Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Learning to Control LLM Agent Harnesses with Offline Reinforcement Learning

About

Large language model (LLM) agents are usually improved by changing prompts, models, or hand-written workflows, while the execution harness around the model is treated as fixed infrastructure. We argue that this harness is itself a learnable control layer. We formalize harness operation as a finite-horizon Harness MDP, where a lightweight controller selects structural execution actions while the LLM executor remains frozen. The controller is trained from offline rollouts using advantage-weighted regression with only terminal task-rubric rewards. We also separate final task quality from a post-hoc Harness Maturity Score, which measures whether the harness follows reliable execution patterns rather than only whether the final answer is correct. This separation gives a finite-buffer view of harness learning: final-quality gains require high-return support in the offline buffer, while process behavior can shift whenever it aligns with advantage-weighted actions. Across six controlled domains and two public-benchmark adapters, the learned controller consistently improves verification behavior and selectively improves final task quality, with the largest gains on adapted tau-bench retail, adapted AgentBench DB-Bench, and coding with a calibrated structural verifier. Ablations against behavior cloning and Forced CHECK show that the gains are not explained by imitation or by simply adding checks. These results identify harness control as a learnable layer for frozen LLM agents, while showing that offline support limits when better process control becomes better final answers.

Haiwen Yi, Xinyuan Song• 2026

Related benchmarks

TaskDatasetResultRank
Agent Task Performanceτ-bench retail 20-task (held-out)
G0.519
2
Agent Task PerformanceAgentBench DB-Bench 20-task (held-out)
G Score54.7
2
Agent Task PerformanceCoding 20-task (held-out)
G Score81.2
2
Agentic Reasoning and Task ExecutionControlled-domain Knowledge-work
Mean Task Rubric Score (G)0.464
2
Agentic Reasoning and Task ExecutionControlled-domain Coding
Mean Task Rubric Score (G)0.812
2
Agentic Reasoning and Task ExecutionControlled-domain Planning
Mean Task Rubric Score (G)0.412
2
Agentic Reasoning and Task ExecutionControlled-domain Multi-tool
Mean Task Rubric Score (G)51.5
2
Agentic Reasoning and Task ExecutionControlled-domain Research
Mean Task Rubric Score (G)0.311
2
Agentic Reasoning and Task ExecutionControlled-domain Long-memory
G (Mean Task-Rubric Score)0.451
2
Deliberative-reasoning verificationAgentBench DB-Bench 20 held-out tasks
CheckBeforeSubmit (CBS)16.7
1
Showing 10 of 13 rows

Other info

Follow for update