gsd-eval-review

To audit AI evaluation coverage, the skill compares a completed agent workflow phase against the evaluation plan in AI-SPEC.md. It detects missing checks or mismatched criteria and produces an EVAL-REVIEW.md report with a score, verdict, and remediation plan.

An evaluation coverage audit verifies whether an executed AI phase actually implemented the intended evaluation strategy. You need it to reliably identify blind spots and correct them before shipping, ensuring retroactive quality assurance for agent workflows.

Yes, you can target a specific phase number for a quality gate review. The skill reviews the most recent completed phase by default, but allows you to specify a phase number for focused iteration and retroactive quality assurance.

You generate a remediation plan by running the skill for the last completed phase. It produces an EVAL-REVIEW.md report containing detected gaps and a concrete, actionable remediation plan while preserving workflow gates.

Yes, you need an AI-SPEC.md file to perform an evaluation coverage audit. The skill compares completed phase behavior against the evaluation plan defined within AI-SPEC.md to detect missing checks and generate the remediation report.

The best way to fix gaps in AI QA phase reviews is to generate an EVAL-REVIEW.md report. This provides a score, a verdict, detected coverage gaps, and an actionable remediation plan to correct blind spots before shipping.