harbor-glovebox

Run a Harbor trial inside a glovebox microVM instead of a shared-kernel container. Each trial boots one hardware-isolated VM behind glovebox's outgoing-traffic filter, and the filter's records land beside the trial's other artifacts.

Status

The harbor-trial live check drives one trial's environment calls on a real microVM in CI, then runs harbor run with Harbor's oracle agent over a one-task dataset through this environment. No run over a published dataset with a model-backed agent has used it yet.

Use

Install the released package:

pip install harbor-glovebox

To run the unreleased code on main instead, install all three packages from the repository at one revision. That needs git on the machine. Set R to the same tag or full commit for all three, since two people who install main on different days get different code. --upgrade --force-reinstall replaces a released glovebox-driver or glovebox-hosts that already satisfies the version pins:

R=git+https://github.com/AlexanderMattTurner/agent-glovebox.git@<tag-or-full-commit>
pip install --upgrade --force-reinstall \
  "glovebox-hosts @ $R#subdirectory=glovebox-hosts" \
  "glovebox-driver @ $R#subdirectory=glovebox-driver" \
  "harbor-glovebox @ $R#subdirectory=harbor-glovebox"

Then run a trial:

harbor run --dataset <dataset> --agent claude-code \
  --env harbor_glovebox.environment:GloveboxEnvironment

The host needs what any glovebox launch needs. glovebox sandbox preflight checks it, and each trial runs that check before it builds anything.

Users

A command runs as the user Harbor names, else the task's [agent].user or [verifier].user. With neither named, it runs as glovebox's de-privileged account, never as the image's own USER. Before that account's first command, it takes over the image's working directory and what the image's USER owned in its home. A task that names the image's own account keeps those files as they are. Every command inherits the image's ENV, as docker exec would carry it. Harbor names root itself for its own setup steps, such as installing the agent. Every command, root's included, sends its traffic through the in-VM filter and then the host filter outside the VM.

What it refuses, and why

• network_mode = "public". glovebox has no unfiltered mode. Use allowlist and list every host the task needs. Put the agent's model endpoint in Harbor's extra_allowed_hosts.
• Wildcard and IP allowlist entries. The filter matches hostnames only. Harbor rejects such a task before it starts.
• A bare top-level name as an allowlist entry, such as com. A public suffix with a dot, such as github.io, is accepted: like every entry, it reaches that one host alone.
• A download larger than 1 GiB. The guest's archive arrives in the host's memory whole, so the transfer fails rather than grow without bound.
• A task image that already ships /logs, /tests, /solution or /harbor. The guest's root is read-only, so each of Harbor's fixed trees is a link onto the guest's writable /tmp disk, and the image build refuses to replace a tree the image carries.
• Multi-container tasks (a docker-compose.yaml in the task's environment directory).

storage_mb is ignored, as Harbor's own Docker environment ignores it.