thkt use-context-root-cause-analysis
Principle
Fix the root cause, not the symptom. Symptom fixes add complexity; root-cause fixes prevent recurrence.
Method
${CLAUDE_SKILL_DIR}/../../rules/core/OPERATION.md § Debug Investigation Protocol is canonical. A forked run does not receive the always-loaded rules, so the steps are copied here.
- Diff working similar code against the broken code and list the differences
- Raise three or more hypotheses for the cause. Candidates come from ${CLAUDE_SKILL_DIR}/references/symptom-patterns.md
- Eliminate each by testing. Reach no conclusion while more than one survives
- The surviving hypothesis is the root cause. Check it with "does fixing this make the symptom go away?"
| Pitfall | Treatment |
|---|---|
| Stopping at the first hypothesis | Start no testing until three are on the table |
| Dropping one without a test | Drop on a run result or on evidence, never on plausibility |
| Drifting into the abstract | Stop at the height where an action exists. Do not reach for design |
| Every hypothesis survives | Sharpen the diff and separate the differences one at a time |
Deciding the Pattern
Pattern turns on whether a recurrence path exists, not on how deep the cause sits. Sweep for code shaped like the root cause and classify.
| Value | Test |
|---|---|
| Isolated | The shape appears nowhere else |
| Recurring | The same shape sits nearby |
| Systematic | It comes from the design and the shape crosses layers |
Output Format
Callers branch on Pattern. /fix decides between applying defense-in-depth and delegating to /research.
| Field | Description |
|---|---|
| Symptom | The failure as the user sees it |
| Root cause | The hypothesis testing left |
| Pattern | Isolated / Recurring / Systematic |
References
| What you are unsure of | File |
|---|---|
| How to raise the hypotheses | ${CLAUDE_SKILL_DIR}/references/symptom-patterns.md |
| How the elimination runs | ${CLAUDE_SKILL_DIR}/references/hypothesis-examples.md |