HumanEval: say which helper files of the workspace the check does not run #20

Priority

medium

Components

• app/nullius (backend)
• app/web (routes)
• templates and static files
• docs and the site
• catalogue (benchmarks, agents, capabilities)
• the standard (ports, kits, relay, runs)
• CI, workflows and skills

Today

HumanEval's check runs solution.py alone, as the upstream's check_correctness runs one completion. An agent that writes a helper beside it (helpers.py) and imports it fails the problem, and the trial's detail carries only the upstream's words, failed: No module named 'helpers'. A reader cannot tell a wrong solution from a right one split over two files.

Wanted

The trial's detail names each file of the workspace that solution.py imports and the check does not run, so the failure reads as what it is. The verdict is unchanged: one function, one file, as published.

Acceptance

• A solution.py that imports a module or package found in the workspace gets detail.not_tested listing each (helpers.py, pkg/).
• A solution.py importing only the standard library has no such field.
• The verdict is still the upstream's check_correctness on solution.py alone.

Dependencies and related

none