__                 _
 _ ____ ____ _ __ _ _ _  ___ _ _   / /_ _ __ _ ___ _ _| |_ ___
| '  \ V / _` / _` | ' \/ _ \ ' \ / / _` / _` / -_) ' \  _(_-<
|_|_|_\_/\__,_\__, |_||_\___/_||_/_/\__,_\__, \___|_||_\__/__/
              |___/                      |___/

AI Workflow

Shared configuration and conventions to bootstrap AI coding assistants. One repo, one command — every tool gets the same rules, skills, and MCP servers.

Supported Tools

Tool	Rules	Skills / Agents	Root File	MCP Config
Claude Code	.claude/rules/	.claude/skills/ · .claude/agents/	CLAUDE.md	.mcp.json
OpenCode	.opencode/rules/	.opencode/skills/ · .opencode/agents/	AGENTS.md	opencode.json
Cursor	.cursor/rules/	.cursor/skills/ · .cursor/agents/	—	.cursor/mcp.json
Codex	—	.agents/skills/	AGENTS.md	.codex/config.toml

Compatibility

Requirement	Details
Node.js	>= 18
macOS	12 Monterey and later
Linux	Any modern distribution
Windows	Not supported (WSL works)

Getting Started

Bootstrap

npx mvagnon-agents ../my-project

Upgrade

Run from within a bootstrapped project (requires .mvagnon-agents/):

cd my-project
npx mvagnon-agents upgrade

Updates generic and dep-sensitive files in .mvagnon-agents/ to match the latest package version. Project-sensitive files are never overwritten. A manifest.json tracks which items are auto-updatable.

Manage

Add tools, rules, skills and agents to an existing project:

cd my-project
npx mvagnon-agents manage

Interactive Walkthrough

The script prompts you to:

1. Select target tools — one or more (Claude Code, OpenCode, Cursor, Codex)
2. Pick resources — rules, skills and agents in a single menu, with category hints
3. Add to .gitignore? — yes to ignore tool directories, no to track everything

AI Workflow → /Users/you/projects/my-app

◆ Select target tools
│ [x] Claude Code
│ [x] Cursor
│ [ ] OpenCode
│ [ ] Codex

◆ Pick resources
│ [x] project.md                            (rules · project-sensitive)
│ [x] nestjs-hexagonal-architecture.md      (rules)
│ [ ] react-hexagonal-architecture.md       (rules)
│ [ ] fastapi-hexagonal-architecture.md     (rules)
│ [x] documentation-writer                  (skills)

◆ Add agents configuration to .gitignore?
│ ○ No

◇ Claude Code Setup
│ Rules:      2 linked
│ Skills:     1 linked
│ CLAUDE.md:  linked
│ .mvagnon-agents/.mcp.json.example: copied
│ .gitignore: not modified

◇ Cursor Setup
│ Rules:      2 linked
│ .mvagnon-agents/mcp.json.example: copied
│ .gitignore: not modified

◇ Next Steps
│ 1. Copy config files and replace the API key placeholders:
│    Claude Code: cp .mvagnon-agents/.mcp.json.example .mcp.json
│    Cursor:      cp .mvagnon-agents/mcp.json.example .cursor/mcp.json
│ 2. Modify the following project-sensitive files to fit your project:
│    - rules/project.md
│ 3. Add rules, skills, agents, MCPs or plugins based on your needs for each tool.

◇ Available Commands
│ npx mvagnon-agents manage    Add tools, rules, skills or agents to an existing project
│ npx mvagnon-agents upgrade   Sync generic resources with the latest package version

