ep local-test

[xzrderek]

Note

Adds a local-test CLI to run evaluation tests locally or in Docker with extra flags, updates EvaluationRow.created_at to UTC, tweaks upload prompts, and adds comprehensive tests.

• CLI:
  • New local-test command in eval_protocol/cli.py and cli_commands/local_test.py to run a selected evaluation test via pytest.
    • Resolves --entry (path or path::function) or uses selector; enforces single selection.
    • Auto-detects Dockerfile; runs in Docker if present (or on host with --ignore-docker).
    • Supports --docker-build-extra and --docker-run-extra; mounts project/logs and maps user IDs.
  • Command dispatch wired into main CLI.
• Upload UX (cli_commands/upload.py): change prompts to say "Select this test?" and "Enter the number to select:".
• Models (models.py): EvaluationRow.created_at now defaults to datetime.now(timezone.utc) (UTC timestamp).
• Tests: add tests/test_cli_local_test.py covering host/Docker execution, multiple Dockerfiles error, extra flag passing, selector behavior, and path normalization.

^{Written by Cursor Bugbot for commit 9b476dc. This will update automatically on new commits. Configure here.}

[cursor]

Bug: Non-interactive --yes fails multiple tests.

When --yes is used without --entry and multiple tests exist, _prompt_select returns all tests (because non_interactive=True), causing the check if len(selected) != 1 to always fail. The error message doesn't guide users to use --entry, making the --yes flag unusable in multi-test scenarios. The function should either fail earlier with a helpful message about requiring --entry, or handle the non-interactive case differently.