benchmark-sandbox - Run benchmark evaluations in Vercel Sandboxes
Runs three-phase benchmark scenarios in ephemeral Vercel Sandboxes, evaluates outputs, extracts hook artifacts, and generates coverage reports.
Tags
Updated: 2026-10-15Capabilities
Typical Inputs
Typical Outputs
What this skill does
- Provision ephemeral sandboxes
- Install evaluation tools
- Run build evaluations
- Verify user stories
- Deploy evaluated applications
- Score phase outputs
- Extract hook artifacts
- Generate coverage reports
Inputs
- Scenario definitions
- Local vercel-plugin
- Anthropic credentials
- Vercel credentials
- Evaluation options
Outputs
- Sandbox evaluation results
- Hook artifacts
- Structured phase scores
- Coverage reports
- Public sandbox URLs
- Deployed application URLs
Requirements
- Bun runtime
- Vercel Sandbox access
- Node 24 sandbox runtime
- Claude Code installation
- Haiku model support
- Vercel team permissions
- Vercel OIDC setup