What problem does it solve? Running plugin benchmark evals locally in WezTerm panes limits parallelism, requires manual verification, and leaves no permanent deploy URLs. This Skill moves eval scenarios into ephemeral Vercel Sandboxes (Firecracker microVMs) so benchmarks run in parallel with automated browser verification, deployment, and structured scoring. ## Core Features & Use Cases - 3-Phase Eval Pipeline: Phase 1 builds the app with Claude Code, Phase 2 uses agent-browser to walk through user stories and fix failures, Phase 3 links to the vercel-labs team and runs vercel deploy with up to 3 build-error retries. - Skill Coverage Tracking: Tracks skill injections across all three phases via hook claim directories and poll history, then compares claimed skills against expected skills per scenario. - Structured Haiku Scoring: Each phase is scored by a separate claude -p --json-schema --model haiku pass returning structured JSON for build completeness, per-story pass/fail, and deploy success. - Use Case: Generate a JSON file of app scenarios with user stories, run run-eval.ts with --scenarios-file, and receive a markdown report with skill coverage, verification results, and permanent *.vercel.app deploy URLs. ## Quick Start Run the eval pipeline by executing bun run .claude/skills/benchmark-sandbox/run-eval.ts with a scenarios JSON file after linking the project with npx vercel link --scope vercel-labs.
