Google open-sources Harness², a recursive agent harnessing framework

Google Cloud AI Research has open-sourced Harness², a framework that improves the harness around a frozen model — prompts, tools, memory and orchestration — without training weights. A single refinement step improved all 4 model-harness setups on all 3 benchmarks, with GLM-5 under OpenCode rising from 64.2 to 75.7 on WorkBuddy-Bench (+11.5 points).

• 8 editable harness components, version 0.1.0, runs on Python 3.12/3.13 with uv
• Best gain: GLM-5 under OpenCode on WorkBuddy-Bench, +11.5 points
• Worst: GLM-5 under Codex on JobBench, only +1.2 points (23.6 to 24.8)
• Works without labels or verifiers, using trajectory contrast instead