feat: additional task tool

[mkultraWasHere]

Summary

• Added TaskOutputBuffer toolset providing a stateful output buffer for accumulating task results
• Agents can incrementally build up output across multiple steps using output() and retrieve results with get_output()

Impetus: for certain evals we now require the model to specifically mark the output that is used to determine score (versus just checking any output supplied throughout entire trajectory). Thus with this tool, all the output content is stored on the tool, then grabbed later by eval scorers to verify.

---

Generated Summary:

• Added TaskOutputBuffer class for stateful output management, allowing agents to accumulate results incrementally.
• Introduced two methods within TaskOutputBuffer :
  • output: Adds content to an internal buffer and returns the total count of outputs.
  • get_output: Retrieves all previously saved outputs in order.
• Updated docstrings for clarity and adherence to markdown formatting.
• Imported PrivateAttr from pydantic for managing internal attributes securely.
• Enhanced error handling by integrating catch=True in tool methods, improving robustness during execution.
• Overall, these changes improve task management and output reporting capabilities within the agent framework.

This summary was generated with ❤️ by rigging

[Copilot]

Pull request overview

This PR adds a new TaskOutput toolset for stateful output management and fixes markdown formatting in existing docstrings.

• Adds a new TaskOutput class that provides an output buffer for agents to accumulate task results
• Fixes markdown formatting in give_up_on_task docstring for better rendering
• Imports necessary dependencies including PrivateAttr from pydantic and Toolset/tool_method from base

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.