debug

<skill id="debug"> <purpose> Debugging is forming and testing a THEORY that explains the bug. Not random changes. Not guessing. Scientific method applied to code. DEFECT (in code) → INFECTION (in state) → FAILURE (visible symptom). The failure you see is NOT where the bug is. Binary search upstream. Systematic methodology beats ad-hoc guessing. The process is the multiplier.

Diagnosis comes before prescription: a surgeon cutting before the X-ray is guessing with a knife. Two modes, same discipline. bug (default) is the steps below. refactor swaps the subject from a failure to a structure, and produces a plan instead of a fix. </purpose>

<prerequisite>Run agentdb recall with the exact error text, subsystem/library, failing test, and known files/symbols. Recall again when the hypothesis changes or a new failure appears; that is a new retrieval question. Reference on demand: skills/debug/reference/debug-research.md.</prerequisite>

<steps> 1. **REPRODUCE**: get specific before touching code. - Document: exact input, expected output, actual output (full stack trace), environment, frequency. - "Sometimes fails" is not a reproduction. Get deterministic. - Sensitive or high-blast-radius bug: investigate read-only (plan mode) and settle the approach before any edit. <!-- Updated 2026-09-20: sitepoint.com / claudelog.com Claude Code debugging guides --> - (gate: can reproduce consistently, OR have added targeted logging to wait for next occurrence)

2. HYPOTHESIZE: list 3 causes before pursuing any.

   • Read ALL error output first (anchoring bias mitigation).
   • Write each hypothesis to AgentDB. Prevents circular re-investigation.
   • (gate: 3 candidate hypotheses written; none pursued yet)

3. ISOLATE: binary search, O(log n) not O(n).

   • Code: call chain A→B→C→D→E fails → check midpoint C → recurse into failing half.
   • Time: git bisect between known-good and known-bad commit. ~10 tests for 1000 commits.
   • Input: large failing input → split in half → recurse to minimal reproduction case.
   • Instrument at boundaries: log inputs/outputs at each layer boundary.
   • Mock external dependencies to isolate which one causes failure.
   <!-- Updated 2026-09-19: dev.to flaky-test patterns; claudelog/sfeir debugging guides -->

   • Diff working vs failing state (env, config, input, commit) before theorizing.

   • Flaky test: classify first as timing/race, shared state, unseeded randomness, or environment difference; only then propose a fix.

   • (gate: failure localized to a specific function/commit/input subset)

4. ROOT CAUSE: the error line is the FAILURE. The DEFECT is upstream.

   • Ask: what assumption was violated? What invariant broke?
   • If you can't explain WHY it broke, you haven't found root cause.
   • Top causes by frequency: wrong input shape/type · off-by-one · missing null check · race condition · shared-state mutation · wrong comparison operator · variable scope · swallowed error · API contract mismatch · environment difference.
   • (gate: can state root cause in one sentence explaining the violated invariant)

5. FIX: root cause, not symptom.

   • Fix the DEFECT, not the FAILURE site. (Null check at crash site = symptom fix.)
   • Write regression test that fails before fix, passes after.
   • Run: original failing case + edge cases + full regression suite.
   • Commit fix + test together.
   • (gate: regression test green; original failing case passes) </steps>

<refactor_mode> Triggers: refactor, restructure, clean up, coupling, dependency, "what breaks if". Same rule: diagnose, then hand off. Do not start cutting inside this mode.

1. MAP: every file/module touching the target. Grep/Glob every reference, or graphify affected <symbol> when the graph is fresh. Build the import/call map.
2. TRACE DEPS: per file, who calls it, who depends on it, what breaks if it changes.
3. MEASURE COUPLING: cross-module reference counts, circular dependencies, the per-function CCN from scripts/complexity.sh.
4. RISKS: current edge cases, what is tested, what is not, invariants to preserve.
5. PLAN: files in change order, tests that must pass before AND after, tier by reversibility x blast radius (file count is only a weak hint). Hand off to /kernel:simplify to execute, which owns the preservation contract and the gate. </refactor_mode>

<diagnosis_output> When the run ends at a diagnosis rather than a fix, emit this and stop:

## Diagnosis: <title>
Mode: bug | refactor · Confidence: high | medium | low
Root cause: <one sentence naming the violated invariant>
Affected: <file - origin> | <file - downstream> ...
Blast radius: N files. Tier 1|2|3.
Hypotheses: 1. <h> -> CONFIRMED | REJECTED (<evidence>) ...
Recommended approach: <what, not how>
Tests required: <fails before> / <passes after>
Next: /kernel:ingest to implement, /kernel:simplify to restructure.

Decide and state the recommendation. Never stop to ask which hypothesis to pursue. </diagnosis_output>

<anti_patterns> Shotgun (random changes until it works) · fix-and-pray (never re-run the original case) · symptom fixing (null check at the crash site) · printf flooding (binary search first, then targeted logging) · blame-the-framework (it's almost never the library) · unscoped "investigate" (scope narrowly or use a subagent so the file reads don't fill context). </anti_patterns>

<when_stuck> Explain the problem in writing · re-read the error message (the answer is there most of the time) · reduce to a minimal reproduction · ask "what changed?" (git log/diff, deps, env) · search the exact error message in quotes · step away, bias accumulates. Re-run the EXACT original failing case before declaring victory; "seems to work" is not evidence. </when_stuck>

<escalation> 30+ min on one hypothesis with no evidence → abandon it. 3+ hypotheses rejected → step back, re-examine assumptions. 2 failed fix attempts → invoke tearitapart; it may be a design problem. Repeated failed corrections in one session → /clear with a minimal reproduction. Bug only in production → add targeted monitoring, document, move on. For 3+ plausible causes, spawn one fresh-context agent per hypothesis (evidence_for / evidence_against / confidence); fresh context catches what a long session anchors past. </escalation> <telemetry> agentdb emit command "debug" "" '{"mode":"bug|refactor","confidence":"high|medium|low","blast_radius":N,"tier":N}' </telemetry>

<on_complete> agentdb write-end '{"skill":"debug","bug":"<description>","root_cause":"<what_broke>","fix":"<what_fixed>","test":"<regression_test_name>","learned":"<pattern_for_future>"}' </on_complete>

</skill>