Search papers, labs, and topics across Lattice.
SkillGrad optimizes LLM agent skills by treating skill packages as structured parameters and iteratively refining them based on task execution losses. Text-based gradients, derived from automatic diagnoses of failures, guide the skill correction process. A momentum agent stabilizes the optimization by accumulating recurring diagnostic patterns, and an LLM-based patcher applies layer-aware edits to the skill package.
SkillGrad's gradient-descent-inspired approach makes optimizing LLM agent skills as straightforward as training neural networks, achieving a 6.7% improvement over training-based baselines.
Agent skills provide a lightweight way to adapt LLM agents to specialized domains by storing reusable procedural knowledge in structured files. However, whether downloaded from third parties or self-generated, these skills are often unreliable, incomplete, or outdated. Existing skill-evolution methods often address these deficiencies through heuristic reflections without an explicit optimization formulation. In this paper, we propose SkillGrad, a gradient-descent-inspired framework for optimizing agent skills. SkillGrad treats the skill package as a structured parameter to optimize in a gradient descent fashion: task executions provide trajectory-level loss evidence, automatic diagnoses then provide text-based gradients that indicate the correction directions. To stabilize optimization across iterations, a momentum agent accumulates recurring diagnostic patterns into a persistent memory overlay. Finally, an LLM-based patcher executes the parameter update by applying layer-aware edits to the skill package. Evaluated on SpreadsheetBench Verified and WikiTableQuestions, SkillGrad consistently outperforms training-based skill evolution baselines across two backbone LLMs, improving over the strongest training-based baseline by $6.7$ percentage points on average. Ablations further show that momentum and contrastive diagnosis both contribute to the final skill quality.
