Harness^2: Recursive Agent Harnessing for an Open World
Quick Answer
This paper shows that Harness^2 introduces a recursive agent harnessing framework that enhances LLM performance by up to 11.5 points on benchmarks like WorkBuddy-Bench.
Quick Take
By utilizing task-level and harnessing-level recursion, it effectively adapts to unseen tasks and improves agent capabilities without ground-truth verifiers. This approach consistently boosts GLM-5 and Gemini 3.5 Flash performance across multiple configurations.
Key Points
- Harness^2 improves GLM-5 and Gemini 3.5 Flash performance across three benchmarks.
- Achieves performance gains of up to 11.5 points on WorkBuddy-Bench.
- Utilizes task-level recursion for reliable edits without ground-truth verifiers.
- Harnessing-level recursion updates continual improvement harness for future tasks.
- Scalable through parallel or sequential recursion for enhanced adaptability.
DeepSignal Analysis
What happened
Harness^2 presents a recursive agent harnessing framework aimed at improving the performance of large language models (LLMs) in open-world settings. It addresses challenges related to unseen tasks and the lack of ground-truth verifiers, achieving performance gains across multiple benchmarks.
Key evidence
- Harness^2 improves GLM-5 and Gemini 3.5 Flash performance by up to 11.5 points on WorkBuddy-Bench and 4.5 points on JobBench.
- The framework employs task-level recursion through a Propose-Probe-Compose step, allowing for reliable edits without ground-truth verifiers.
- Harnessing-level recursion enables continual improvement by extracting experience from previous edits, enhancing adaptability to future tasks.
Why it matters
The development of Harness^2 is significant as it provides a method for LLMs to adapt to new tasks without relying on predefined benchmarks or verifiers. This could lead to more robust AI systems capable of functioning effectively in dynamic environments, which is crucial for real-world applications.
What to watch
Paper Resources
📖 Reader Mode
~2 min readAbstract:LLM agents' capabilities depend not only on their underlying models but also on their execution harnesses. Recursive self-improvement (RSI) offers a promising approach to automatically evolve these harnesses. However, deploying RSI in open-world settings still faces two critical challenges. First, open-world tasks are typically unseen during the harness design phase and novel at test time, leaving no ground-truth verifiers to reliably guide online harness edits. Second, the RSI process remains highly task-specific and fails to generalize to novel scenarios. We introduce Harness^2, a plug-and-play, recursive agent harnessing framework that addresses these challenges through two connected levels of recursion. Task-level recursion refines the current task's harness using a contrastive Propose-Probe-Compose step that reliably guides edits without ground-truth verifiers, and can be scaled further via parallel or sequential recursion. Harnessing-level recursion then extracts experience from these edits to update a continual improvement harness, enabling the agent to better adapt to future tasks. Across three professional benchmarks and four model-harness configurations, a single step consistently improves GLM-5 and Gemini 3.5 Flash over their base harnesses, yielding gains of up to 11.5 points on WorkBuddy-Bench, 4.5 points on JobBench, and 11.8 percentage points on LAB, with further gains as the task-level recursion budget increases.
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.14224 [cs.AI] |
| (or arXiv:2610.14224v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.14224 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yanfei Chen [view email]
[v1]
Fri, 9 Oct 2026 18:34:25 UTC (6,866 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →VEPA: Verified Evolving Formal Policy for Lifelong Agent Guardrails
VEPA introduces a framework for verified evolving formal policies for LLM agents, enhancing safety by adapting guardrails based on deployment feedback. It outperforms existing methods across benchmarks like ASSEBench-Safety and ASSEBench-Security, ensuring deterministic policy enforcement even under challenging conditions.