Harness²
Open-source (Apache 2.0) framework from Google Research for recursive agent harnessing. The Propose–Probe–Compose loop refines an agent's prompts, tools, memory, skills, subagents, and guardrails without changing model weights, and it keeps lessons for future tasks. It supports OpenCode and Codex executors. The repository says it is not an officially supported Google product.
Get notified when this tool is tested
No logged test yet. This joins the same daily newsletter, and it does not assign a tier.
What's new · October 21, 2026Harness², from Google Research, is an open-source (Apache 2.0) framework for recursive agent harnessing. Its Propose–Probe–Compose loop refines an agent's prompts, tools, memory, skills, subagents, and guardrails without changing model weights, and it keeps lessons for future tasks. It supports OpenCode and Codex executors. Google reports gains of up to 11.5 points on WorkBuddy-Bench. That figure is a vendor claim. The repository says it is not an officially supported Google product. Agent test pending.Source
No affiliate link on file. Outbound goes to github.com. Identity checked October 21, 2026. Scores were not part of that check.
Verdict
Pending agent test
Use-it-for, skip-if, and the one-line verdict publish with the test log. This page will not guess them.
- Use it for
- Not assigned
- Skip it if
- Not assigned
- Tier
- Unranked
Scores
One 0–10 score from the logged run. Time and cost are measured separately. None of that is set.
Run score
—
out of 10 · Pending agent test
- OutputNot scored
- SpeedNot scored
- CostNot scored
- ChecksNot scored
Pros and cons
Pros
Pending agent test. Nothing is listed until a run is logged.
Cons
Pending agent test. Nothing is listed until a run is logged.
Best for
Not a tested recommendation. These are the jobs this tool is filed under.
Test log
Test pending: not run. Harness² is an open-source framework. No run was logged.
- Status
- Not run
- Date
- —
- Time to first output
- —
- Cost
- —
- Result
- —
- Evidence
- None
Queued task. Run a fixed multi-step workflow with one tool call and one failure case. Log setup time and cost.
FAQ
What is the agent-tested verdict for Harness²?
Test pending: not run. Harness² is an open-source framework. No run was logged.. Harness² is in the catalog with its official URL, category, and pricing model. No score, pros, cons, or tier have been published.
How much does Harness² cost?
The pricing model on file is Open source. A dollar price is not stored until someone checks the vendor page again.
Do sponsors outrank this tool?
No. A sponsored slot is labeled and sold separately from tiers. It cannot move this listing.