Summary
Harbor 0.22.0 can now run Terminal-Bench and other agent evaluations inside Vercel Sandbox microVMs, letting teams execute isolated benchmark runs on Vercel's ephemeral compute.
What changed
Harbor 0.22.0 added Vercel Sandbox as a supported eval environment, enabling Terminal-Bench and other Harbor evals to run in isolated microVMs on Vercel.
Why it matters
Reproducible, isolated environments are the bottleneck for evaluating coding and terminal agents; wiring Vercel Sandbox into Harbor gives agent builders on-demand microVMs for benchmark runs without standing up their own sandbox infrastructure.
Evidence excerpt
Run Terminal-Bench and other Harbor evals on Vercel Sandbox — Harbor 0.22.0 adds Vercel Sandbox as an eval environment for running benchmarks in isolated microVMs.