Done

Project Structure

config/
├── rules/
│   ├── project-sensitive/
│   │   └── project.md                           # Per-project rules (editable)
│   └── generic/
│       ├── nestjs-hexagonal-architecture.md      # NestJS hexagonal
│       ├── react-hexagonal-architecture.md       # React hexagonal
│       └── fastapi-hexagonal-architecture.md     # FastAPI hexagonal
├── skills/
│   ├── generic/
│   │   ├── documentation-writer/                 # Documentation writing skill
│   │   └── docs-lookup/                          # External library docs lookup
│   └── dep-sensitive/
│       └── Playwright MCP/
│           └── ui-check/                         # UI verification via Playwright
├── agents/                                       # Reserved (empty)
├── AGENTS.md                                     # Master rules for all agents
├── claudecode.settings.json                      # Claude Code MCP config
├── opencode.settings.json                        # OpenCode MCP config
├── cursor.mcp.json                               # Cursor MCP config
└── codex.config.toml                             # Codex MCP config (TOML)

bootstrap.mjs                                     # Interactive setup script
lib/
├── manage.mjs                                    # Manage subcommand
└── manifest.mjs                                  # Manifest read/write helpers

How It Works

Files are organized into three types:

• project-sensitive — meant to be edited per project. Never overwritten on upgrade.
• generic — kept in sync with the package. Updated on upgrade.
• dep-sensitive — like generic, but only relevant when a specific dependency is present. Organized as dep-sensitive/{dependency}/{skill-name}/ in the config directory. Updated on upgrade.

All items are stored flat in .mvagnon-agents/<category>/. A manifest.json tracks each item's type so the upgrade mechanism knows which items to sync.

Tool directories (.claude/rules/, .cursor/rules/, etc.) contain relative symlinks pointing into .mvagnon-agents/. This means all tools share the same source files.

.mvagnon-agents/
├── AGENTS.md                                 # Root file
├── manifest.json                             # Tracks item types for upgrade
├── rules/
│   ├── project.md                            # Project-sensitive (editable)
│   ├── nestjs-hexagonal-architecture.md      # Generic (updated via upgrade)
│   ├── react-hexagonal-architecture.md
│   └── fastapi-hexagonal-architecture.md
└── skills/
    ├── documentation-writer/                 # Generic
    └── docs-lookup/

Configuration Storage

MCP config files are created from .example templates in .mvagnon-agents/. Copy them to the correct path and replace the API key placeholders with your actual keys. Config files are always gitignored.

Recommended Claude Code Hooks

Claude Code supports PostToolUse hooks that automatically lint and format files after every Edit or Write. Add the following to your ~/.claude/settings.json to enable auto-formatting across all projects:

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Edit|Write",
        "hooks": [
          {
            // JS/TS: ESLint fix then Prettier (sequential to avoid race conditions)
            "type": "command",
            "command": "jq -r '.tool_input.file_path | select(test(\"\\\\.([jt]sx?|mjs|cjs)$\"))' | xargs -I {} sh -c 'npx eslint --fix \"$1\" && npx prettier --write \"$1\"' _ {}",
          },
          {
            // Other files (md, json, css, etc.): Prettier only
            "type": "command",
            "command": "jq -r '.tool_input.file_path | select(test(\"\\\\.([jt]sx?|mjs|cjs)$\") | not)' | xargs npx prettier --write --ignore-unknown",
          },
          {
            // Python: Ruff check then format (sequential to avoid race conditions)
            "type": "command",
            "command": "jq -r '.tool_input.file_path | select(test(\"\\\\.py$\"))' | xargs -I {} sh -c 'uvx ruff check --fix \"$1\" && uvx ruff format \"$1\"' _ {}",
          },
        ],
      },
    ],
  },
}

Why sequential? Claude Code runs hooks in parallel. Tools that write to the same file (e.g. ESLint + Prettier, or Ruff check + Ruff format) must be chained with && inside a single hook to prevent race conditions.

Prerequisites

• Node.js (for npx eslint and npx prettier)
• uv (for uvx ruff) — install via curl -LsSf https://astral.sh/uv/install.sh | sh
• Project-level ESLint and Prettier configs (.eslintrc.* / eslint.config.*, .prettierrc*) for JS/TS rules to take effect

Manual Installation

If you prefer not to use the bootstrap script:

1. Copy the config/ folder contents to your project
2. Rename files according to your target tool (see Supported Tools table)
3. Replace API key placeholders in config files with your actual keys
4. Update .gitignore as needed