Workspace sandbox and repository environment

The idea

Worker commands run in a sandbox (local or Docker) confined to the checkout, with a scrubbed environment, allowed binaries, limits and one path guard. Repository setup and readiness are recorded and reused.

The problem
Commands could reach secrets, other paths or the network, and setup was done again each time.
Why it matters
Secrets never enter the sandbox, pushes to the main line are guarded, and old workspaces are cleaned up within quota.
Where it stands
In production as of

A step-by-step replay of this idea is not published yet. The library adds one when a recorded run shows it.

  • Private execution host

    A root service on the host runs sandbox commands under signed, bounded grants from an issuer. It listens on a private socket and keeps a journal to clean up the Docker objects it owns.

  • Own accounts, execution profiles and payers

    Hosted work uses API keys. A subscription runs only through the vendor's product or the person's own runner. Executor, model, host and payer inherit from the platform through org, role and worker.

Turn the numbers into shipped work.

Crewshal runs these choices for you: a persistent AI worker with memory and rules, on your Claude and Codex subscriptions.