oh-my-dev

DIP (Dual-phase Isomorphic Documentation) protocol generator. Generate P1 root maps, P2 module docs, and P3 file headers automatically.

npm i oh-my-dev    # install
npx oh-my-dev      # auto-detect and install

Quick Start

Step 1: Install

npx oh-my-dev --dest ./skills

This installs the skill to ./skills/oh-my-dev/.

Step 2: Use in your AI agent

/init                # Bootstrap full documentation (SCAN → P1 → P2 → P3 → VERIFY)
/scan                # Analyze project structure
/p1                  # Generate root agent config (AGENTS.md/CLAUDE.md)
/p2                  # Generate module documentation
/p3                  # Add P3 headers to files
/verify              # Check doc-code consistency

Supported AI Tools

Cursor

npx oh-my-dev --dest ~/.cursor/skills

Then restart Cursor and enable Agent Skills in Settings → Rules.

Claude Code

# Project-local (recommended)
npx oh-my-dev --dest ./.claude/skills

# Global (all projects)
npx oh-my-dev --dest ~/.claude/skills

Codex CLI

# Project-local
npx oh-my-dev --dest ./.agents/skills

# User-wide
mkdir -p ~/.agents/skills && npx oh-my-dev --dest ~/.agents/skills

Gemini CLI

npx oh-my-dev --dest ~/.gemini/skills

Then restart Gemini CLI and type /skills to verify.

VS Code (GitHub Copilot)

npx oh-my-dev --dest ./.github/skills

Trae

# Trae China
npx oh-my-dev --dest ~/.trae-cn/skills

# Trae International
npx oh-my-dev --dest ~/.trae/skills

All Commands

Documentation Bootstrap

/init        # Full bootstrap: SCAN → P1 → P2 → P3 → VERIFY
/scan        # Analyze codebase structure
/document    # Generate DESIGN.md from existing code

Layer Generation

/p1          # Generate root AGENTS.md / CLAUDE.md
/p2          # Generate module AGENTS.md (member lists)
/p3          # Add P3 headers to source files

Quality Control

/verify      # Check doc-code consistency
             # Reports: FATAL (must fix) / SEVERE (fix this session)

DIP Protocol Overview

DIP = Dual-phase Isomorphic Documentation

Three fractal layers that keep docs in sync with code:

| Layer | File | Content | |-------|------|---------| | P1 | AGENTS.md (root) | Global topology, stack overview, key abstractions | | P2 | AGENTS.md (module) | Member list: files, responsibilities, technical points | | P3 | File header | WHO exports, FROM deps, TO consumers, HERE role |

P3 Header Example:

/**
 * [WHO]: Provides buildSystemPrompt(), BuildSystemPromptOptions
 * [FROM]: Depends on config, skills, tools
 * [TO]: Consumed by agent runtime, SDK
 * [HERE]: core/prompt/system-prompt.ts - prompt building
 */

Workflow Examples

New project:

/init          # Full bootstrap with 5 context questions
               # → Creates P1 + P2s + P3s + VERIFY

Existing project:

/scan          # Analyze structure and gaps
/init          # Bootstrap with detected context
/verify        # Check consistency

After structural changes:

/verify        # Check what broke
/p3            # Add missing headers
/verify        # Re-check

Design Decision Priority

When in doubt:

P1 (root) > P2 (module) > P3 (file)

Docs must stay isomorphic with code. Run /verify after any structural change.

Options

--dest <dir>       # Destination directory (required)
--repo <owner/repo> # Download from GitHub instead
--ref <branch>     # Git ref (default: main)
--help             # Show help

Examples:

npx oh-my-dev --dest ./skills
npx oh-my-dev --dest ~/.claude/skills --repo O-Pencil/skills --ref v1.0.0

Language Intelligence

Skill content: English (SKILL.md + references)

Generated output: Adapts to user's language

• Section headers, descriptions → user's language
• Code symbols, file paths → always original
• Build commands → always original

FORBIDDEN

| Code | What | |------|------| | FATAL-001 | Modify code without updating docs | | FATAL-002 | Skip P3 on modified files | | FATAL-003 | Delete file without updating P2 | | FATAL-004 | New module without P2 |

License

MIT · GitHub