跳到正文
FunCoding

搜索

搜索文档、Skill 和 MCP

agent-md-refactor

Refactor bloated AGENTS.md, CLAUDE.md, or similar agent instruction files to follow progressive disclosure principles. Splits monolithic files into organized, linked documentation.

代码质量与审查2.5kskills/agent-md-refactor/SKILL.md

安装

把这段话发给 Claude Code、Codex 或 Cursor。智能体会先检查安全性,你确认后才安装。

读取 https://funcoding.ai/skills/softaworks/agent-toolkit/agent-md-refactor/install.md ,按里面的步骤帮我安装这个 Skill。

SKILL.md

Agent MD Refactor

Refactor bloated agent instruction files (AGENTS.md, CLAUDE.md, COPILOT.md, etc.) to follow progressive disclosure principles - keeping essentials at root and organizing the rest into linked, categorized files.


Triggers

Use this skill when:

  • "refactor my AGENTS.md" / "refactor my CLAUDE.md"
  • "split my agent instructions"
  • "organize my CLAUDE.md file"
  • "my AGENTS.md is too long"
  • "progressive disclosure for my instructions"
  • "clean up my agent config"

Quick Reference

PhaseActionOutput
1. AnalyzeFind contradictionsList of conflicts to resolve
2. ExtractIdentify essentialsCore instructions for root file
3. CategorizeGroup remaining instructionsLogical categories
4. StructureCreate file hierarchyRoot + linked files
5. PruneFlag for deletionRedundant/vague instructions

Process

Phase 1: Find Contradictions

Identify any instructions that conflict with each other.

Look for:

  • Contradictory style guidelines (e.g., "use semicolons" vs "no semicolons")
  • Conflicting workflow instructions
  • Incompatible tool preferences
  • Mutually exclusive patterns

For each contradiction found:

## Contradiction Found

**Instruction A:** [quote]
**Instruction B:** [quote]

**Question:** Which should take precedence, or should both be conditional?

Ask the user to resolve before proceeding.


Phase 2: Identify the Essentials

Extract ONLY what belongs in the root agent file. The root should be minimal - information that applies to every single task.

Essential content (keep in root):

CategoryExample
Project descriptionOne sentence: "A React dashboard for analytics"
Package managerOnly if not npm (e.g., "Uses pnpm")
Non-standard commandsCustom build/test/typecheck commands
Critical overridesThings that MUST override defaults
Universal rulesApplies to 100% of tasks

NOT essential (move to linked files):

  • Language-specific conventions
  • Testing guidelines
  • Code style details
  • Framework patterns
  • Documentation standards
  • Git workflow details

Phase 3: Group the Rest

Organize remaining instructions into logical categories.

Common categories:

CategoryContents
typescript.mdTS conventions, type patterns, strict mode rules
testing.mdTest frameworks, coverage, mocking patterns
code-style.mdFormatting, naming, comments, structure
git-workflow.mdCommits, branches, PRs, reviews
architecture.mdPatterns, folder structure, dependencies
api-design.mdREST/GraphQL conventions, error handling
security.mdAuth patterns, input validation, secrets
performance.mdOptimization rules, caching, lazy loading

Grouping rules:

  1. Each file should be self-contained for its topic
  2. Aim for 3-8 files (not too granular, not too broad)
  3. Name files clearly: {topic}.md
  4. Include only actionable instructions

Phase 4: Create the File Structure

Output structure:

project-root/
├── CLAUDE.md (or AGENTS.md)     # Minimal root with links
└── .claude/                      # Or docs/agent-instructions/
    ├── typescript.md
    ├── testing.md
    ├── code-style.md
    ├── git-workflow.md
    └── architecture.md

Root file template:

# Project Name

One-sentence description of the project.

## Quick Reference

- **Package Manager:** pnpm
- **Build:** `pnpm build`
- **Test:** `pnpm test`
- **Typecheck:** `pnpm typecheck`

## Detailed Instructions

For specific guidelines, see:
- [TypeScript Conventions](.claude/typescript.md)
- [Testing Guidelines](.claude/testing.md)
- [Code Style](.claude/code-style.md)
- [Git Workflow](.claude/git-workflow.md)
- [Architecture Patterns](.claude/architecture.md)

Each linked file template:

# {Topic} Guidelines

## Overview
Brief context for when these guidelines apply.

## Rules

### Rule Category 1
- Specific, actionable instruction
- Another specific instruction

### Rule Category 2
- Specific, actionable instruction

## Examples

### Good
\`\`\`typescript
// Example of correct pattern
\`\`\`

### Avoid
\`\`\`typescript
// Example of what not to do
\`\`\`

Phase 5: Flag for Deletion

Identify instructions that should be removed entirely.

