reticlehq/reticle — reverse-engineered prompt
Reverse engineered prompt
Build me a dev only, localhost only tool called Reticle that works like a proof layer for AI coding agents. I want an agent to make a code change, automatically open the running app, inspect what is really happening inside the program, and then return a simple pass or fail verdict with evidence and the exact file and line to fix.
It should catch the kinds of bugs a screenshot or basic browser check misses, like mock data still being used, wrong APIs, hidden 500s, broken flows, bad store state, console problems, request count issues, and React render or signal problems. Make it especially good for React apps.
I also want it to be easy to plug into coding agents through an MCP server and a simple setup command, so tools like Claude Code or Cursor can use it right away. Support replaying recorded flows deterministically, keep reruns fast and cheap, allow multiple agents to verify in parallel on one browser, and keep everything private with no telemetry. Look up current docs online if you need to.
Want more depth? Deep Reverse