Metadata-Version: 2.4
Name: quibbler
Version: 0.1.2
Summary: AI monitoring for Claude Code agents
Requires-Python: >=3.12
Requires-Dist: claude-agent-sdk>=0.1.0
Requires-Dist: fastapi>=0.117.1
Requires-Dist: requests>=2.30.0
Requires-Dist: uvicorn>=0.37.0
Description-Content-Type: text/markdown

# Critic

Independent AI monitoring for Claude Code agents. Critic watches your agent's work through hook events and provides quality feedback to prevent common mistakes.

## Installation

```bash
pip install critic
```

## Usage

### 1. Start the monitoring server

```bash
critic-server [port]
```

Default port is 8081.

### 2. Configure Claude Code hooks

Add to your Claude Code project configuration:

```json
{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "critic-hook {session_id} {source_path}"
          }
        ]
      }
    ],
    "UserPromptSubmit": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "critic-display"
          }
        ]
      }
    ]
  }
}
```

### 3. Work normally

Critic monitors in the background and writes feedback to `.critic-messages.txt` when it has observations. The feedback is displayed at the next user prompt.

## How it works

1. **critic-hook** forwards Claude Code hook events to the monitoring server
2. The server runs a monitoring agent that watches for quality issues
3. When issues are detected, the monitor writes feedback to `.critic-messages.txt`
4. **critic-display** shows the feedback to the agent on the next interaction

## Environment Variables

- `ANTHROPIC_API_KEY`: Required for the monitoring agent (same as Claude Code)
- `CRITIC_MONITOR_BASE`: Override server URL (default: `http://127.0.0.1:8081`)
- `CLAUDE_MONITOR_SKIP_FORWARD`: Set to `1` to disable hook forwarding

## License

MIT
