task helper

[lorenss-m]

Note

Medium Risk
Introduces a new Task.run() execution path that wraps run_eval() and is now used by Chat.send(), so any behavior differences could affect evaluation tracing/reward propagation. Also changes file/path handling for coding tools on Windows, which may have platform-specific edge cases.

Overview
Adds Task.run() as a one-liner helper to execute a single task (accepting a model string or agent instance), internally creating the eval context via run_eval() and propagating ctx.reward onto the returned Trace.

Refactors Chat.send() and multiple docs/examples to use task.run(...)/env(...).run(...) instead of manually creating an eval context and running the agent.

Improves Windows compatibility in coding tools by giving a Windows-specific absolute-path error in EditTool.validate_path() and falling back to direct file I/O in read_file_async/write_file_async when shell utilities/heredocs aren’t available.

^{Written by Cursor Bugbot for commit e27a71d. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

[cursor]

Test patches class method but instance is MagicMock

Medium Severity

The patch targets hud.eval.task.Task.run on the Task class, but dummy_task is a MagicMock, not a Task instance. When Chat.send() calls task.run(...), task is the MagicMock returned by model_copy, so task.run(...) invokes MagicMock's auto-generated run attribute — not the patched Task.run. This returns a non-awaitable MagicMock, causing await task.run(...) to raise TypeError. The patch has no effect on the actual code path being tested.