# google-research/harness2

> A framework for recursive optimization of AI agent harnesses (prompts, tools, memory, skills) without fine-tuning model weights, with JobBench, LAB and WorkBuddy-Bench benchmarks.

- Magnitude: 8.9 out of 10 — Breakout
- Stars: 2,911 total · +1,349 stars measured 2026-10-18, 02:46–11:45 UTC, ≈ 2,364 by evening
- Star trust: star growth looks organic
- Category: AI agents · Language: Python · License: Apache-2.0 · Created: 2026-10-06 · Last push: 2026-10-14
- GitHub: https://github.com/google-research/harness2 · Page: https://gitnova.dev/en/r/google-research/harness2

## Useful for

- Run the local harness2-demo without model credentials
- Run the LAB benchmark with parallel recursion and evaluate results
- Compare base and refined harnesses on JobBench or WorkBuddy-Bench

## Why it’s here

- Star-counter measurements on 2026-10-18 (UTC), 02:46–11:45: 1562 → 2911 stars (+1349). This is the change over that interval.
- Estimated end-of-day forecast: about +2,363 stars, using observed gains and the previous day.
- Over the last two days the pace is 154× that of the previous week and a half.
- The repository is 12 days old and already has 2,911 stars. With less than two weeks of history, there's no usual pace to compare the spike against yet.
- Top new repositories this month: #53.
- About 280 forks a day — people are taking the code.

## Star trust

Star growth looks organic.

Star-trust labels are heuristics based on the repository’s behavior, not a check of every stargazer.

## Numbers

- Forks: 249
- Issues and pull requests: 43
- Watchers: 11
- Average over the last week: 629 per day
- Usual pace: too little history (under two weeks)
- Stars in the last hour (measured): 278

## Stars per day, last 13 days (oldest → newest, today is partial)

2026-10-06 … 2026-10-18: 0, 0, 1, 1, 0, 0, 1, 4, 5, 108, 589, 1330, 1349

## Spotted in now

- Top new repositories this month: #53

## Similar by description

1. **Danielxu0208/rsi_eval** — 2.4 · Early signal · Language models · Python · +7 stars measured 2026-10-18, 01:14–11:48 UTC, ≈ 17 by evening
   A Python SDK and CLI that evaluates recursive self-improvement methods by running agent benchmark tasks and returning scores, status, and token usage. It lets you plug in your own improvement loop and reuse pinned benchmark tasks for…
   Full card: https://gitnova.dev/en/r/Danielxu0208/rsi_eval.md
2. **zzzz7788990213-ops/EvoVLM** — 0.1 · Cooling · Language models · Python · +0 stars measured 2026-10-18, 00:24–11:54 UTC, ≈ 0 by evening
   Research project on evolutionary search over vision-language inference programs: without fine-tuning weights, it tunes prompts, image preprocessing, and generation budgets to improve the accuracy–latency trade-off on scientific chart…
   Full card: https://gitnova.dev/en/r/zzzz7788990213-ops/EvoVLM.md

---

Magnitude (0–10) measures how fast and how unusually interest in a repository is growing right now. It is not a quality score.
Days are UTC. “So far today” is a fact; “expected by the end of the day” is a forecast.
Summaries and use cases are written by an LLM (DeepSeek V4.1 Flash) from the README and may be inaccurate: verify specific claims (benchmarks, speed, hardware) in the repository itself.
Data as of 2026-10-18 11:55 UTC, updated every 30 minutes.