Delete if:

CriterionExampleWhy Delete
Redundant"Use TypeScript" (in a .ts project)Agent already knows
Too vague"Write clean code"Not actionable
Overly obvious"Don't introduce bugs"Wastes context
Default behavior"Use descriptive variable names"Standard practice
OutdatedReferences deprecated APIsNo longer applies

Output format:

## Flagged for Deletion

| Instruction | Reason |
|-------------|--------|
| "Write clean, maintainable code" | Too vague to be actionable |
| "Use TypeScript" | Redundant - project is already TS |
| "Don't commit secrets" | Agent already knows this |
| "Follow best practices" | Meaningless without specifics |

Execution Checklist

[ ] Phase 1: All contradictions identified and resolved
[ ] Phase 2: Root file contains ONLY essentials
[ ] Phase 3: All remaining instructions categorized
[ ] Phase 4: File structure created with proper links
[ ] Phase 5: Redundant/vague instructions removed
[ ] Verify: Each linked file is self-contained
[ ] Verify: Root file is under 50 lines
[ ] Verify: All links work correctly

Anti-Patterns

AvoidWhyInstead
Keeping everything in rootBloated, hard to maintainSplit into linked files
Too many categoriesFragmentationConsolidate related topics
Vague instructionsWastes tokens, no valueBe specific or delete
Duplicating defaultsAgent already knowsOnly override when needed
Deep nestingHard to navigateFlat structure with links

Examples

Before (Bloated Root)

# CLAUDE.md

This is a React project.

## Code Style
- Use 2 spaces
- Use semicolons
- Prefer const over let
- Use arrow functions
... (200 more lines)

## Testing
- Use Jest
- Coverage > 80%
... (100 more lines)

## TypeScript
- Enable strict mode
... (150 more lines)

After (Progressive Disclosure)

# CLAUDE.md

React dashboard for real-time analytics visualization.

## Commands
- `pnpm dev` - Start development server
- `pnpm test` - Run tests with coverage
- `pnpm build` - Production build

## Guidelines
- [Code Style](.claude/code-style.md)
- [Testing](.claude/testing.md)
- [TypeScript](.claude/typescript.md)

Verification

After refactoring, verify:

  1. Root file is minimal - Under 50 lines, only universal info
  2. Links work - All referenced files exist
  3. No contradictions - Instructions are consistent
  4. Actionable content - Every instruction is specific
  5. Complete coverage - No instructions were lost (unless flagged for deletion)
  6. Self-contained files - Each linked file stands alone

相似的 Skill

claude-api
anthropics/skills180k

claude-api

Reference for the Claude API / Anthropic SDK — model ids, pricing, params, streaming, tool use, MCP, agents, caching, token counting, model migration. TRIGGER — read BEFORE opening the target file; don't skip because it "looks like a one-liner" — whenever: the prompt names Claude/Anthropic in any form (Claude, Anthropic, Fable, Opus, Sonnet, Haiku, `anthropic`, `@anthropic-ai`, `claude-*`, `us.anthropic.*`, `[1m]`); the user asks about an LLM (pricing/model choice/limits/caching) — never answer from memory; OR the task is LLM-shaped with provider unstated (agent/MCP/tool-definition/multi-agent/RAG/LLM-judge/computer-use; generate/summarize/extract/classify/rewrite/converse over NL; debugging refusals/cutoffs/streaming/tool-calls/tokens). SKIP only when another provider is being worked on (overrides all triggers): OpenAI/GPT/Gemini/Llama/Mistral/Cohere/Ollama named in the query; OR `grep -rE 'openai|langchain_openai|google.generativeai|genai|mistralai|cohere|ollama'` over the project hits (run this grep FIRST if no provider named — don't Read the file).

代码质量与审查

ponytail-review
DietrichGebert/ponytail158k

ponytail-review

Quality review of a change: is the logic right, is it safe, does it hold under real load, is risky code tested, is it fast enough, and is every line needed. Reads the connected code, not only the diff. Each finding is explained in plain English. Use for "review this", "code review", "review the last commit", "review my PR", "is this over-engineered", /ponytail-review.

代码质量与审查

code-review-and-quality
addyosmani/agent-skills103k

code-review-and-quality

Conducts multi-axis code review. Use before merging any change. Use when reviewing code written by yourself, another agent, or a human. Use when you need to assess code quality across multiple dimensions before it enters the main branch. Use when asked to review a diff or a pull request, even when the diff is pasted inline.

代码质量与审查

documentation-and-adrs
addyosmani/agent-skills103k

documentation-and-adrs

Records decisions and documentation. Use when you need to document an architecture decision (ADR) or the reasoning behind a design choice, when changing public APIs, shipping features, or when you need to record context that future engineers and agents will need to understand the codebase.

代码质量与审查

code-simplification
addyosmani/agent-skills103k

code-simplification

Simplifies code for clarity. Use when refactoring code for clarity without changing behavior. Use when code works but is harder to read, maintain, or extend than it should be. Use when reviewing code that has accumulated unnecessary complexity.

代码质量与审查

understand
Egonex-AI/Understand-Anything86k

understand

Analyze a codebase to produce an interactive knowledge graph for understanding architecture, components, and relationships

代码质量与审查