freu-ai/freu-cli — reverse-engineered prompt

Build me a Python command line tool called Freu CLI that helps AI agents stop wasting tokens on repeated browser tasks.

I want a user to run a learn command with an optional goal like “star a GitHub repo by URL”, then do the task once in Chrome. The tool should capture their clicks, typing, and navigation through a Chrome extension bridge. When they stop recording, it should use an LLM to turn the session into a reusable skill folder with a readable SKILL.md, one JSON command file per action, and logs for each learning stage.

The saved skill should be easy for agents like Claude Code, Codex CLI, Cursor, OpenClaw, or Hermes to load and run later. It should replay the workflow without sending the whole page DOM back to the model every time. Please make element matching resilient by using semantic context around elements instead of brittle CSS selectors. If the goal is to retrieve information, compile that as an output too.

Include a simple quickstart, tests, and look up current docs online if needed.