English | 中文

🔍 code2doc

Reverse-engineer business logic from code. Generate professional documentation covering the entire software delivery lifecycle.

Documentation drift is entropy, not negligence. Code has CI/lint/compiler errors as feedback loops — documentation has none, so it inevitably rots. code2doc is the solution: treat code as the single source of truth, and continuously generate or calibrate documentation.

---

ClawHub: https://clawhub.ai/suifei/code-to-doc-generator

Skillhub:https://skillhub.cn/skills/code-to-doc-generator

Coverage

---

What It Does

Given any code repository (web, backend services, mobile, desktop), code2doc can reverse-engineer and generate 12 standard software delivery documents:

Phase	Document Type	Target Audience
Initiation	BRD (Business Requirements) · MRD (Market Requirements)	Executives, Investors
Requirements	PRD (Product Requirements)	Product Managers, Tech Leads
Architecture	HLD (High-Level Design) · DDD (Domain-Driven Design)	Architects, Tech Leads
Implementation	LLD (Low-Level Design) · Coding Guidelines	Developers
Verification	Test Documents (BDD/SIT/E2E/UAT)	QA Engineers
Operations	Ops Manual · Admin Quick Reference · SLI/SLO Monitoring	Ops Staff, SRE
Release	CI/CD Pipeline Documentation	DevOps Engineers

Core Features

• Code as Truth: All content is extracted directly from code. Inferences are marked with 〔INFER〕, code facts are traced with 〔FACT｜file:line〕
• Audience-Adaptive: Ops manuals avoid technical jargon; HLD/LLD retain precise engineering language; BRD/MRD are fully business-oriented for executives
• Reverse Sync: Run Reverse Sync Mode after code changes to automatically detect documentation drift and calibrate
• Rigorous Structure: Ops documents require 🔗 prerequisite dependency chains + 📊 impact tracking matrices (including underlying algorithm/strategy names)
• Any Tech Stack: Go · Python · Node.js · Java · Swift · Kotlin · Flutter · React · Vue · Electron

---

Installation & Usage

Claude Code (Recommended)

Claude Code is Anthropic's official CLI tool with native skill format support.

Option A: Project-Level (current project only)

# Run in project root
mkdir -p .claude/skills
git clone https://github.com/suifei/code2doc .claude/skills/code2doc

Add to your project's CLAUDE.md:

## Skills

@.claude/skills/code2doc/SKILL.md

Option B: Global (all projects)

mkdir -p ~/.claude/skills
git clone https://github.com/suifei/code2doc ~/.claude/skills/code2doc

Add to ~/.claude/CLAUDE.md:

## Skills

@~/.claude/skills/code2doc/SKILL.md

Usage:

# Generate an operations manual
> Generate operations manual

# Generate HLD
> Write HLD for me

# Reverse sync — code changed but docs are out of date
> Sync documentation

---

OpenCode

OpenCode supports the same skill directory structure as Claude Code.

# Clone to OpenCode's skills directory
mkdir -p ~/.opencode/skills
git clone https://github.com/suifei/code2doc ~/.opencode/skills/code2doc

Register in ~/.opencode/config.toml:

[[skills]]
path = "~/.opencode/skills/code2doc/SKILL.md"

Or via the OpenCode UI: Settings → Skills → Add from path

Usage: Type /code2doc in the chat or directly describe your documentation needs.

---

Cursor

Cursor loads rules from the .cursor/rules/ directory (.mdc format).

# In project root
mkdir -p .cursor/rules

# Convert SKILL.md to Cursor rules format
curl -fsSL https://raw.githubusercontent.com/suifei/code2doc/main/SKILL.md \
  -o .cursor/rules/code2doc.mdc

Or manually:

1. Open Cursor → Cursor Settings → Rules for AI
2. Paste the contents of SKILL.md into the User Rules area
3. Request a document type directly in the chat:

Generate HLD for this project

Global rules (all projects):

Cursor Settings → General → Rules for AI → Paste SKILL.md content

---

Cline / RooCode (VSCode Extensions)

Cline and RooCode support custom System Prompts.

