code-documenter（代码记录器）

Code Documenter Skill

An intelligent documentation system that analyzes codebases and generates thorough, comprehensive documentation tailored to your project type and audience needs.

Core Philosophy

Documentation serves readers, not authors. Every decision about structure, depth, and content is evaluated from the reader's perspective:

• Will this help them understand?
• Will this help them succeed?
• Will this answer their questions?
• Does this earn its place?

Comprehensive without overwhelm. Thorough coverage of what matters, ruthless cutting of what doesn't. The goal is complete, accurate, useful documentation—not exhaustive documentation of every line of code.

Documentation as code artifact. Docs should be versioned, tested, and maintained with the same rigor as code. They're not afterthoughts; they're essential.

When to Use This Skill

Primary use cases:

• After completing significant new work (epic, major feature, new app)
• When shipping an open source project
• When documentation has fallen out of sync with code
• When onboarding requires better documentation
• When setting up a new project properly

Not for:

• In-progress features (wait until stable)
• Code comments or docstrings (this generates external docs)
• API reference generation from code (use language-specific tools for that)

Session Flow

Mode Selection

The skill operates in two modes:

The skill will ask: "Quick update or comprehensive documentation?"

Phase 1: Multi-Agent Project Analysis

The skill deploys specialized analysis agents to understand your project:

Analysis Phase:
├─ Agent 1: Project Structure
│  └─ Scanning file tree, identifying project type, tech stack
├─ Agent 2: Code Surface Analysis
│  └─ Finding APIs, components, commands, exports
├─ Agent 3: Dependency Analysis
│  └─ Reviewing packages, frameworks, key dependencies
├─ Agent 4: Git History Analysis
│  └─ Analyzing commits since last doc update
├─ Agent 5: Existing Documentation Review
│  └─ Reading current docs, assessing state
└─ Synthesis: Generating documentation plan...

Quick Mode: Focused analysis on what changed since last update Comprehensive Mode: Full analysis of entire project

Each agent reports findings in structured format. You see everything happening.

Phase 2: Documentation State Assessment

The skill compares code state vs. documentation state:

If manifest exists (.doc-state.json):

• Loads documentation state from last run
• Compares current code vs. last documented state
• Identifies what's changed (added/modified/removed)
• Shows you the documentation debt

If no manifest:

• First time documenting this project
• Will create fresh manifest after completion

Key metrics shown:

• Health Score: Current documentation health (0-100)
• Coverage: What % of public surface is documented
• Freshness: How current are docs vs. code
• Debt: What needs attention (prioritized)

Phase 3: Audience & Scope Discovery

The skill asks:

1. "Who needs these docs?"

- Developers only - Users only - Both developers and users

1. "What depth of documentation?"

- Standard: Right-sized coverage (4,000-7,000 words) - Deep: Comprehensive with internals (8,000-12,000 words)

Quick Mode: Uses stored preferences from manifest Comprehensive Mode: Asks explicitly, shows examples

Phase 4: Documentation Boundaries

The skill proposes what should be documented:

Based on my analysis of your Express + PostgreSQL API:

PUBLIC SURFACE (Always Document)
├─ 12 API endpoints in /routes
├─ Database schema (3 tables)
├─ 14 environment variables
└─ Authentication flow

INTERNAL IMPLEMENTATION (Recommended for Deep)
├─ 6 middleware functions
├─ Error handling patterns
└─ Database migration strategy

INFRASTRUCTURE (Essential for all levels)
├─ Docker setup
├─ CI/CD pipeline
└─ Deployment process

EXCLUDED (Recommend skip)
├─ Test files (obvious from names)
├─ Build scripts (standard tooling)
└─ Internal helpers (<10 lines each)

Adjust these boundaries? [approve/modify]

You approve or adjust before documentation generation begins.

Phase 5: Documentation Plan Presentation

The skill presents a complete plan:

DOCUMENTATION PLAN

Structure:
└─ /docs
   ├─ developers/
   │  ├─ api.md (API endpoint reference)
   │  ├─ architecture.md (System design + diagrams)
   │  ├─ contributing.md (How to contribute)
   │  ├─ deployment.md (Deploy & operate)
   │  ├─ troubleshooting.md (Common issues)
   │  ├─ examples/ (12 runnable examples)
   │  └─ adr/ (4 architecture decisions)
   ├─ users/
   │  ├─ getting-started.md (Quick start guide)
   │  ├─ features.md (Feature overview)
   │  ├─ troubleshooting.md (User-facing issues)
   │  └─ examples/ (5 user examples)
   ├─ CHANGELOG.md (Documentation change log)
   └─ documentation-map.md (Navigation guide)

Files to update:
- README.md (complete rewrite, progressive disclosure)
- docs/developers/api.md (2 new endpoints, 1 modified)
- docs/CHANGELOG.md (new entry)

Files to create:
- docs/adr/004-caching-strategy.md (new decision)
- docs/developers/examples/pagination.js (new example)

Files to remove:
- docs/developers/legacy-auth.md (endpoint removed)

Estimated scope: ~5,200 words total
Target health score: 92/100

Proceed? [yes/adjust scope/abort]

