AgentSkillsCN

incremental-fixes

适用于调试复杂问题,或在先前的修复尝试均告失败时使用。通过在每一步之间严格执行“一次只做一项变更”的原则,并辅以验证环节,有效减少“过度修改”的风险。

SKILL.md
--- frontmatter
name: incremental-fixes
description: Use when debugging complex issues or when previous fix attempts have failed. Reduces "excessive changes" by enforcing one-change-at-a-time discipline with verification between each step.

Incremental Fixes

Use this skill to prevent "excessive changes" friction - the pattern where many edits are made but only partial outcomes are achieved.

The Problem

From insights analysis:

  • 474 sessions flagged "excessive_changes" as friction
  • High activity (many edits) but only "partially_achieved" outcomes
  • Multiple changes before testing = can't isolate what worked

The Solution

One change. Verify. Next change.

Default Prompt Template

When starting any debugging or fix task, use this prompt:

code
Let's fix this incrementally. Propose ONE change, I'll approve it, then
we verify it works before moving to the next. Start with the most likely
root cause.

The Process

Step 1: Diagnose First

Before ANY changes:

  1. Read error messages completely
  2. Identify the most likely root cause
  3. Explain your hypothesis

Do not propose fixes yet.

Step 2: Propose Single Change

Propose exactly ONE change:

  • What file/line to change
  • What the change is
  • Why this addresses the root cause
  • How to verify it worked

Wait for approval before implementing.

Step 3: Implement and Verify

After approval:

  1. Make the single change
  2. Run verification (tests, health checks, etc.)
  3. Report result: worked or didn't work

Step 4: Iterate or Complete

  • If it worked: Ask if more changes needed, or declare complete
  • If it didn't work: Return to Step 1 with new information, propose different change

Never stack multiple unverified changes.

Red Flags - STOP

If you catch yourself:

  • Proposing multiple changes at once
  • Making changes without verification between
  • "While I'm here, let me also..."
  • "These are related, so I'll do them together"

STOP. Return to single-change discipline.

When to Use

  • Debugging any issue
  • After a fix attempt failed
  • When the problem is unclear
  • When previous session ended with "confused" state
  • When you've already tried 2+ things that didn't work

Relationship to Other Skills

This skill complements:

  • superpowers:systematic-debugging - Use that for root cause investigation
  • /platform-healthcheck - Use that for verification steps
  • /linear-python - Use that to checkpoint progress between changes

Why This Works

Without IncrementalWith Incremental
5 changes, something broke, which one?1 change, verify, know exactly what happened
"Partially achieved"Clear pass/fail per change
Can't resume mid-debugEach change is a checkpoint
Cascading issues hiddenCascading issues caught immediately