What problem does it solve? Running plugin benchmark evals locally in WezTerm terminal panes limits parallelism, requires manual verification, and leaves no permanent deploy URLs. This Skill moves eval scenarios into ephemeral Vercel Sandbox microVMs so builds, browser verification, and deployments run in parallel with structured scoring. ## Core Features & Use Cases - 3-Phase Eval Pipeline: Phase 1 builds the app with Claude Code, Phase 2 uses agent-browser to verify user stories and fix issues, Phase 3 links to the vercel-labs team and deploys with up to 3 build-error retries. - Structured Haiku Scoring: Each phase is scored via claude -p --json-schema --model haiku, returning structured JSON for build completeness, per-story pass/fail, and deploy success. - Skill Coverage Tracking: Skills injected by the plugin are polled and re-extracted after every phase, producing coverage reports comparing expected vs. claimed skills per scenario. - Use Case: Generate a JSON file of app-building scenarios with user stories, then run run-eval.ts --scenarios-file to execute them across up to 10 parallel sandboxes and get a markdown report with deploy URLs. ## Quick Start Run the benchmark sandbox eval with my scenarios file using bun run .claude/skills/benchmark-sandbox/run-eval.ts --scenarios-file /tmp/my-scenarios.json.
