Human-Eval Review Skill

Turn analysis output into a single self-contained HTML review file. A human opens it in any browser (no server, no build step, no internet), works through the items, and clicks Download Results to export their judgments as JSON. Progress is autosaved in the browser, and a previous results file can be re-loaded to resume.

This generalizes a family of "did the model get it right?" review tasks into one builder.

When to Use

• You ran a matching / retrieval / classification / extraction / translation step and a human needs to confirm, correct, score, or rank the output before it's trusted.
• You want a clean review UI you can hand to a collaborator who has only a browser.
• You're building a labeled gold set from model proposals.

Critical: Things Claude Won't Know Without This Skill

The output is ONE file, rendered from embedded data