gsd-eval-review

Auditing AI evaluation coverage involves reviewing a completed agent workflow against the declared AI-SPEC.md evaluation strategy to verify implemented gates. This process identifies missing or weak elements and generates a scored verdict with an EVAL-REVIEW.md remediation plan.

An AI-SPEC.md evaluation strategy defines the intended quality gates for an AI delivery phase. You need to review it retroactively after phase completion to confirm the actual agent workflow matched the declared strategy and to document any coverage gaps.

Generating a remediation plan for missing evaluation gates requires running a retroactive audit on the finished phase. The audit produces a scored pass/fail verdict and outputs an EVAL-REVIEW.md file detailing explicit gaps and concrete next steps to fix them.

Phase-based delivery reviews are designed for iterative agent workflows, applying retroactive audits to completed phases. They preserve existing workflow gates while scoring evaluation coverage and producing actionable remediation plans for subsequent runs.

Retroactive evaluation coverage audits are limited to reviewing completed phases against the declared AI-SPEC.md strategy. They cannot fix missing gates in past workflows, but they do produce a prioritized EVAL-REVIEW.md remediation checklist for future iterations.