SkillAdam: Stable and Efficient Skill Evolution for Agents
Gaoyuan Li, Meihao Fan, Shaolei Zhang and Ju Fan at Renmin University present SkillAdam, which ports Adam's two moment estimates to the optimization of discrete, non-differentiable skill documents so that skill self-evolution stops oscillating.
Ask this paper
Two named failure modes: Direction Stability requires that effective corrections accumulate rather than be overwritten by iteration-local feedback; Update Adaptivity requires each revision's scope to reflect how consistent recent case-level improvements have been.
First moment as optimization memory: An optimization memory records identified problems and the outcomes of prior solution attempts, which stabilizes the update direction the way Adam's first moment smooths a noisy gradient.
Second moment as an edit budget: A volatility-driven edit budget tracks history-weighted variation in recent case-level improvements and adaptively controls how large each revision is allowed to be.
State of the art on seven benchmarks: SkillAdam leads across seven benchmarks spanning short- and long-horizon tasks, with more stable optimization dynamics than heuristic skill-revision loops.
Cheaper as well as better: It reaches stronger skills in substantially fewer optimization iterations and at lower cost than prior methods. Code is at github.com/ruc-datalab/SkillAdam.
Abstract
Agent skills provide a lightweight way to equip frozen language-model agents with domain knowledge and procedural guidance, yet obtaining high-quality skills remains costly and difficult to scale. Expert-written skills require substantial human effort. Recent skill self-evolution methods automate an iterative loop that uses execution feedback to revise skills, but their heuristic update strategies often yield unstable optimization and low iteration efficiency. We identify two challenges in realizing stable and efficient skill self-evolution. Direction Stability requires effective corrections to accumulate rather than be overwritten by iteration-local feedback. Update Adaptivity requires the scope of each revision to reflect the consistency of recent case-level improvements. We introduce SkillAdam, an Adam-inspired framework for optimizing discrete and non-differentiable skill documents. As a functional analogue of Adam's first moment, an optimization memory records identified problems and the outcomes of prior solution attempts to stabilize the update direction. As a functional analogue of Adam's second moment, a volatility-driven edit budget tracks the history-weighted variation of recent case-level improvements and adaptively controls the update magnitude. Across seven benchmarks that span short- and long-horizon tasks, SkillAdam achieves state-of-the-art performance with more stable optimization dynamics. It also obtains stronger skills with substantially fewer optimization iterations and lower cost than prior methods. Code repository: https://github.com/ruc-datalab/SkillAdam