debug

建立系统的调试工作流程。当用户说“调试这个”、“/debug”、报告Bug、分享错误信息或堆栈跟踪，或询问“为什么这段代码出问题”时使用。遵循“复现→追踪→假设→修复→验证”的结构化流程。

SKILL.md

--- frontmatter

name: debug
description: Systematic debugging workflow. Use when the user says "debug this", "/debug", reports a bug, shares an error/stacktrace, or says "why is this broken". Follows a structured reproduce → trace → hypothesize → fix → verify cycle.
allowed-tools:
  - Bash
  - Read
  - Write
  - Edit
  - Glob
  - Grep

Debug

系统化调试：复现 → 追踪 → 假设 → 修复 → 验证。

Workflow

1. Reproduce

First, make the bug visible:

•If the user gave a command that fails → run it, capture output
•If the user gave a stacktrace → read it carefully
•If the user described a behavior → write a minimal reproduction

The bug is NOT understood until you can reproduce it. Don't guess.

2. Read the Stacktrace

Parse the error:

•Error type: what category (TypeError, 404, panic, etc.)
•Location: file:line where it originates
•Call chain: how execution got there
•Data: what values caused the error

Read the source file at the error location. Read surrounding context (±20 lines).

3. Trace the Code Path

Follow the execution path backwards from the error:

•What function called this?
•What data was passed in?
•Where did that data come from?

Use Grep to find callers. Read each file in the chain.

4. Hypothesize

Based on the trace, form a hypothesis:

•"X is null because Y didn't initialize it"
•"The API returns Z format but code expects W"
•State it clearly before attempting a fix.

5. Fix

Apply the minimal fix that addresses the root cause:

•Don't refactor surrounding code
•Don't add unrelated improvements
•Fix the specific issue

6. Verify

Run the original reproduction:

•The error must be gone
•Run the related tests to confirm no regression

bash

# Run the failing command again
[original command]

# Run related tests
[test command]

Show the output. The fix is NOT done until verification passes.

Anti-patterns

•不要猜——先复现
•不要一次改多处——每次只改一个地方，验证后再改下一个
•不要添加 workaround 而不理解 root cause
•不要说 "should work now" 而不运行验证命令