fix: handle multiple EndTurn tools with Union type for allow_fail

[zzstoatzz]

summary

fixes infinite loop and "multiple endturn tools" warning when using allow_fail=True

root cause

• when a task has allow_fail=True, it creates two endturn tools: MarkTaskSuccessful and MarkTaskFailed
• previous code (agent.py:186) set output_type=type(None) when >1 endturn tool detected
• this caused the agent to not validate its output as an EndTurn, leading to infinite loop with repeated "Multiple EndTurn tools detected" warnings

solution

• use Union[Tool1, Tool2] as output_type when multiple endturn tools exist
• this allows pydantic_ai to properly validate the agent's response as one of the union types
• the model can now choose between marking the task successful or failed

testing

• added 3 new tests in tests/basic/tasks/test_tasks.py::TestAllowFail:
  • test that tasks with allow_fail complete without infinite loop (regression test)
  • test that allow_fail creates both MarkTaskSuccessful and MarkTaskFailed tools
  • test that multiple EndTurn tools result in Union output type
• manually tested with repros - tasks can now properly succeed OR fail
• 61/62 existing tests pass (1 failure due to API connection issue, not our change)

related

• builds on PR feat: enable true concurrent execution for independent tasks #1229 which fixed concurrent task execution

🤖 Generated with Claude Code

[Copilot]

Pull Request Overview

This PR fixes an infinite loop issue that occurred when using tasks with allow_fail=True, which creates multiple EndTurn tools. The solution replaces the previous workaround of setting output_type=type(None) with a proper Union type that allows the model to choose between success and failure tools.

• Replaced type(None) with Union[Tool1, Tool2] for multiple EndTurn tools to enable proper validation
• Removed the warning about multiple EndTurn tools since this is now properly handled
• Added comprehensive test coverage for the allow_fail functionality

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
src/marvin/agents/agent.py	Updated logic to use Union type for multiple EndTurn tools instead of type(None)
tests/basic/tasks/test_tasks.py	Added new test class with 3 tests covering allow_fail functionality and regression testing

Comments suppressed due to low confidence (1)

tests/basic/tasks/test_tasks.py:1

• The test accesses private attribute _output_type and doesn't verify the Union contains the expected tool types. Consider using public APIs if available, and assert that the Union contains the specific MarkTaskSuccessful and MarkTaskFailed tool types.

import enum

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}