SkillOpt-Lite: Better and Faster Agent Self-evolution via One Line of Vibe

2026-07-03

AI Summary

Existing approaches to agent skill optimization suffer from procedural complexity and a lack of theoretical grounding. This work proposes a minimalist self-evolution framework grounded in zeroth-order optimization, which requires only a single line of “vibe” feedback to drive the optimization process. The framework integrates PAC learning theory with the Claude Code philosophy and naturally extends to test harness optimization (HarnessOpt). It is guided by three core principles: filesystem trajectory exploration, consensus property mining, and independent verification gating. On LiveMath, the method boosts GPT-5.4-nano’s performance by 25.4 points, surpassing GPT-5.4, and improves GPT-5.5 by 8.8 points. On SpreadsheetBench, HarnessOpt elevates GPT-5.4-nano’s accuracy to 0.7758, outperforming GPT-5.5’s 0.7620.

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.

Problem

• skill optimization
• autonomous agents
• minimal pipeline
• Zeroth-Order optimization
• generalization

Innovation

• Zeroth-Order Optimization
• Skill Self-evolution
• Minimal Pipeline
• Trajectory-based Feedback
• Harness Optimization