I built this because I wanted to use agentic tools like Claude Code, but I was worried about running claude with --dangerously-skip-permissions on my local machine.

Axon is a Custom Resource Definition (CRD) that spins up ephemeral Pods for every agent task. It creates a safe blast radius.

The Dogfooding: To test it, I used Axon to develop Axon. Over this weekend, I ran agents that read the repo, implemented features/fixes, and opened PRs. I merged 47 of them. It feels like having a team of juniors working in parallel.

The repo is written in Go. I'd love feedback on the controller architecture or the CRD design.