Claude Code plugin
Before Claude adds code to a file, coderot’s plugin checks whether the file is one to be careful with. It reads your repo’s git history on your machine and names three kinds of file:
- Files agents keep growing: most of their recent PRs came from AI agents. Claude puts new code in a new file next to them.
- Hotspots: high cognitive complexity and many commits in the last 90 days. Claude prefers a new module next to them.
- Rising: the files whose complexity rose most in about 30 days. Claude keeps new code out of them where it can.
These are the same rules as the dashboard’s hotspots and worst changes. Nothing is uploaded and no sign-in is needed: the skill runs npx -y @coderot-ai/cli hotspots --json in your repo and reads its output.
Install
claude plugin marketplace add coderot-ai/claude-plugin
claude plugin install coderot@coderot-ai
Claude uses the skill on its own before it edits or refactors existing files. You can also run it as /coderot:hotspots. The source is on GitHub.
It needs Node.js and the repo’s full history: a shallow clone gives a short list. On a large repo the first run takes a few minutes, so Claude runs it once per session.
Without the plugin
To keep the list in the repo for every agent, not only Claude Code, run this and paste the block it prints into CLAUDE.md or AGENTS.md:
npx -y @coderot-ai/cli hotspots --md
The plugin uses that block instead of running the CLI while it’s less than a week old. Run the command again now and then to refresh it.