L/small changes 12

[lorenss-m]

Note

Medium Risk
Moderate risk: changes core evaluation helpers to be async (including subprocess execution) and extends CLI task discovery/import behavior, which could affect existing integrations and task loading in real projects.

Overview
Adds a significantly expanded hud.native evaluation toolkit: graders are now fully async, BashGrader uses async subprocesses with timeout_seconds, and a new Grade.gather() runs multiple graders/subscores in parallel. This also introduces LLMJudgeGrader (rubric-based LLM judging) plus built-in answer comparison/normalization helpers, and exports them via hud.native.

Improves CLI task collection to better match real project layouts: supports importing a directory as a package (running its own discovery), recursively finds **/task.py, and adds project-root sys.path injection to fix cross-module imports; includes extensive new tests for these edge cases.

Updates agent console output to show a tool discovery table at init and a structured per-step tool call/result summary at INFO, and refreshes docs for the new graders API plus minor doc link/typo fixes.

^{Written by Cursor Bugbot for commit 2d5b407. This will update automatically on new commits. Configure here.}

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
hud	🟢 Ready	View Preview	Apr 1, 2026, 1:05 AM

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

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