AgentSkillsCN

systematic-debugging

在尝试修复前进行四阶段根本原因调查。适用于调试bug、测试失败或意外行为时使用。

SKILL.md
--- frontmatter
name: systematic-debugging
description: Four-phase root cause investigation before attempting fixes. Use when debugging bugs, test failures, or unexpected behavior.

Systematic Debugging

ALWAYS find root cause before attempting fixes. Symptom fixes are failure.

When to Use

  • •Encountering bugs or test failures
  • •Unexpected behavior in code
  • •Before proposing any fix

The Four Phases

Phase 1: Root Cause Investigation

Before proposing any solution:

  1. •Read error messages thoroughly - Don't skip warnings or stack traces; they often contain exact solutions
  2. •Reproduce consistently - Verify you can trigger the issue reliably with documented steps
  3. •Check recent changes - Examine git diff, dependencies, and configuration modifications
  4. •Gather diagnostic evidence - In multi-component systems, add instrumentation at component boundaries
  5. •Trace data flow - Backward trace from the error to find where bad values originate

Phase 2: Pattern Analysis

Establish the pattern before fixing:

  1. •Locate similar working code in the codebase
  2. •Read reference implementations completely (not skimmed)
  3. •List every difference between working and broken code
  4. •Understand all dependencies and assumptions

Phase 3: Hypothesis Testing

Apply scientific method:

  1. •State your hypothesis clearly: "I believe X is failing because Y, evidenced by Z"
  2. •Test with the smallest possible change
  3. •Change only ONE variable at a time
  4. •Verify results before proceeding

Phase 4: Implementation

Fix the root cause systematically:

  1. •Create a failing test case first (TDD)
  2. •Implement a single fix addressing only the root cause
  3. •Verify the fix resolves the issue without breaking other tests
  4. •If fix doesn't work, return to Phase 1

Red Flags - STOP Immediately

  • •Proposing fixes without understanding the issue
  • •Attempting multiple simultaneous changes
  • •Assuming problems without verification
  • •Skipping evidence gathering
  • •Making "quick fixes" before investigation

When 3+ Fixes Fail

STOP. This signals an architectural problem, not a fixable bug:

  1. •Do not attempt another fix
  2. •Return to Phase 1
  3. •Question whether the underlying pattern/design is sound
  4. •Ask: "Should we refactor architecture vs. continue fixing symptoms?"

Random fixes waste time and create new bugs. Quick patches mask underlying issues.

Results

Systematic approach: 15-30 minutes to resolution with 95% first-time success vs. Trial-and-error: 2-3 hours of thrashing with 40% success and new bugs introduced