Проверка и создание тестов для eval

testing

ensure-test-coverage is a Claude Code skill that reviews existing tests and creates missing ones for a single evaluation, ensuring thorough test coverage across all meaningful code paths. It operates in structured phases: first scanning `src/inspect_evals/<eval_name>/` for functions decorated with `@task`, `@solver`, `@scorer`, and `@tool`, as well as dataset patterns and sandbox usage; then running `run_autolint.py` and `pytest --cov` with `--runslow` to measure line-level coverage and identify gaps. Rather than enforcing a fixed threshold, the skill judges whether every important branch — calculations, parsing, error handling — would be caught if broken. It's the right choice when asked to review, add, or improve tests for an eval; for fixing specific failing CI tests, ci-maintenance-workflow should be used instead.