TeleTune: Evolving Agent Skills From Offline Telemetry

2026-10-04

AI Summary

This study addresses the challenges of absent goals, non-replayability, and interleaved trajectories when learning reusable skills from offline telemetry logs. We propose an offline skill optimization framework for computer-use agents that constructs a textual skill library and leverages action prediction errors to guide iterative skill editing. Relevant demonstrations are retrieved based on workflows to enhance agent decision-making, while a skill-guided progress mechanism enables purely offline optimization without real-time interaction. Experimental results demonstrate that our method achieves success rates of 77.1% and 80.6% on the WorkArena and Online-Mind2Web benchmarks, respectively, significantly outperforming existing baselines. Furthermore, it substantially reduces token consumption and improves system robustness.

Abstract

Computer-use agents need to capture procedural knowledge of how people use software. User telemetry offers a scalable source of this knowledge. However, learning reusable skills from these logs requires addressing three challenges: (1) Goal Underspecification, since logs do not record the goal behind each action; (2) Non-Replayability, since past activity cannot be replayed to evaluate skill updates; and (3) Interleaved Trajectories, since logs may mix several tasks without marking their boundaries. To address these, we introduce TeleTune, a framework for learning a textual skill library from offline logs without recorded goals, cannot be replayed during optimization, and may interleave tasks. TeleTune uses action-prediction errors on logged trajectories to propose library edits and keep only those that improve held-out action-prediction accuracy, which we call skill-guided progress. The learned workflows also enable retrieval of demonstrations that cover the subgoals of a new task. At test time, the agent is provided with the learned library and the workflow-based retrieved demonstrations. Experiments on WorkArena and Online-Mind2Web show that TeleTune outperforms random retrieval, Agent Workflow Memory (AWM), and their combination. We find that the best baseline varies by setting, whereas TeleTune achieves average success rates of 77.1% and 80.6%, respectively, improving over the strongest baseline on each benchmark by 6.7% and 7.7%. Under the heaviest perturbation of the WorkArena training data,TeleTune keeps the highest average success rate at 68.5%, 6.3% above the strongest baseline. Our analyses show (1) skill optimization and workflow-based retrieval are complementary, (2) optimizing on fixed logs costs 5 to 75 times fewer tokens than validating the same edits with live episodes, (3) skill-guided progress tracks the live success rate.

Problem

• Computer-use agents
• Offline telemetry
• Skill learning
• Goal underspecification
• Interleaved trajectories

Innovation

• Offline Telemetry
• Skill Library
• Action Prediction
• Workflow-based Retrieval
• Computer-use Agents