Comprehensive Mode: Review and approve before generation Quick Mode: Brief preview, option to review or proceed

Phase 6: Documentation Generation

Comprehensive Mode (with Quality Gates)

The skill works through documentation in phases, pausing for review:

Gate 1: Core Documentation (README + Getting Started)

• Generates README with progressive disclosure
• Creates getting started guides
• Quality Check: Does this hook and onboard effectively?
• You review and approve or request changes

Gate 2: Reference Documentation (API/Commands/Components)

• Generates reference documentation
• Creates working examples
• Quality Check: Is everything covered? Examples clear?
• You review and approve or request changes

Gate 3: Architecture & Decisions

• Documents architecture with Mermaid diagrams
• Creates ADRs for key decisions
• Quality Check: Does this explain the WHY?
• You review and approve or request changes

Gate 4: Supporting Documentation

• Generates troubleshooting guides
• Creates contributing guidelines
• Documentation map for navigation
• Quality Check: Complete and helpful?
• You review and approve or request changes

Gate 5: Polish & Verification

• Generates test scripts for examples
• Creates link validation script
• Runs accessibility check
• Final health score calculation
• Final review

Quick Mode (Autonomous)

The skill executes the plan efficiently:

• Updates only changed sections
• Preserves manual edits in unchanged areas
• Generates new content as needed
• Updates manifest and changelog
• Reports final results

Phase 7: Session Completion

The skill produces:

Files Created/Updated:

• All documentation files as planned
• .doc-state.json (updated manifest)
• docs/CHANGELOG.md (new entry)
• Test and validation scripts

Documentation Health Report:

DOCUMENTATION HEALTH REPORT

Overall Health Score: 92/100 (↑ from 78)

Component Scores:
├─ Coverage Health: 95/100 (↑ from 82)
│  └─ 97% of public surface documented
├─ Freshness Health: 98/100 (↑ from 65)
│  └─ All docs current as of commit a3f2b1c
├─ Quality Health: 88/100 (↑ from 81)
│  └─ 12 examples, 4 ADRs, troubleshooting complete
└─ Consistency Health: 90/100 (↑ from 84)
   └─ Tone consistent, terminology standardized

Debt Status:
├─ Critical: 0 items (was 2)
├─ Important: 1 item (was 4)
└─ Minor: 3 items (unchanged)

Next session recommendations:
- Document the new webhook system (flagged as potential)
- Add performance troubleshooting section
- Consider hosting docs on GitHub Pages

Session Notes:

• Decisions made
• Scope adjustments
• What was included/excluded and why
• Next steps

File Outputs

Core Documentation Structure

Developer-only projects:

/docs
├── CHANGELOG.md
├── api.md / commands.md / components.md
├── architecture.md
├── contributing.md
├── deployment.md
├── troubleshooting.md
├── examples/
│   ├── example-1.js
│   ├── example-2.js
│   └── test-examples.sh
├── adr/
│   ├── 001-initial-decisions.md
│   ├── 002-database-choice.md
│   └── ...
└── scripts/
    ├── validate-links.sh
    └── accessibility-check.md

Multi-audience projects:

/docs
├── CHANGELOG.md
├── documentation-map.md
├── users/
│   ├── getting-started.md
│   ├── features.md
│   ├── troubleshooting.md
│   └── examples/
│       ├── basic-usage.js
│       └── advanced-usage.js
├── developers/
│   ├── api.md / architecture.md
│   ├── contributing.md
│   ├── deployment.md
│   ├── troubleshooting.md
│   ├── examples/
│   │   ├── integration.js
│   │   ├── extending.js
│   │   └── test-examples.sh
│   └── adr/
│       ├── 001-framework-choice.md
│       └── ...
└── scripts/
    ├── validate-links.sh
    └── accessibility-check.md

Documentation Manifest (.doc-state.json)

Tracks complete documentation state:

{
  "version": "1.0",
  "project": {
    "name": "my-api",
    "type": "express-api",
    "lastScanned": "2025-01-10T14:30:00Z",
    "gitCommit": "a3f2b1c"
  },
  "preferences": {
    "audiences": ["developers", "users"],
    "depthLevel": "standard",
    "tone": "professional"
  },
  "healthScore": {
    "overall": 92,
    "components": {
      "coverage": 95,
      "freshness": 98,
      "quality": 88,
      "consistency": 90
    },
    "trend": [78, 85, 92]
  },
  "coverage": {
    "endpoints": { "total": 12, "documented": 12, "changed": 0 },
    "components": { "total": 8, "documented": 8, "changed": 0 },
    "schemas": { "total": 3, "documented": 3, "changed": 0 }
  },
  "debt": {
    "critical": [],
    "important": [
      {
        "item": "Document webhook system",
        "effort": "medium",
        "status": "to-fix"
      }
    ],
    "minor": [
      {
        "item": "Add performance troubleshooting",
        "effort": "low",
        "status": "accepted"
      }
    ]
  },
  "documentationMap": {
    "README.md": {
      "lastUpdated": "2025-01-10T14:30:00Z",
      "covers": ["overview", "quickstart"],
      "wordCount": 850
    },
    "docs/developers/api.md": {
      "lastUpdated": "2025-01-10T14:30:00Z",
      "covers": ["endpoints"],
      "wordCount": 2400
    }
  }
}

Collaboration Behaviors

In Comprehensive Mode

Proactive contributions:

• "I notice you're using Redis but there's no ADR explaining why—should I document that decision?"
• "Your error handling is sophisticated. This deserves explanation in architecture docs."
• "The authentication flow is non-standard. Users will have questions—let me address them."

