fix resolves reproduction case
|
v
Phase 5: Verification & Report
Run full test suite
Check for regressions
Generate structured debug report to .workflow/.debug/bashmkdir -p .workflow/.debug
.workflow/.debug/
debug-report-{YYYY-MM-DD}-{slug}.json # Structured debug reportThis skill follows the Completion Status Protocol defined in _shared/SKILL-DESIGN-SPEC.md sections 13-14.
| Status | When | |--------|------| | DONE | Root cause confirmed, fix applied, regression test passes, no regressions | | DONE_WITH_CONCERNS | Fix applied but partial test coverage or minor warnings | | BLOCKED | Cannot reproduce bug, or 3-strike escalation triggered in Phase 3 | | NEEDS_CONTEXT | Missing reproduction steps, unclear error conditions |
| Document | Purpose | |----------|---------| | phases/01-root-cause-investigation.md | Evidence collection and reproduction | | phases/02-pattern-analysis.md | Codebase pattern search and scope assessment | | phases/03-hypothesis-testing.md | Hypothesis formation, testing, and 3-strike rule | | phases/04-implementation.md | Minimal fix with Iron Law gate | | phases/05-verification-report.md | Test suite, regression check, report generation | | specs/iron-law.md | Iron Law rules definition | | specs/debug-report-format.md | Structured debug report JSON schema |
This skill leverages ccw cli for multi-model analysis at key points:
| Phase | CLI Usage | Mode | |-------|-----------|------| | Phase 1 | Initial diagnosis from error evidence | --mode analysis | | Phase 2 | Cross-file pattern search | --mode analysis | | Phase 3 | Hypothesis validation assistance | --mode analysis |
All CLI calls use --mode analysis (read-only). No write-mode CLI calls during investigation phases 1-3.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-18 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-07 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-05 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-10 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-06 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-01 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-11 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-02 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-21 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-09 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-03 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-17 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-14 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-15 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-04 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-16 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-19 | pass→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-22 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-20 | pass→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-08 | pass→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-12 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-13 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
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 17 counted toward the lift figure. The other 5 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 +50 percentage points is the difference between those two pass rates over the 17 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
The per-case answers from this run were removed by the retention sweep, so the case table below shows the verdicts without the text either arm produced. The counts above were recorded at the time and are unaffected. Answers are now kept for 180 days.
Other measured skills in the registry, with their headline benchmark lift.