SkillOpt-Lite: Better and Faster Agent Self-evolution via One Line of Vibe
2026-07-03 • Software Engineering
Software EngineeringArtificial IntelligenceMachine Learning
AI summaryⓘ
The authors study how to improve skill learning in autonomous agents using a simple and well-justified approach. They connect skill optimization to Zeroth-Order optimization methods and introduce three key principles that help agents learn better and more reliably. Their streamlined method, SkillOpt-Lite, speeds up learning and outperforms previous techniques on several AI models. They also show it can be used in real coding tools and scaled up to optimize entire agent workflows, beating larger models on benchmarks. The authors provide their code for others to use and build on.
skill optimizationautonomous agentszeroth-order optimizationtrajectory explorationconsensus attribute miningvalidation gatingLiveMath benchmarkagent skill learninggeneralizationHarnessOpt
Authors
Yifei Shen, Bo Li, Xinjie Zhang
Abstract
While skill optimization for autonomous agents has gained traction, existing methods rely on complex pipelines. This leaves a fundamental question unaddressed: What constitutes a minimal viable pipeline for skill optimization, where every component is justified by theory or empirical necessity? We formalize skill optimization via Zeroth-Order (ZO) optimization, mapping classical counterparts (central difference, trust regions) to recent literature. Noting that unlike blind numerical perturbations in classical ZO, skill trajectories serve as interpretable debugging feedback. Grounded in Claude Code philosophy and PAC learning, we establish three principles for convergence and generalization: file-system-based trajectory exploration, consensus attribute mining, and independent validation gating. Eliminating redundancies, we propose SkillOpt-Lite. It accelerates convergence and outperforms full SkillOpt: improving LiveMath by +8.8 points on GPT-5.5 and +25.4 points on GPT-5.4-nano, allowing the nano model to surpass standard GPT-5.4 optimized by SkillOpt. Finally, we integrate our framework into production coding agents like VSCode Copilot, enabling developers to evolve agent skills via one line of vibe. Because our framework treats all agent components simply as standard editable code, this minimal pipeline naturally generalizes to full harness optimization (HarnessOpt). On SpreadsheetBench, HarnessOpt enables GPT-5.4-nano to achieve 0.7758 accuracy, outperforming the larger GPT-5.5 running standard pipelines (0.7620). Code is available at https://github.com/EvolvingLMMs-Lab/SkillOpt-Lite.