ai

AI — Centralized Claude Code Plugin

A centralized Claude Code plugin containing agents, skills, hooks, and coding rules. Use this across all your projects without duplicating setup.

Quick Start

1. Clone to a permanent location

git clone https://github.com/judigot/ai.git ~/ai

2. Create a shell alias

Add to your .bashrc, .zshrc, or shell config:

alias cc='claude --plugin-dir ~/ai'

Reload your shell:

source ~/.bashrc

3. Use in any project

cd /path/to/any/project
cc

All your agents, skills, hooks, and rules are now available!

Directory Structure

ai/
├── .claude-plugin/
│   └── plugin.json           # Plugin manifest (required)
├── agents/                   # Agent definitions (.md files)
│   ├── code-reviewer.md      # Git-based PR review
│   ├── task-master.md        # Worktree task execution
│   ├── multitasker.md        # Parallel worktree management
│   └── agentic-workflow.md   # Multi-agent coordination
├── skills/                   # Specialized skills (subdirectories)
│   ├── lint-master/
│   │   └── SKILL.md          # Multi-tool linting workflow
│   └── test-master/
│       └── SKILL.md          # Testing infrastructure
├── hooks/
│   └── hooks.json            # SessionStart, PreToolUse, Stop hooks
├── commands/                 # Slash commands (.md files)
├── scripts/                  # Helper scripts
├── settings/                 # Personal settings (compartmentalized)
│   └── rules.md              # Coding rules loaded on session start
├── CLAUDE.md                 # Plugin entry point
└── README.md

How It Works

Personal Settings (Compartmentalized)

Your personal coding rules are stored in settings/rules.md, separate from ~/.claude. This provides:

• Single source of truth: One repository for all global settings
• Version control: Track changes to your rules over time
• Portability: Same settings across all machines

The hooks/hooks.json loads your rules on every session start:

{
  "SessionStart": [
    {
      "matcher": "*",
      "hooks": [
        {
          "type": "command",
          "command": "cat ${CLAUDE_PLUGIN_ROOT}/settings/rules.md"
        }
      ]
    }
  ]
}

Global Plugin Loading

The --plugin-dir flag tells Claude Code to load this plugin for any project:

claude --plugin-dir ~/ai

This applies your agents, skills, hooks, and settings globally without copying files to each project.

Available Agents

Agent	Purpose
code-reviewer	Git-based PR review with enterprise-grade analysis
task-master	Execute single task in a worktree autonomously
multitasker	Sprint orchestrator - creates worktrees and spawns parallel subagents
agentic-workflow	Multi-agent coordination architecture

Compatibility: All agents work with both Claude Code and OpenCode.

Available Skills

Skill	Purpose
lint-master	Multi-tool linting workflow (ESLint > Oxlint > Biome)
test-master	Testing infrastructure and implementation

Sprint Modes

Mode A: Parallel Worktrees (Recommended)

Give the multitasker a list of tasks - it creates worktrees and spawns subagents:

"I need to work on auth, dashboard, and API features"

The multitasker will:

1. Create .worktrees/feat-auth, .worktrees/feat-dashboard, .worktrees/feat-api
2. Spawn task-master subagents via Task tool (in parallel)
3. Report results when done

Philosophy: Git is the only source of truth. No Context.md, no .state files, no metadata.

Mode B: Sequential Ralph Loop

For tasks that touch the same files or need strict sequencing:

./ai/scripts/ralph/ralph.sh 10

Start here: ai/scripts/ralph/

Combining with Project-Specific Config

Local projects can have their own settings that extend the global ones:

my-project/
├── CLAUDE.md                 # Project-specific rules
├── .claude/
│   └── settings.local.json   # Project-specific settings
└── agents/                   # Project-specific agents (optional)

Claude Code loads in this order:

1. Global plugin (from --plugin-dir) ← This repository
2. Project CLAUDE.md
3. Local .claude/ settings

Updating

cd ~/ai
git pull

Changes apply to the next Claude Code session.

Adding New Components

New Agent

Create a new .md file in agents/:

---
name: my-agent
description: Use this agent when [conditions]. Examples:

<example>
Context: [Situation]
user: "[Request]"
assistant: "[Response]"
<commentary>
[Why this agent triggers]
</commentary>
</example>

model: inherit
color: blue
tools: ["Read", "Write", "Bash"]
---

You are an expert at...

[Agent instructions here]

New Skill

Create a new subdirectory in skills/ with a SKILL.md file:

skills/
└── my-skill/
    └── SKILL.md

The skill file follows the same format as agents.

New Command