SkillAlign: Aligning Skill Interfaces for LLM-based Agents
Shuo Ren, Xiaomian Kang and Jiajun Zhang at the Institute of Automation, Chinese Academy of Sciences argue that how a skill is exposed to an agent changes its value as much as which skill is chosen, and build SkillAlign to measure that by holding everything else fixed.
Ask this paper
The assumption being challenged: Work on skill acquisition, retrieval, compression and composition generally treats the interface as fixed once a skill is selected. The same skill can help, distract or mislead depending on how it is shown.
Multi-view procedural cards: Candidate skills are represented provider-agnostically as cards and rendered through alternative exposure interfaces: full instructions, hints, compressed summaries, workflows, or no exposure at all.
A clean counterfactual design: Task, agent and candidate skills stay fixed while only the exposure interface varies, which is what isolates the effect of presentation from the effect of selection.
Compact beats complete: Across ALFWorld and SkillsBench, exposure form substantially changes task success and rendered context cost, and compact top-k exposure can outperform injecting the full library.
Adaptive exposure is learnable but unsolved: A replay-based policy-learning analysis on ALFWorld finds learnable signal in choosing the exposure, while remaining far from oracle selection.
Abstract
Language-model agents increasingly rely on skills: reusable procedural knowledge for reasoning, tool use, and interaction. Existing work studies how skills are acquired, retrieved, compressed, or composed, but often assumes that once a skill is selected, its interface to the agent is fixed. We argue that this overlooks a key source of skill utility: the same skill can help, distract, or mislead depending on how it is exposed. We propose SkillAlign, a provider-agnostic framework that represents candidate skills as multi-view procedural cards and renders them through alternative exposure interfaces, including full instructions, hints, compressed summaries, workflows, or no exposure. This enables counterfactual evaluation where the task, agent, and candidate skills are fixed while only the exposure interface varies. Across ALFWorld and SkillsBench, we show that exposure form substantially affects task success and rendered context cost, and that compact top-k exposure can outperform full-library injection. We further conduct a replay-based policy-learning analysis on ALFWorld, showing that adaptive exposure contains learnable signal but remains far from oracle selection. Our results suggest that skill-augmented agents should optimize not only which skills to use, but also how those skills are presented.