Arc agi 3 env

[eligotts]

Description

Based off of discussions, model has 4 tools: bash, arc_step (progresses arc game), arc_reset (reset arc game), stop_episode (quit...) . on every turn, the model sees its full assistant message + tool response for previous 3 turns + the contents of a persistent MEMORY.md file in its sandbox.

go to here to create an ARC key: https://docs.arcprize.org/

export it in your terminal and you can see the full run on their dashboard

also currently printing tool calls to: arc_tool_calls.jsonl for convenience to see what the model is doing

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

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