Harbor integrates Docker Sandboxes to stop agents from gaming their own evals

joecole · x · 2026-10-16

The Harbor framework now integrates Docker Sandboxes to address a subtle eval problem: agents that edit files and run commands can game their own evaluations. With the integration, every eval task runs in an isolated microVM, with network policies enforcing fair, reproducible runs.

Original post →

More from coding & agent

coding & agent channel →