Hi my name is Quinn, and I have been testing and using LLMs for various coding projects for about a year now. And at this point I run all my agents overnight or for even days at a time. And I’ve sunk countless hours and tokens into coming back to a job that's half done with issues and bugs. That's why I built Krites. It blocks the edit tools on the paths you name, runs your tests when your agent tries to stop and blocks the stop until they pass (up to 3 tries by default), and prints a receipt with /krites:receipt that contains commands, exit codes, diff and commit. The biggest known limits: Bash can still write protected files, and an agent with a shell could forge a receipt. Those and the full docs are in the repo’s README: https://github.com/krites-dev/kr... Install it with: claude plugin install krites@anthropic-plugin-directory. It's free and open source, Claude Code CLI only, needs Node 20 or later. Comment if you have any questions, suggestions or similar experiences, or hop on a quick call to tell me about when your coding agents failed https://cal.com/quinnl/krites!

Krites

Don't ask your agent to run the tests. Force it.

Don't ask your agent to run the tests. Force it.