1. Open VSCode → Click the Cline/RooCode icon in the sidebar
2. Click the settings gear → Custom Instructions
3. Paste the full contents of SKILL.md
4. Click save

Usage:

# Enter directly in the Cline chat
Generate operations manual

# Or specify modules
Generate LLD for user and channel modules

---

Continue.dev

Continue supports custom slash commands via config.json.

Edit ~/.continue/config.json:

{
  "customCommands": [
    {
      "name": "code2doc",
      "description": "Generate project documentation from code",
      "prompt": "{{{ input }}}\n\n---\n{{{ read_file '.claude/skills/code2doc/SKILL.md' }}}"
    }
  ]
}

Install the skill file:

mkdir -p .claude/skills
git clone https://github.com/suifei/code2doc .claude/skills/code2doc

Usage:

/code2doc Generate PRD for this project

---

Windsurf

Windsurf loads instructions via .windsurfrules files or global rules.

Project-level:

# Create rules file in project root
curl -fsSL https://raw.githubusercontent.com/suifei/code2doc/main/SKILL.md \
  -o .windsurfrules

Global:

1. Windsurf Settings → AI Rules
2. Paste the contents of SKILL.md

Usage:

Generate HLD

---

Any LLM Tool (Universal)

Use the contents of SKILL.md as a System Prompt. Works with any tool that supports custom System Prompts:

• ChatGPT / OpenAI API: Paste into System Message
• API calls:

import anthropic

with open("SKILL.md", "r") as f:
    skill_content = f.read()

client = anthropic.Anthropic()
message = client.messages.create(
    model="claude-opus-4-5",
    max_tokens=8192,
    system=skill_content,
    messages=[
        {"role": "user", "content": "Generate operations manual"}
    ]
)

---

File Structure

code2doc/
├── SKILL.md                    # Main skill file (4-round workflow + quality checks)
└── references/
    ├── document-types.md       # Section structure templates for 12 document types
    ├── document-structure.md   # Formatting specs for ops/admin manuals
    ├── analysis-dimensions.md  # Seven analysis dimensions (ops documents)
    ├── exploration-strategy.md # Code exploration strategy (scripts + high-info locations)
    └── reverse-sync.md         # Reverse sync protocol (documentation drift calibration)

---

Workflow

User specifies document type
       │
       ▼
  Round 1: Identify project type & entry points
  (Read root dir → Identify tech stack → Existing docs → Initial module list)
       │
       ▼
  Round 2: Structural scan & information skeleton extraction
  (Routes/menus → i18n strings → Data models → Permission rules → Business rules)
       │
       ▼
  Round 3: Per-module deep analysis
  (Analysis dimensions vary by doc type: steps/state impact/business rules/dependencies...)
       │
       ▼
  Round 4: Document generation
  (Template skeleton → Fill content → FACT/INFER annotations → Save)
       │
       ▼
  Post-completion: Documentation sync rule
  (Search related docs → Choose one: Consistent/Update/Adjudicate)

---

Reverse Sync Mode

Code changed but docs didn't keep up? Use reverse sync:

Code changed but docs are out of date

The skill automatically enters Practice-as-Truth Reverse Sync mode:

1. Build artifact inventory (distinguish PRACTICE vs DESCRIPTION)
2. Full code review, extract FACTS + OBSERVATIONS
3. Diff comparison (Consistent / Drifted / Missing / Zombie)
4. Write back revisions
5. Generate OBSERVATIONS.md (issue tracking)

---

Quality Standards

Check	Standard
Audience Fit	Language depth matches target readers (ops ≠ dev ≠ QA)
Annotation Completeness	Inferences have 〔INFER〕, code facts have 〔FACT｜file:line〕; zero annotations = execution failure
Flowcharts	Multi-step operations / decision branches include Mermaid diagrams
Unambiguous Terminology	Technical terms are translated for the audience; BRD/MRD/PRD must not use raw terms like API/Token
Prerequisite Dependencies	Each chapter of ops documents has 🔗 dependency chains
Impact Matrix	Each chapter of ops documents has 📊 matrix with strategy names in column 4
No "TBD" Promises	Uncertain items are resolved (pick one of three options); "follow up later" is prohibited

---

License

MIT © suifei