What problem does it solve? Testing a Claude Code plugin's skill injection and hook behavior requires real interactive sessions, since print-mode runs never fire hooks or create files. This Skill provides the exact commands, monitoring queries, and verification checks to run reliable end-to-end evals of the Vercel plugin. ## Core Features & Use Cases - Interactive Eval Launch: Spawns WezTerm panes running real Claude Code sessions with the plugin installed via npx add-plugin, using timestamped project directories. - Injection Monitoring: Checks skill claim directories and debug logs to count injected skills, hook firings, and PostToolUse validation catches. - Code Verification: Greps generated projects for correct patterns such as withWorkflow config, AI Gateway usage, Gemini image models, and AI Elements components. - Use Case: After updating a skill's prompt signals, launch three parallel eval sessions with natural-language prompts, confirm the expected skills were injected, inspect the generated Next.js code, then write a coverage report before releasing. ## Quick Start Launch a benchmark eval by creating a timestamped directory under ~/dev/vercel-plugin-testing, installing the Vercel plugin with npx add-plugin, and spawning a WezTerm pane running an interactive Claude session with a natural-language product prompt.
