Description

Allow for prompt rewriting in setup_state. If the returned state contains the prompt field, then overwrite prompt with the returned value.

This enables environments that, for example, contain grounding information in the initial prompt.

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Test improvement

This could be considered a breaking change in the sense that if somebody was already mutating the prompt field in setup_state, then the system behavior would change. I don't know why somebody would be doing that as it would have no effect, but it's possible.

Testing

• All existing tests pass when running uv run pytest locally.
• New tests have been added to cover the changes

Checklist

• My code follows the style guidelines of this project as outlined in AGENTS.md
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• Any dependent changes have been merged and published

Additional Notes

This resolves a problem for a use-case that I have. Since verifiers doesn't currently support generators for datasets, my other options are:

• Regenerate my dataset for each run
• Decorate rollout in my subclass to modify prompt before super.rollout is called. However, I need to store metadata in state that corresponds to the modification to prompt. So I would then also need to mutate info to pipe that data through into the state object, which seems kind of lame.
• Completely override rollout with a custom implementation

But this seemed like the best approach and I thought that others would find this change useful.