Agent Zero vs Harness²

Agent Zero

Self-hosted under the MIT licence, with nine releases in the 90 days to 12 October 2026 and breaking changes listed in the release notes. The agent runs shell commands in its container with no approval step, the web interface has no login by default, and five CVEs from the past year have no published GitHub advisory.

Harness²

Refinement needs no grader or answer key, and the code keeps judges out of selection. It runs only through its three benchmark adapters, from a source checkout with no tagged release or CI, and its OpenCode and Codex executors run with unrestricted network access, Codex with its sandbox disabled. The benchmark gains are the authors' claims.

Agent Zero

1. Set a UI login and password in Settings before publishing the port or opening a tunnel. The default has none
2. Fill allowed_users, allowed_numbers or sender_whitelist before enabling a channel. An empty list accepts every sender
3. Enable the Infection Check plugin when the agent reads mail, web pages or MCP results. It ships disabled
4. Run v2.12 or later. It fixes a Telegram webhook authentication bypass and a WhatsApp media path traversal
5. Call POST /api_message with the X-API-KEY header. The token changes when the UI login or password changes

Harness²

1. Don't plan to call Harness² as a service. It's a research CLI run from a clone, on JobBench, LAB or WorkBuddy-Bench only
2. Ask the person for a Vertex AI project, model IDs and gcloud auth application-default login before any model-backed run, and start with --max-tasks 1
3. Run executors behind an egress firewall and block huggingface.co and github.com after setup, as the docs advise, so rubrics can't be fetched
4. Write .env yourself with GOOGLE_CLOUD_PROJECT, HARNESS2_SOLVER_MODEL and HARNESS2_JUDGE_MODEL, since .env.example isn't shipped, and never commit it
5. Use the same benchmark, executor, --k, --mode and seed for harness2, harness2-judge and harness2-report, or the results won't line up