firstbatchxyz/watchmen — reverse-engineered prompt

Reverse engineered prompt

GitHub

Build me a local developer helper called watchmen. I want it to sit quietly behind tools like Claude Code, Codex, Cursor, OpenCode, and pi.dev, read my past and new coding sessions from the normal local folders, and learn the repeated things I do. When it sees useful patterns, it should create reusable skill bundles and keep CLAUDE.md and AGENTS.md updated for each repo so the next coding agent session starts smarter.

Add a simple watchmen init wizard that asks for an LLM provider, saves the key locally, imports sessions, lets me choose projects, shows a cost estimate before analysis, then sets up a background service and a local dashboard. The dashboard should run on my machine, show active repos, skill usage, tool errors, project health, next actions, and a per project before and after impact view. Also include a doctor command, privacy minded local storage, and basic plugin support for Claude Code and Codex. Look up current docs online if you need to.

Want more depth? Deep Reverse