Skip to main content
Generaldraphonix

debugging

Use when blocked workers, test failures, build errors, runtime crashes, or integration issues need systematic root-cause debugging. Reads history/learnings/critical-patterns.md to avoid re-solving known issues. Writes debug notes that compounding can later capture.

Stars
218
Source
draphonix/skills
Updated
2026-05-20
Slug
draphonix--skills--debugging
View on GitHubRaw SKILL.md

// install — copy + paste into any project

mkdir -p .claude/skills && curl -fsSL https://raw.githubusercontent.com/draphonix/skills/HEAD/plugins/khuym/skills/debugging/SKILL.md -o .claude/skills/debugging.md

Drops the SKILL.md into .claude/skills/debugging.md. Works with Claude Code, Cursor, and any agent that loads SKILL.md files from .claude/skills/.

Debugging

If .khuym/onboarding.json is missing or stale for the current repo, stop and invoke khuym:using-khuym before continuing.

Resolve blockers and failures systematically. Do not guess. Triage first, reproduce second, diagnose third, fix fourth.

When To Use This Skill

  • a build fails
  • a test fails
  • a runtime crash or exception occurs
  • an integration breaks
  • a worker is blocked by dependencies or reservations
  • reviewing or executing hands off with a failure that needs root-cause analysis

Step 1: Triage

Classify the issue before you investigate it.

Type Signals
Build failure compiler error, type error, missing module, bundler failure
Test failure assertion mismatch, timeout, snapshot diff, flake
Runtime error crash, uncaught exception, undefined behavior
Integration failure HTTP 4xx/5xx, auth failure, env mismatch, schema mismatch
Worker blocker circular bead dependency, conflicting reservations, no safe execution path

Output a one-line classification:

[TYPE] in [component]: [symptom]

Step 2: Reproduce

Check history/learnings/critical-patterns.md first. If a matching pattern already exists, start from that fix path.

If not, rerun the exact failing command and capture the exact output.

Examples:

npm run build 2>&1 | tee /tmp/debug-output.txt
pytest tests/specific_test.py -v 2>&1 | tee /tmp/debug-output.txt

Run it twice. If it is intermittent, treat it as a flaky failure rather than a deterministic one.

Step 3: Diagnose

Work through these checks in order.

3a. Read the relevant files

Use the failing output to identify the smallest relevant slice. Do not read the entire repo.

3b. Check recent changes

git log --oneline -20
git blame <file> -L <line>,<line>
git diff HEAD~3 -- <file>

3c. Check bead intent

br show <bead-id>

Ask whether the code drifted from the bead, or the bead itself is wrong.

3d. Check locked decisions

Read the relevant CONTEXT.md entries and confirm the implementation did not violate a locked decision.

3e. Check local reservation state

node .codex/khuym_reservations.mjs list --active-only --json

Look for:

  • overlapping reservations
  • leaked reservations from a finished worker
  • a worker that still holds files after a blocker or timeout

Also inspect .khuym/state.json for the active worker list.

3f. Check recent worker results in the parent thread

If this debugging pass was spawned from swarming, use the parent-thread context and the saved worker status in .khuym/state.json as the coordination surface. Do not assume an external inbox exists.

3g. Write the root cause sentence

Do not proceed until you can write:

Root cause: <file>:<line> — <what is wrong and why>

If you cannot write that sentence, you do not have the root cause yet.

Step 4: Fix And Verify

Small fix

If the fix is obvious and low risk:

  • implement directly
  • run the exact failing command again
  • run the next-wider verification that protects against regressions

Larger fix

If the fix is cross-cutting or changes the intended behavior:

br create "Fix: <root cause summary>" -t task --blocks <original-bead-id>

Then implement against that new bead.

Decision violation

If a locked decision was violated:

  • do not silently "fix" it by changing behavior on your own
  • return or report a blocker summary to the parent thread or user
  • propose the conservative fix that honors CONTEXT.md

Reservation-related fixes

If the failure is caused by leaked or stale reservations:

  1. inspect the holder
  2. release the reservation only if the holder is clearly done or abandoned
  3. note the release explicitly in your final report

Use:

node .codex/khuym_reservations.mjs release --agent "<codex-name>" --bead "<bead-id>" --json
node .codex/khuym_reservations.mjs sweep --json

Verify

The original failing command must pass cleanly. If it still fails, return to diagnosis.

Step 5: Report

If you are inside a swarm, return the debugging result to the parent thread using a clear status heading:

  • [DONE] if the fix is complete and verified
  • [BLOCKED] if the problem needs a decision, another worker release, or a broader redesign

At minimum include:

  • root cause sentence
  • fix summary
  • verification result
  • reservation impact, if any
  • next action needed

If you are working directly for the user, give the same information in the final response.

Step 6: Learn

If this exposed a new reusable failure pattern, write a debug note for compounding so the lesson can be promoted later.

If the failure matched an existing pattern from critical-patterns.md, verify whether that guidance still works. If not, flag it for compounding.

Blocker-Specific Protocol

When a worker is stuck rather than code-broken:

  1. inspect cycles and dependencies:
    bv --robot-insights 2>/dev/null | jq '.Cycles'
  2. inspect local reservations:
    node .codex/khuym_reservations.mjs list --active-only --json
  3. determine whether the worker is:
    • waiting on another bead
    • blocked by an overlapping reservation
    • blocked by a real product decision

If it is only waiting, return [BLOCKED] with the dependency or reservation holder and stop.

If it is a real dead-end, return [BLOCKED] with concrete options for the parent or user.

Do not spin.

Red Flags

  • fixing symptoms instead of root cause
  • skipping reproduction
  • ignoring critical-patterns.md
  • patching around a locked decision violation
  • reporting success without rerunning the original failing command
  • forgetting to account for local reservations during swarm debugging

Quick Reference

Situation First action
Build fails rerun the exact build command
Test fails rerun the exact test and capture assertion output
Runtime crash read the stack trace and find the first line in your code
Integration error check env/config, then the real response body
Worker stuck inspect bv plus local reservations
Recurring issue check history/learnings/critical-patterns.md first