SkillAlchemy: Open-World Agent Skill Creation
- ID: 342fc898
- 原文链接: https://arxiv.org/abs/2608.23417
- PDF: https://arxiv.org/pdf/2608.23417v1
- 作者: Hengjun Wang, Shuyue Wei, Boyi Liu, Jun Yang, Yongxin Tong
- 日期: 2026-08-24
- 更新: 2026-08-24
- 分类: agents
- 来源类型: paper
- 标签: agent-skills, skill-creation, skillsbench, evidence-grounded, admission-control
- 质量评分: 4/5
- 抓取时间: 2026-08-26T15:43:01Z
中文导读
agent 技能的可靠来源长期依赖人工撰写、模型先验或执行轨迹，陌生任务这三样都缺。SkillAlchemy 研究从开放世界材料造技能：给定欠规范的技能简述与源访问规格，创建者要用对比证据发现简述漏掉的行为相关要求，按证据支持的范围决定每条源派生流程能推广多宽，再编译成语法引导的技能包。87 个 SkillsBench v1.1 任务上，比无技能执行高 19.9 个百分点、比最强自动化基线高 8.6 个百分点，达到与人工策展技能相当的水平。以准入为中心的设计是关键：证据决定范围，范围决定能不能进技能包。
为什么值得关注
「证据决定范围」的准入设计对任何技能型 agent 系统都可直接借鉴：技能包的每条流程都应能回答「凭什么推广到这个宽度」。
关键信息
- 论文标题: SkillAlchemy: Open-World Agent Skill Creation
- 作者: Hengjun Wang, Shuyue Wei, Boyi Liu, Jun Yang, Yongxin Tong
- arXiv: https://arxiv.org/abs/2608.23417
- 发布时间: 2026-08-24
- arXiv 分类: cs.AI
- 备注: 无
- 关联标签: agent-skills, skill-creation, skillsbench, evidence-grounded, admission-control
English Abstract
Agent skills are reusable procedural artifacts that extend language agents with specialized workflows, tool conventions, and domain behaviors at inference time. However, creating reliable skills still depends largely on human authorship, model priors, or execution traces. These sources are often unavailable for unfamiliar tasks, suggesting the need to create skills from open-world materials. In this paper, we study open-world skill creation: given an underspecified skill brief and a source-access specification, a creator must discover behavior-relevant requirements omitted by the brief and determine how broadly each source-derived procedure is justified. We propose SkillAlchemy, an admission-centered framework for source-grounded skill creation. SkillAlchemy identifies implicit requirements through contrastive evidence, admits candidate procedures based on evidence-supported scope, and compiles the admitted content into a grammar-guided skill package. Extensive experiments across 87 SkillsBench v1.1 tasks demonstrate that SkillAlchemy improves pass rate over no-skill execution by 19.9 percentage points and the strongest automated baseline by 8.6 percentage points, while achieving performance comparable to human-curated skills.
Obsidian Notes
- 元数据与摘要由
opencli arxiv paper 2608.23417 -f json拉取。 - 中文导读锚定在论文摘要声明的 claim 与数字上；未补充摘要之外的实验细节。
- 选题来源: OpenClaw定时任务/论文流水线/2026-08-26-论文流水线.md（评分 4/5 对应流水线 8.x/10 档）。
- 本页由 AAIF daily-intake-evening 任务生成（2026-08-26）。