# Harness-1: Reinforcement Learning for Search Agents with State-Externalizing Harnesses

arXiv: https://arxiv.org/abs/2606.02373 · HTML: https://mlsanity.com/paper/2606.02373/

## Key facts

- Cited by 10 papers in the map (data through 14 Oct 2026).
- Placed in Knowledge- and tool-augmented reasoning, one of 17 directions in LLM reasoning.
- Builds on 13 of its 56 references that are in the map.
- First posted on arXiv 1 Jun 2026.

## Placed in

- LLM reasoning / [Knowledge- and tool-augmented reasoning](https://mlsanity.com/lr/knowledge-and-tool-augmented-reasoning/)

## Builds on (most cited first, 13 of 13)

- [ReAct: Synergizing Reasoning and Acting in Language Models](https://mlsanity.com/paper/2210.03629/) (2022)
- [Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning](https://mlsanity.com/paper/2503.09516/) (2025)
- [Agentless: Demystifying LLM-based Software Engineering Agents](https://mlsanity.com/paper/2407.01489/) (2024)
- [R1-Searcher: Incentivizing the Search Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2503.05592/) (2025)
- [Meta-Harness: End-to-End Optimization of Model Harnesses](https://mlsanity.com/paper/2603.28052/) (2026)
- [MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention](https://mlsanity.com/paper/2506.13585/) (2025)
- [StepSearch: Igniting LLMs Search Ability via Step-Wise Proximal Policy Optimization](https://mlsanity.com/paper/2505.15107/) (2025)
- [SealQA: Raising the Bar for Reasoning in Search-Augmented Language Models](https://mlsanity.com/paper/2506.01062/) (2025)
- [SPA-RL: Reinforcing LLM Agents via Stepwise Progress Attribution](https://mlsanity.com/paper/2505.20732/) (2025)
- [DeepRetrieval: Hacking Real Search Engines and Retrievers with Large Language Models via Reinforcement Learning](https://mlsanity.com/paper/2503.00223/) (2025)
- [Tongyi DeepResearch Technical Report](https://mlsanity.com/paper/2510.24701/) (2025)
- [General Modular Harness for LLM Agents in Multi-Turn Gaming Environments](https://mlsanity.com/paper/2507.11633/) (2025)
- [HARBOR: Automated Harness Optimization](https://mlsanity.com/paper/2604.20938/) (2026)

## Built on by (most cited first, 10 of 10)

- [Dr-DCI: Scaling Direct Corpus Interaction via Dynamic Workspace Expansion](https://mlsanity.com/paper/2606.14885/) (2026)
- [Fetch-then-Explore: Decoupling Selection from Extraction over a Persistent Workspace for Search Agents](https://mlsanity.com/paper/2608.02097/) (2026)
- [EvoHarness-RL: Learning Self-Evolving Runtime Harness for Long-Horizon LLM Agents](https://mlsanity.com/paper/2608.05446/) (2026)
- [Self-Correcting Long-Horizon Search Agents via Tree-Structured Memory](https://mlsanity.com/paper/2608.10676/) (2026)
- [Harness-G: A Graph-Structured Harness for Search Agents](https://mlsanity.com/paper/2607.27652/) (2026)
- [Harness Annealing: Learning to Act with Less External Control](https://mlsanity.com/paper/2610.01235/) (2026)
- [Harness-Search: Guiding Long-Horizon Search through Multi-Agent Coordination](https://mlsanity.com/paper/2610.05382/) (2026)
- [T-Search: An Open Agentic Retriever and Playground for Hard Multi-Step Search](https://mlsanity.com/paper/2610.06782/) (2026)
- [Bounded Autonomy and Verifiable Safety for Agentic AI Enabled Automation](https://mlsanity.com/paper/2610.08815/) (2026)
- [Webhell: A Unified Interface for Agentic Web Access](https://mlsanity.com/paper/2610.13416/) (2026)

## Abstract (arXiv)

Search agents are often trained as policies over growing transcripts: the model must decide how to search while also remembering what it has seen, which evidence is useful, which constraints remain open, and which claims have actually been checked. We argue that this formulation puts too much routine state management inside the policy: reinforcement learning is forced to optimize both semantic search decisions and recoverable bookkeeping that the environment can maintain more reliably. We introduce Harness-1, a 20B search agent (retrieval subagent) trained with reinforcement learning inside a stateful search harness. The harness maintains environment-side working memory, including a candidate pool, an importance-tagged curated set, compact evidence links, verification records, compressed and deduplicated observations, and budget-aware context rendering. The policy retains the semantic decisions: what to search, which documents to keep or discard, what to verify, and when to stop. Across eight retrieval benchmarks spanning web, finance, patents, and multi-hop QA, Harness-1 achieves 0.730 average curated recall, outperforming the next strongest open search subagent by +11.4 points and remaining competitive with much larger frontier-model searchers. Its gains are especially strong on held-out transfer benchmarks, suggesting that reinforcement learning over explicit search state can produce retrieval behaviors that generalize beyond the training domains. Our code is available at https://github.com/pat-jj/harness-1.