Start at the top and follow the sequence. Each guide has one outcome, commands you can run, a verification step, and the point where human judgment belongs.
Open an h5i browser session, drive a page by @ref handle, then audit the fail-closed request log the engine wrote before any bytes moved.
Create your first h5i sandbox, run an agent inside it, inspect the diff and execution record, then export or apply the result.
Fetch an untrusted pull request into a detached h5i box, build and exercise it, inspect denied activity, and export a review bundle.
Define an h5i profile with an explicit isolation tier, filesystem grants, default-deny networking, and resource limits, then verify it.
Run a dev server and browser inside an h5i box, watch it through a loopback-only viewer, and safely transfer control from agent to human.