Hi!
I noticed that during one iteration of a roborev refine loop, it tried to commit changes but was stopped by a linting pre-commit hook. Then the entire command exited. In another case, something was wrong with a post-checkout hook, and when roborev tried to create a new worktree and failed, the command exited.
Could we instead provide that failure as feedback to the agent to fix?