Suggesting instead of always making fixes with roborev fix/refine #623
Description
I would to have an option to make fix and refine less confident, especially for the skill versions.
It seems like there are at least three levels of autonomy for agents to make fixes:
- Suggest all fixes, asking users to explicitly approve all of them
- Do easy fixes, but only make suggestions in cases where the desired fix is ambiguous
- Do all fixes
Right now, it seems like roborev defaults to (3), which is a little too confident for me. I would rather the agent asks for guidance than makes a best guess.
I could try to fix this with prompting in my own CLAUDE.md, but perhaps we could update the default prompting to be a bit less assertive? Alternatively, we might add options for explicitly choosing a different "autonomy level"