Autonomous mode / wrapper for Claude Code?

At covenance.ai we've recently open sourced our internal tool doing exactly that. https://github.com/ikamensh/kodo - it will use a backend, such as claude code, cursor or codex (or free google-cli) and an LLM orchestrator. It either works towards a goal you specify, or you run it with --improve flag and it tests your software, looks for bugs, brainstorms how to improve architecture / test coverage. Try it :)

There is multiple ways this can be achieved, you can use ralph loops which is a claude code plugin that makes the agent repeatedly run until all items are complete.

Also you can write an .md file and explain how you want things to be done in details, maybe you want to write failing tests before coding maybe you want to run tests in a certain manner etc ..., basically steering the AI to write more code and run for longer.

Telling the AI to "fix this" or "do this" without clear instructions makes it lazy where it just wants to do its thing and stop.