{"name":"harness2","repo":"google-research/harness2","url":"https://midorreal.com/project/google-research-harness2","repository":"https://github.com/google-research/harness2","description":"","facts":{"license":{"value":"Apache-2.0","source":"https://github.com/google-research/harness2","checked":"22 Nov 2026"},"language":{"value":"Python","source":"https://github.com/google-research/harness2","checked":"22 Nov 2026"},"releases":{"value":"0","source":"https://github.com/google-research/harness2/releases","checked":"19 Nov 2026"},"contributors":{"value":"11","source":"https://github.com/google-research/harness2/graphs/contributors","checked":"19 Nov 2026"},"archived":{"value":"no","source":"https://github.com/google-research/harness2","checked":"22 Nov 2026"}},"stars":24716.0,"writeup":{"what_it_is":"Harness² is a plug-and-play framework for recursive agent harnessing, which optimizes the system around a model (prompts, tools, memory, skills, orchestration) while keeping model weights fixed. It supports task-level refinement through a Propose–Probe–Compose step and harnessing-level updates to a persistent improvement harness, and it runs on JobBench, LAB, and WorkBuddy-Bench.","audience":"Researchers and developers evaluating agent harness optimization on professional benchmarks","claims":[{"kind":"specific","claim":"A single Harness² step improves GLM-5 and Gemini 3.5 Flash over their base harnesses","excerpt":"a single Harness² step improves GLM-5 and Gemini 3.5 Flash over their base harnesses","status":"not_checked"},{"kind":"specific","claim":"Gains of up to 11.5 points on WorkBuddy-Bench","excerpt":"with gains of up to 11.5 points on WorkBuddy-Bench","status":"not_checked"},{"kind":"specific","claim":"Gains of up to 4.5 points on JobBench","excerpt":"4.5 points on JobBench","status":"not_checked"},{"kind":"specific","claim":"Gains of up to 11.8 percentage points on LAB's partial pass rate","excerpt":"11.8 percentage points on LAB's partial pass rate","status":"not_checked"},{"kind":"broad","claim":"Increasing the task-level recursion budget brings further gains","excerpt":"Increasing the task-level recursion budget brings further gains.","status":"too_broad"}],"alternatives":[],"written_by":"AI, from the project's README","date":"2026-10-20"},"why_now":null,"owner_supplied":null,"score":{"verdict":"too_early","hype":87,"reality":39,"gap":49,"momentum":null,"confidence":60,"version":"score-2.0","rule":"verdict-2.0","calculated_at":"2026-11-22T08:00:00+00:00","verdict_source":"rule"},"early_read":{"estimate":true,"lean":"too_close","gap_low":17,"gap_high":60,"rule":"early-2.0"},"changes":[{"date":"2026-10-06T21:48:11+00:00","text":"Repository created","source":"https://github.com/google-research/harness2"}],"cite":"harness2 is a GitHub project at https://github.com/google-research/harness2 written mainly in Python.","attribution":"Data from Mid or Real, https://midorreal.com/project/google-research-harness2, checked 22 November 2026. Attribution with a link is required.","still_real":null}