cathrynlavery/repo-atlas ? reverse-engineered prompt
Reverse engineered prompt
Build me a Claude Code skill that creates a persistent context system for any git repo, so I can drop it into a codebase and quickly understand what’s going on without digging around forever.
I want to be able to run one command like /repo-atlas and have it generate a docs/atlas/ folder with a repo map, directory tree, entry points, file stats, changelog, and a set of structured docs that explain the architecture, domain model, important flows, state sources, gotchas, and testing notes. It should also update CLAUDE.md with a simple agent on ramp that helps with a plan first, verify second workflow, and add build targets so this can be checked in CI too.
Please make it work in a practical way for common languages and build tools, and keep it simple enough that it can run from a Python script with standard library only if possible. If you need to look up current Claude Code skill docs online, do that.
Have a live product UI? Try website reverse