Challenge assumptions:

• "You said 'standard depth' but you have 47 endpoints. That needs Deep documentation."
• "This 'getting started' guide assumes too much. Your users won't know X."
• "Three examples aren't enough here. The concept is complex."

Surface insights:

• "Your git history shows you refactored auth 3 times. That's an ADR waiting to be written."
• "These three files handle all business logic but aren't documented at all. Gap."
• "You have inline JSDoc but it contradicts what's in the markdown docs. Consistency issue."

In Quick Mode

Efficient execution:

• "Updating 3 files based on changes in commit a3f2b1c..."
• "2 new endpoints detected, adding to api.md..."
• "Removed documentation for deleted legacy-auth endpoint..."

Flag concerns:

• "Warning: Manual changes detected in architecture.md, preserving your edits"
• "Note: Health score dropped from 92 to 85 due to new undocumented features"

Quality Standards

The skill maintains high documentation quality through:

Clarity

• Concepts explained before they're used
• Technical terms defined on first use
• Examples precede or immediately follow concepts
• Progressive disclosure (simple → complex)

Completeness

• Every public-facing element documented
• Edge cases and gotchas addressed
• Troubleshooting for predictable failures
• Examples for common use cases

Accuracy

• All facts verified against code
• Examples tested and working
• Links validated
• No outdated information

Accessibility

• Diagrams include alt text
• Links have descriptive text
• Heading hierarchy is logical
• Code snippets have language labels

Consistency

• Terminology used uniformly
• Tone maintained throughout
• Formatting standardized
• Structure predictable

Special Features

Architecture Decision Records (ADRs)

Captures WHY decisions were made:

The skill identifies decision points through:

• Major dependency additions (git history)
• Non-standard architectural patterns (code analysis)
• Framework/library choices
• Flagged commits with "decision" language
• Your explicit identification

Each ADR documents:

• Context (what was the situation?)
• Decision (what did we decide?)
• Rationale (why this choice?)
• Consequences (trade-offs accepted)
• Alternatives considered

Living, Tested Examples

Examples are runnable code in /examples:

• Actually work (not pseudocode)
• Cover common use cases
• Include test script to verify they run
• Referenced from documentation
• Maintained as code evolves

Troubleshooting Database

Two flavors:

• User troubleshooting: Common errors, how to fix them
• Developer troubleshooting: Debugging guides, edge cases, gotchas

The skill seeds initial content in Comprehensive Mode, grows it in Quick Mode as you add real troubleshooting content.

Documentation Map

For large projects, a navigation guide:

• Different learning paths (beginner/advanced)
• How docs connect to each other
• What to read when
• Visual/textual navigation aid

Mermaid Diagrams

The skill generates code-based diagrams:

• Architecture diagrams
• Sequence diagrams
• Entity-relationship diagrams
• State diagrams
• Flowcharts

All version-controllable, all render in GitHub.

Documentation Hosting Integration

Detects if you're using:

• GitHub Pages
• ReadTheDocs
• GitBook
• MkDocs

Generates appropriate config files and optimizes structure for static site generation.

Reference Documents

Load contextually when needed:

• references/project-types-guide.md — How to document different project types
• references/documentation-patterns.md — Common documentation patterns and structures
• references/quality-standards.md — Detailed quality criteria and examples
• references/manifest-spec.md — Technical specification for .doc-state.json
• references/depth-levels-guide.md — Standard vs. Deep explained with examples
• references/health-score-formula.md — How health score is calculated
• references/adr-guide.md — Writing effective Architecture Decision Records

Templates

Available in templates/:

• README-template.md — Progressive disclosure README structure
• api-docs-template.md — API endpoint documentation
• component-docs-template.md — Component/module documentation
• troubleshooting-template.md — Troubleshooting guide structure
• contributing-template.md — Contributing guidelines
• documentation-map-template.md — Navigation guide template

Success Criteria

Documentation is complete when:

✓ Health score ≥ 85 ✓ All critical and important debt resolved ✓ Examples run without errors ✓ Links validate successfully ✓ Accessibility check passes ✓ Tone is consistent throughout ✓ You're confident someone could use your project from the docs alone

Key Reminders

• Reader-first always: Every decision serves the reader
• Git-aware: Smart incremental updates, not regeneration
• Transparency: You see all analysis and decisions
• Preserve quality: Don't overwrite good manual edits
• Test examples: Generate test scripts for working code
• Track health: Health score trends over time
• Prioritize debt: Not all debt is equal
• Comprehensive without exhaustive: Document what matters