claude plugin eval now loads every case. Grader files lacked type/weight front matter, so all cases failed with invalid case.yaml: graders: Required.evals/<skill>/<case>/prompt.md + graders/*.md (149 cases); the 65 flat .md cases are converted. Case names are unique (<skill>-<case>).skill-fired grader (tool_used: Skill, with-only arm).fixture.sh + case.yaml scaffolds (run with --scaffold); code-review cases embed target code; postgresql cases supply MCPg results inline.not-covered-japan, not-covered-de-overlay, Cuba as the unpacked-jurisdiction example).docs/evals.md and evals/README.md describe the layout, --scaffold, and rubric guidance.pr-review-merge-ready-clean does not reliably fire its skill. See RELEASE-NOTES-v0.18.0.md.typescript-code-review — 11-domain scored TypeScript review (config, report template, domain refs, evals).nodejs-code-review — 11-domain scored Node.js service/API review.react-code-review — 11-domain scored React/Next UI review.ui-system-review when the ask is design-system tokens/components across platforms.regulatory-compliance-applicability-scan: Stage I geographic close-out, force/thin-pack re-checks, EU overlay article-level depth (DE FR IT ES NL AT SE PL).RELEASE-NOTES-v0.16.0.md.regulatory-compliance-applicability-scan first skill release cut:
ui-system-review. Plugin 0.14.0.See git history.