Skip to content

Source paper

Harness^2: Recursive Agent Harnessing for an Open World

Ruiyao Xu, Yanfei Chen, Zhongying CuiZhu and 6 more

Ruiyao Xu, Yanfei Chen, Zhongying CuiZhu, Bhavana Dalvi Mishra, Yifei Ming, Han Yu, Rujun Han, Chen-Yu Lee, Tomas Pfister

arXiv 2610.14224 · 202670 upvotes on Hugging Face

Research ideas from this paper (5)

Generated by AI. The paper’s authors did not write these.

Utility-Tracked Consolidation and Failure-Aware Retrieval for Harness Improvement Libraries

Consolidate overlapping agent scaffolding components and rank past editing lessons by empirical utility so long-running self-improvement libraries stay compact and relevant.

Harness^2: Recursive Agent Harnessing for an Open World
🟡Fairly easy💡Moderate InnovationSingle GPU or lessRelated work found#Agents#Retrieval & memory

Difficulty- and Saturation-Aware Adaptive Task-Level Recursion Budgeting

Dynamically allocate test-time exploration budgets using early rollout divergence and early-stopping checks so agents avoid wasting inference tokens on easy or saturated tasks.

Harness^2: Recursive Agent Harnessing for an Open World
🟡Fairly easy💡Moderate InnovationSingle GPU or lessRelated work found#Agents#Efficiency

Requirement-Coverage and Execution-Trace-Aware Candidate Deliverable Selection

Augment final output selection with trajectory-derived behavioral invariants and requirement-coverage checklists to prevent selectors from picking incomplete or non-conforming deliverables.

Harness^2: Recursive Agent Harnessing for an Open World
🟢Easy🔧Minor TweakSingle GPU or lessRelated work found#Agents#Data & evaluation

Adversarial Specification Probing and Multi-Seed Contrast Verification for Verifier-Free Harness Edits

Cross-check proposed agent edits with specification-derived adversarial probes and multi-seed reruns to prevent shared model assumptions from passing self-written verification scripts.

Harness^2: Recursive Agent Harnessing for an Open World
🟠Medium🔬Significant ResearchSingle GPU or lessRelated work found#Agents#Safety & alignment

Meta-Recursive Co-Evolution of Improver Stage Instructions and Composition Rules

Co-evolve the meta-prompt rules and composition heuristics using retrospective edit-attribution logs so an agent improves how it edits, not just its stored examples.

Harness^2: Recursive Agent Harnessing for an Open World
🔴Hard🔬Significant ResearchSingle GPU or lessClosely matching work found#Agents#LLMs

AI summary

This paper introduces Harness^2, a plug-and-play recursive agent harnessing framework that enables LLM agents to optimize their execution harnesses at test time in open-world settings without ground-truth verifiers. It couples task-level contrastive refinement (Propose-Probe-Compose) with harnessing-level recursion that distills reusable harness components and lessons into an evolving improvement library across tasks.

How it works (AI summary)

Operating around frozen executor LLMs (Gemini 3.5 Flash and GLM-5) inside an eight-component file-level harness (system prompt, guardrails, memory, skills, tool guidance, scripts, subagents, and middleware plugins), the framework uses an agentic editor (Claude Opus 4.8) to propose invariant and domain-specific edits, probe them on the current task, and compose a successor harness via parallel or sequential recursion. After blind deliverable selection, an Experience Reflector attributes edit effects from execution contrasts and updates a domain-indexed improvement library with adopted harnesses and EXPERIENCE.md lessons for subsequent tasks.