Summary

This project offers an autonomous, self-improving product system that analyzes performance reports to identify and implement the highest-priority fixes. It automates continuous enhancement for developers and product teams, delivering ready-to-review pull requests.

How It Works

The system orchestrates an AI-driven workflow: daily reports are analyzed to pinpoint priorities, which are then planned into granular, executable tasks. An iterative execution loop implements these tasks, runs configurable quality checks (e.g., tests, type checks), commits successful changes, and updates task status. This repeats until completion or a maximum iteration limit, culminating in a pull request for human review. The philosophy emphasizes compounding small, correct changes.

Quick Start & Requirements

• Installation: Clone the repo and run ./install.sh /path/to/your/project to integrate into an existing codebase.
• Prerequisites: Requires Amp CLI or Claude Code, jq, gh CLI (authenticated), and a configured LLM provider (Anthropic, OpenRouter, AI Gateway). Browser testing needs npm install -g agent-browser.
• Configuration: Customize config.json for report directories, quality checks, and iteration limits.
• Running: Use ./scripts/compound/auto-compound.sh --dry-run for simulation or ./scripts/compound/auto-compound.sh for execution.
• Automation: macOS launchd can schedule runs, requiring careful PATH setup.
• Docs: Key scripts and skills are