---
type: "post"
id: "google-harness2-for-recursive-self-improvement-github-repo-89c90b"
slug: "google-harness2-for-recursive-self-improvement-github-repo-89c90b"
title: "Google Harness² for Recursive Self Improvement (GitHub Repo)"
description: "Harness² makes the harness, rather than the model weights, the object of recursive improvement."
retrieval_nugget: "Harness² makes the harness, rather than the model weights, the object of recursive improvement."
published_at: "2026-10-15"
updated_at: "2026-10-15"
record_date: "2026-10-15"
date_kind: "published_at"
topics: ["google"]
source_urls: ["https://github.com/google-research/harness2"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:google-harness2-for-recursive-self-improvement-github-repo-89c90b"
status: "published"
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:905a7751bb00a39b1cbfff2353849dbaeb04a02c4198dd7ef254f259776219f3","reviewed_at":"2026-10-15T16:09:22.316Z","source_evidence_count":1,"verified_claim_count":3,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":3,"implication_count":2,"watch_condition_count":1,"related_record_count":0}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"the harness can be refined as an editable, evidence-bearing system around a fixed model.","observed_facts":[{"text":"The repository describes task-level Propose–Probe–Compose refinement without ground-truth feedback, alongside a persistent harness-level loop that retains adopted harnesses and evidence-supported lessons.","source_urls":["https://github.com/google-research/harness2"]},{"text":"It lists eight editable components: system prompt, guardrails, memory, tool guidance, subagents, skills, scripts, and plugins.","source_urls":["https://github.com/google-research/harness2"]},{"text":"Google reports maxima of 11.5 points on WorkBuddy-Bench, 4.5 on JobBench, and 11.8 percentage points in LAB partial pass rate.","source_urls":["https://github.com/google-research/harness2"]}],"mechanism":"An agent proposes harness edits, probes their effects on the same task, and composes a refinement from behavioral contrasts; adopted harnesses and evidence-supported lessons can then guide subsequent task-level refinement.","why_now":"the release turns harness engineering from an abstract practice into a runnable, benchmarked workflow for recursive refinement.","implications":["Teams can treat harness changes and their evidence as a persistent engineering asset, while holding model weights fixed.","The two loops separate local adaptation for a task from retained learning intended to inform future tasks."],"evidence_boundary":"the supplied repository describes the method and reports its own benchmark results; independent replication and broader deployment outcomes are not established here.","watch_conditions":["independent runs should test whether gains persist on unseen tasks and across different model–harness configurations."],"related_records":[],"new_branch_reason":"No prior New Runtime record was supplied; this release's linked task-level and persistent harness-improvement loops warrant a distinct analytical record."}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/google-harness2-for-recursive-self-improvement-github-repo-89c90b/","markdown":"https://newruntime.com/posts/google-harness2-for-recursive-self-improvement-github-repo-89c90b.md","json":"https://newruntime.com/posts/google-harness2-for-recursive-self-improvement-github-repo-89c90b.json"}
related_patterns: ["harness-architecture-outlives-model-choice"]
---

# Google Harness² for Recursive Self Improvement (GitHub Repo)

## Retrieval answer

Harness² makes the harness, rather than the model weights, the object of recursive improvement.

Harness² makes the harness, rather than the model weights, the object of recursive improvement. Its thesis is: the harness can be refined as an editable, evidence-bearing system around a fixed model. The repository describes two distinct loops: task-level changes to the execution harness and harness-level retention of adopted configurations and lessons.

The mechanism is: An agent proposes harness edits, probes their effects on the same task, and composes a refinement from behavioral contrasts; adopted harnesses and evidence-supported lessons can then guide subsequent task-level refinement. The repository lists eight editable components: system prompt, guardrails, memory, tool guidance, subagents, skills, scripts, and plugins. Google reports gains across three professional benchmarks and four model–harness configurations, including maxima of 11.5 points on WorkBuddy-Bench, 4.5 on JobBench, and 11.8 percentage points in LAB partial pass rate. These reported figures make the release concrete, but they do not by themselves show that the method generalizes beyond the tested configurations.

Why now: the release turns harness engineering from an abstract practice into a runnable, benchmarked workflow for recursive refinement. The first implication is: Teams can treat harness changes and their evidence as a persistent engineering asset, while holding model weights fixed. The evidence boundary is: the supplied repository describes the method and reports its own benchmark results; independent replication and broader deployment outcomes are not established here. First watch condition: independent runs should test whether gains persist on unseen tasks and across different model–harness configurations. If they do, evidence-backed harness evolution may become a durable production layer; if not, the reported improvements may remain benchmark- or setup-specific.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Google](https://newruntime.com/topics/google.md): Explore the google topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)