CHILL-Harness: Counterfactual Harness Learning for Efficient Reasoning in Long-Horizon Agents

2026-07-28

AI Summary

Existing agent harness systems predominantly rely on manual or static strategies, which struggle to adapt to dynamic task requirements and execution states, often resulting in high computational overhead and low success rates. This work addresses this limitation by formulating adaptive harness orchestration as a causal learning problem and proposes a causal intervention–driven orchestration mechanism. Leveraging counterfactual reasoning, confidence-weighted evidence analysis, and advantage-guided policy selection, the approach achieves efficient inference and execution while maintaining high task success rates. The method incorporates a success-preserving objective and an advantage-marginal empowerment constraint to ensure reliable workflow adaptation. Experimental results demonstrate that, across long-horizon tasks in information retrieval, software engineering, and terminal interaction, the proposed framework significantly reduces token consumption and execution time while sustaining or even improving overall task performance.

Abstract

Agent harnesses have become the operational infrastructure of modern large language model agents, coordinating context, tools, verification, and execution control to translate latent model capability into reliable long-horizon behavior. However, reliable long-horizon behavior requires harness control to adapt to task demands, execution environments, and evolving execution states, whereas current harnesses predominantly rely on hand-crafted or globally fixed policies; this mismatch manifests as unnecessary computational overhead and, in adverse cases, reduced task success. To address this limitation, we formulate the task of enabling adaptive orchestration in harness systems as a causal learning problem and propose Counterfactual Harness Intervention Learning for Long-Horizon Agents (CHILL-Harness). CHILL-Harness intervenes at the orchestration layer to enable advantage-guided workflow adaptation, thereby improving reasoning and execution efficiency while preserving task performance. Specifically, we develop causal intervention effect learning as the effect-estimation component of CHILL-Harness to estimate intervention-relative workflow advantage from confidence-weighted execution evidence and identify advantageous workflow adaptations. We further introduce advantage-realizing causal orchestration as its realization component to adaptively allocate counterfactual reasoning and realize only workflow adjustments supported by sufficient expected advantage. Finally, we incorporate a success-preserving objective and advantage-margin authorization constraints into CHILL-Harness to promote reliable adaptation. Extensive experiments on heterogeneous long-horizon tasks spanning information seeking, software engineering, and terminal interaction show that CHILL-Harness consistently preserves or improves task success while substantially reducing token consumption and execution time.

Problem

• long-horizon agents
• agent harnesses
• adaptive orchestration
• computational overhead
• task success

Innovation

• counterfactual intervention
• causal learning
• adaptive orchestration
• workflow adaptation
• long-horizon reasoning