eval-quality-workflow

Use the fix mode of this workflow, which walks through every item in EVALUATION_CHECKLIST.md top to bottom and refactors the evaluation code to comply. It records progress in NOTES.md, UNCERTAINTIES.md, and a final SUMMARY.md under agent_artefacts/<eval_name>/fix.

Use the review mode, which assesses the evaluation against the agent-runnable checks in EVALUATION_CHECKLIST.md without modifying files. It documents each issue with the standard violated, location, recommended fix, and a GitHub-ready comment prefixed with (Agent).

Fix mode actively refactors the evaluation to pass each checklist item and checks items off, while review mode only assesses compliance and reports issues. Both modes create the same artefact structure but under fix/ and review/ subfolders respectively.

No, the evaluation report is explicitly out of scope. The workflow notes in SUMMARY.md that producing evaluation reports requires the separate /eval-report-workflow skill, though it still runs initial smoke tests to confirm the evaluation executes.

Do not use it to review all evaluations against a single code quality standard; that scenario belongs to the code-quality-review-all skill. This workflow targets one evaluation against the full checklist.