Claude Code Hooks & Rules

注意

LLMにこのREADMEを作らせたので、全然間違っていることを言っている可能性があります。

概要

A collection of opinionated hooks and best‑practice rules for Claude Code. The initial release ships with stop_hook.py, a plug‑and‑play Stop‑Event hook that:

• Echoes Claude’s last assistant message back to it
• Appends 「これであってる？」 to prompt clarification
• Fires a macOS notification with the echoed text
• Unblocks the Stop event once, using stop_hook_active to avoid infinite loops

---

Table of Contents

1. Demo
2. Requirements
3. Quick Start
4. Configuration
5. How It Works
6. Customising the Hook
7. Project Roadmap
8. Contributing
9. License

---

Demo

$ claude --debug
> …Claude answers…
# Notification ✨ pops up (macOS)
> assistant: <previous answer>

これであってる？

---

Requirements

Component	Version
Python	3.8 +
Claude Code	v0.33 +
OS	macOS (for notifications) Linux/Windows run fine—notification call is skipped

---

Quick Start

git clone https://github.com/1ulce/.claude.git
cd .claude/hooks
chmod +x stop_hook.py

1. Open Claude Code → Settings → Hooks
2. Add the snippet below to settings.json (or use the UI):

{
  "hooks": {
    "Stop": [
      {
        "matcher": "",          // run on every Stop
        "hooks": [
          {
            "type": "command",
            "command": "/absolute/path/to/hooks/stop_hook.py"
          }
        ]
      }
    ]
  }
}

3. Start a session – the hook fires automatically whenever Claude reaches a Stop Event.

---

Configuration

Flag / Section	Purpose
stop_hook_active check	Prevents infinite trigger loops
osascript call	Sends a notification (macOS only)
assistant_last parsing	Extracts latest assistant reply

Feel free to fork & tweak for your workflow.

---

How It Works

1. Claude signals a Stop decision.

2. The hook reads the conversation transcript (JSONL).

3. The latest assistant chunk is extracted & echoed.

4. A macOS notification is displayed via AppleScript.

5. The hook prints

   {"decision":"block","reason":"<echoed>\n\nこれであってる？"}

   which Claude treats as the next user turn.

6. On the subsequent Stop, stop_hook_active is already true; the script exits silently, letting the conversation end normally.

---

Customising the Hook

• Disable notifications – set NOTIFY=false env var or comment out the send_notification call.
• Change the suffix – edit SUFFIX = "\n\nこれであってる？".
• Different platforms – swap the osascript block for notify-send (Linux) or PowerShell (Windows).

---

Project Roadmap

• 🔧 Pre‑prompt rules & context validators
• 🪝 Additional hook templates (MessageRewriter, Guardrails, Metrics)
• 🧩 Schema‑driven plugin loader
• ✅ GitHub Actions CI (flake8, mypy, unit tests)

Contributions welcome—see below!

---

Contributing

1. Fork the repo & create your branch: git checkout -b feature/my-awesome-hook
2. Commit with conventional commits style.
3. Run pre-commit run --all-files (config coming soon).
4. Open a PR; describe what & why clearly.
5. Be nice ❤️ – we review fast!

---

License

Released under the MIT License – see LICENSE for details.

---

Acknowledgements

• ChatGPT & Claude community for inspiration
• Original discussion: [link to chat or issue]