Winzyss-code/sentinel-sfa — reverse-engineered prompt

Reverse engineered prompt

Build me a small tool called Sentinel SFA that lets an AI use websites through a numbered text menu instead of screenshots. I want it to open a real browser, scan the page for visible links, buttons, and inputs, print a menu like [1] search box and [2] submit, and then let me do things like click 2 or type 1 hello. Start with Google working really well, but make it usable on normal sites too. It should have a terminal mode where I can open a page, print the menu once, or give a plain English goal and let an LLM try a short sequence of actions.

Also include a simple desktop app so I can see runs, failures, sessions, and token savings live, and expose the same browser controls through an MCP server for tools like Claude Desktop or Cursor. Please include basic tests, examples, and clear setup instructions so I can run it locally and try a demo right away. Look up current docs online if you need to.

Want more depth? Deep Reverse