Vercel Plugin Eval

This is a QA harness for testing the Vercel plugin end to end. It launches real Claude Code sessions in WezTerm, tails debug logs, and verifies all eight hook types fire correctly with proper deduplication. The skill walks you through creating timestamped test projects, monitoring skill injection counts against atomic claim files in tmpdir, and checking that PostToolUse validation catches real problems like API key bypasses. It includes a scenario design guide targeting different coverage zones, from AI chat apps to monorepos, with call outs for the eight hard to trigger skills that need explicit mentions. You write results to a coverage matrix showing which hooks and skills fired across sessions. Useful if you're maintaining the plugin or debugging why certain skills don't inject when expected.