Code Forge — Debug

⚡ Execution Entry Point (READ THIS FIRST)

When this skill is loaded, you MUST immediately begin executing the Workflow below — do not wait, do not summarize, do not ask "what should I do now". Skills are operational manuals, not reference documents. Read the first executable step, perform it, then the next, etc., until the workflow completes or you reach an AskUserQuestion checkpoint.

If the harness shows you Successfully loaded skill · N tools allowed, that message means the SKILL.md content was injected into your context — it does NOT mean the skill has run. Skills do not "run" autonomously; you run them by executing the Detailed Steps below.

If you find yourself about to say "the skill didn't produce output", "skill 仍未输出", "falling back to manual debugging", "回退到手动 debug", or anything similar, STOP. You have misunderstood how skills work. Go directly to the first executable step and start.

The first user-visible action of this skill should be either (a) the output of the first step, or (b) an AskUserQuestion if the first step needs disambiguation. Never an apology, never a fallback, never silence.

---

Systematic root cause debugging for any technical issue.

When to Use

• Any test failure, bug report, or unexpected behavior
• Performance degradation, build failures, integration issues
• ESPECIALLY when under time pressure or when "one quick fix" seems obvious

For code-forge features: Use /code-forge:fix instead — it adds upstream document tracing and state tracking on top of this methodology.

Iron Law

NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST.

No "let me just try this." No "obvious fix." No guessing. Investigate first.

Workflow

Root Cause Investigation → Pattern Analysis → Hypothesis Testing → Implementation (TDD fix)

Phase 0: Understand the Project

Before investigating, build situational awareness of the codebase:

@../shared/project-analysis.md

Execute steps PA.1 (Project Profile) and PA.2 (Architecture Analysis) at minimum. For complex bugs, also run PA.3 (Language-Specific Deep Scan) on the affected module and PA.4 (Relationship Mapping) to understand how the bug might propagate.

This context informs Phase 1-4: knowing the architecture tells you WHERE to look; knowing relationships tells you WHAT ELSE might be affected.

Four Phases

Complete each phase before moving to the next.

Phase 1: Root Cause Investigation

1. Read error messages carefully — complete messages, not skimmed
2. Reproduce consistently — can you trigger it reliably?
3. Check recent changes — git diff, git log for what changed
4. Gather evidence — add diagnostic instrumentation at each boundary in multi-component systems
5. Trace data flow backward — from the error, walk back through the call chain

Phase 2: Pattern Analysis

1. Find working examples — is there a similar feature that works?
2. Compare against references — read reference code COMPLETELY, not skimmed
3. Identify differences — list EVERY difference between working and broken
4. Understand dependencies — what does this code depend on? What depends on it?

Phase 3: Hypothesis and Testing

1. Form a single hypothesis — state it clearly, write it down, be specific
2. Test minimally — smallest possible change that tests the hypothesis
3. One variable at a time — never change multiple things simultaneously
4. Verify before continuing — did the test confirm or refute the hypothesis?

If you don't know: say so. Don't pretend. "I don't understand why X happens" is valuable information.

Phase 4: Implementation

1. Create a failing test case first — use TDD (see code-forge:tdd)
2. Implement a single fix — ONE change, not "while I'm here" improvements
3. Verify the fix worked — run the test, confirm it passes
4. Run the full test suite — ensure no regressions

When Fixes Fail

Count your fix attempts:

• < 3 attempts: Return to Phase 1, gather more evidence
• >= 3 attempts: STOP. This is NOT a hypothesis failure — it's likely a wrong architecture or wrong mental model. Discuss with the user before proceeding.

Example

Bug: "UserService.getProfile() returns null for valid user IDs"

Phase 1 — Investigate:
  $ grep -r "getProfile" src/  →  found in UserService.ts:42, UserController.ts:15
  $ git log --oneline -5 src/services/UserService.ts  →  recent refactor changed query

Phase 2 — Compare:
  Working: getById() uses parameterized query with $1
  Broken:  getProfile() concatenates id into string (bug introduced in refactor)

Phase 3 — Hypothesis:
  "getProfile builds wrong SQL — id is treated as string, not integer"
  Test: hardcode known id=1 → returns null. Confirmed.

Phase 4 — Fix:
  Write test: expect(getProfile(1)).resolves.toMatchObject({id: 1})
  Fix: use parameterized query ($1) instead of string interpolation
  Verify: test passes, full suite 42/42 green

Decision Rules

Apply these checks before acting:

If you're about to...	Instead...	Why
Skip Phase 1 because the cause seems obvious	Run Phase 1 anyway — it will be fast if you're right	Obvious causes are often symptoms of deeper issues
Apply a fix without reproducing first	Reproduce the bug with a reliable trigger	A fix you can't verify is not a fix
Revert to a working state without understanding	Investigate WHY it broke, then fix	Blind reverts leave the root cause active
Trust the error message at face value	Trace backward from the error through the call chain	Error messages point to symptoms, not causes
Add logging everywhere	Add targeted logging at component boundaries only	Shotgun logging creates noise and obscures signal
Try the same approach again	STOP after 3 failed attempts — reassess your mental model	Repeated failure means wrong diagnosis, not wrong execution

Hard Stops

Halt and reassess if:

• You've attempted the same fix approach 3+ times
• You're adding workarounds instead of addressing the root issue
• You're suppressing or swallowing errors to make tests pass
• You're making changes without a specific hypothesis to test
• The fix is growing larger than the feature it supports