Run evaluations with Inspect AI

This page describes the planned workflow for running Inspect AI evaluations in Capsem and points to the automation available today. The Inspect AI integration is not installed or published yet; its qualification is tracked in #342.

This guide will cover choosing an evaluation image, configuring authorized inputs, running a small task, inspecting results alongside VM activity, and cleaning up interrupted runs. Installation and run commands will be added after the published integration passes that workflow.

For current automation, use the SDK. For retained activity, see Investigate a VM.