benchmark-sandbox

This Skill automates Vercel plugin benchmark testing by orchestrating a three-phase pipeline—BUILD, VERIFY, and DEPLOY—within isolated Vercel Sandboxes. It uses Claude Code and agent-browser to execute scenarios, followed by structured scoring via a 'haiku' model for comprehensive performance and integration testing.

Running Vercel plugin evals in ephemeral microVMs is the best way to test without affecting your local setup. This Skill provisions isolated Vercel Sandboxes that execute benchmark scenarios independently, ensuring robust testing and performance analysis in clean, reproducible environments.

Structured scoring in Vercel sandbox benchmarking utilizes a 'haiku' model to generate structured JSON scores after each pipeline phase. This mechanism evaluates the BUILD, VERIFY, and DEPLOY stages individually, providing comprehensive performance and integration testing metrics for the Vercel plugin.

Yes, the Vercel sandbox evaluation pipeline supports automated browser testing during the VERIFY phase. It leverages agent-browser alongside Claude Code to execute verification steps, ensuring comprehensive integration testing before the final DEPLOY phase occurs within the isolated microVM environment.

Yes, you can test how the Vercel plugin integrates with new Vercel features by running a suite of benchmark scenarios in clean Vercel Sandboxes. This approach provides reproducible microVM environments specifically designed for robust integration and performance analysis across Vercel deployments.

Vercel Sandboxes for plugin performance testing rely on ephemeral microVMs and a rigid three-phase BUILD, VERIFY, DEPLOY pipeline orchestrated by Claude Code. This architecture is specifically tailored for Vercel plugin benchmark scenarios and may not suit testing workflows outside this ecosystem.