Agent behavior clone for browser using, targeting general GUI using and distributed trajectory collecting. This project allows users to record their browser tasks, which are then distilled into reusable skills for AI agents like Claude. It provides a local, single-user environment for skill creation, benefiting developers and power users looking to automate or integrate browser workflows into AI applications without extensive manual coding.

How It Works

The system captures user interactions via a browser extension, sending these "traces" (containing intent and events) to a local server. A background Python pipeline then atomizes these traces, classifies user actions into distinct capabilities (e.g., "login," "search"), buckets them by domain, and uses an LLM to distill them into structured skill definitions (SKILL.md) and usage guides (TRACE_GUIDE.md). This automated distillation process transforms raw user behavior into reusable, LLM-interpretable skills.

Quick Start & Requirements

• Installation: Clone the repository. Copy config.example.env to .env.local and set SF_LLM_KEY with your LLM API key. Run ./scripts/start.sh or python entry/main.py to start the local server. For the extension, navigate to extension/, run pnpm install && pnpm build, and load the dist/chrome-mv3 folder as an unpacked extension in Chrome.
• Prerequisites: An LLM API key (Anthropic Messages API or OpenAI-compatible gateway). pnpm for building the extension. Python runtime.
• Resource Footprint: Local server and distillation process. Data stored locally under data/.
• Links: Control panel at http://127.0.0.1:8099/. Documentation for Claude Desktop setup is available in docs/claude-desktop-setup.md.

Highlighted Details

• Local-First Architecture: All processing and data storage occur locally, with only LLM API calls leaving the user's machine.
• Automated Skill Distillation: Leverages LLMs to convert recorded browser interaction traces into structured, reusable skills (SKILL.md).
• Dual Integration: Skills can be automatically installed for Claude Code or exported as .zip files for Claude Desktop.
• Extensibility: Supports using OpenAI-compatible LLM gateways by setting SF_LLM_BASE.

Maintenance & Community

No specific details regarding contributors, sponsorships, community channels (like Discord/Slack), or roadmaps were found in the provided README snippet.

Licensing & Compatibility

The license type and any compatibility notes for commercial use or closed-source linking are not explicitly stated in the provided README snippet.

Limitations & Caveats

Skills generated are instructional (SKILL.md) and require a separate browser automation engine (like Playwright) to be configured and executed, particularly for Claude Desktop integration. The project is described as "not a benchmark." The extension build process requires pnpm.