搜索

PythonReactSEOPDF
GitHub Skill

reset

Reset after failures, blockers, or when stuck. Use when execution hits a wall, something breaks, or we need to step back and reassess. This is the fifth system in the 5-system framework.

ioku24☆ 1 仓库 Stars0 仓库 Forks

安装与使用

请根据来源仓库和你的 AI 工具说明安装此技能。先阅读技能文件和依赖要求,再添加到工作流。

查看来源 ↗使用指南

技能说明

Reset System (Failure Recovery)

Purpose: Gracefully recover from setbacks, learn, and continue. When to trigger: When stuck, something breaks, approach isn't working, or need to step back.

Failure Response Protocol

Step 1: STOP

Immediately cease current approach.

Do NOT:

  • Try "one more thing"
  • Keep pushing the same failing approach
  • Make changes hoping they'll work
  • Ignore the signals

Do:

  • Acknowledge something isn't working
  • Take a breath
  • Prepare to capture state
Step 2: CAPTURE STATE

Document what happened:

## Failure Capture

**What we were trying to do:**
[Description of goal]

**What actually happened:**
[Description of failure/block]

**Error/symptom observed:**
[Specific error messages, unexpected behavior]

**What we tried:**
1. Attempt 1: [what] → [result]
2. Attempt 2: [what] → [result]
3. Attempt 3: [what] → [result]

**Current git state:**
[Branch, commit, uncommitted changes]

Log this to .claude/failure-log.md.

Step 3: CHECKPOINT

Ensure state is preserved:

  1. Git status - Are there uncommitted changes to save?
  2. Working state - Note the last known working commit
  3. Context - Save any important context that might be lost
git status
git stash  # if needed to preserve work-in-progress
Step 4: DIAGNOSE

Determine the root cause category:

Clarity Problem (misunderstanding)

  • Requirements were ambiguous
  • Success criteria unclear
  • Scope was wrong → Return to Clarity System

Identity Problem (hidden issue)

  • Discovered unknown constraint
  • Missed a dependency
  • Technical limitation found → Run Identity System

Priority Problem (wrong focus)

  • Solving the wrong problem
  • This should have waited
  • Dependencies not addressed → Return to Priority System

Execution Problem (bad approach)

  • Approach was flawed
  • Implementation strategy wrong
  • Need different technique → Try different approach, same task

External Block (outside control)

  • API down
  • Missing credentials
  • Need user input → Document and wait/ask
Step 5: RESET TO APPROPRIATE SYSTEM

Based on diagnosis:

Root Cause Action
Clarity Update active-context.md, restart with clarity
Identity Run identification, log new issues
Priority Re-evaluate, might deprioritize
Execution New approach, same goal
External Document, notify user, wait
Step 6: LEARN

Update .claude/learnings.md with:

## Learning Entry - [Date]

**Failure:** [Brief description]

**Root Cause:** [What actually went wrong]

**Pattern:** [Is this a recurring issue?]

**Prevention:** [How to avoid this in future]

**New Rule/Check:** [Any process improvement to add]

Failure Log Format

Maintain .claude/failure-log.md:

# Failure Log

## Entry Template

| Date | Task | What Failed | Root Cause | Resolution | Learning |
|------|------|-------------|------------|------------|----------|
| 2026-01-01 | Add auth | Type errors cascaded | Didn't validate early | Reset, fix types first | Run tsc after every change |

## Detailed Entries

### [Date] - [Brief Title]

**Context:**
What we were doing

**Failure:**
What went wrong

**Attempts:**
What we tried

**Resolution:**
How it was fixed

**Takeaway:**
What we learned

Recovery Strategies

For Code Failures
  1. git stash or commit current state
  2. git checkout . to return to last working state
  3. Re-approach with new strategy
For Conceptual Failures
  1. Step back to clarity
  2. Re-read requirements
  3. Ask user for clarification
For Integration Failures
  1. Isolate the failing component
  2. Test in isolation
  3. Gradually reintegrate
For Persistent Failures
  1. Document thoroughly
  2. Ask user for help
  3. Consider if this is the right approach at all

Rules

  1. No shame in reset - It's better to restart cleanly than compound mistakes
  2. Document first - Don't reset without capturing what happened
  3. Learn always - Every failure is information
  4. Stay calm - Frustration compounds errors
  5. Ask for help - The user is a partner, not a judge

Transition

After reset:

  • Clarity problem → Clarity System
  • Hidden issue found → Identity System
  • Wrong priority → Priority System
  • Bad approach → Execution System with new approach
  • Resolved → Continue with task

This is System 5 of 5: Clarity → Identity → Priority → Execution → Reset

The Complete Cycle

          ┌─────────────────────────────────────────────┐
          │                                             │
          ▼                                             │
    ┌─────────┐     ┌──────────┐     ┌──────────┐      │
    │ CLARITY │ ──▶ │ IDENTITY │ ──▶ │ PRIORITY │      │
    └─────────┘     └──────────┘     └──────────┘      │
                                           │            │
                                           ▼            │
                                    ┌───────────┐       │
                                    │ EXECUTION │       │
                                    └───────────┘       │
                                           │            │
                              ┌────────────┴────────┐   │
                              ▼                     ▼   │
                        ┌─────────┐           ┌───────┐ │
                        │ FAILURE │           │SUCCESS│ │
                        │ (Reset) │           └───────┘ │
                        └─────────┘                     │
                              │                         │
                              └─────────────────────────┘

使用示例

查看来源仓库中的使用示例与配置说明。

查看来源 ↗

来源历史

收录时间
2026-01-20
索引更新时间
2026-01-26
Branch
main

收录与同步日期属于本站记录。源码修改历史请查看来源仓库。

GitHub ↗

chat Comments (0)

chat_bubble_outline

No comments yet. Be the first to share your thoughts!