The Anatomy of an Agentic Skill: Triple-Interface Design and the Generative Mechanisms of Artifact Adoption
Type: article
Abstract: AI agents can perform new tasks by following small instruction files called skills. A skill should work for three groups at once: people deciding whether to use it, AI agents carrying out its instructions, and contributors who update and share it. We analyze 49,332 skills from the ClawHub registry to learn what helps them attract users and evolve. Skills receive more installs when they combine clear explanations for people with precise instructions for AI agents. Files with technical instructions but little readable context receive fewer installs. Features supporting sharing, versioning, and reuse are linked to more updates, although they are also linked to fewer initial installs. A three-wave analysis shows that the advantage of combining human and machine readability weakens as common technical practices spread. We describe skills as triple-interface artifacts because one file coordinates people, AI agents, and contributor communities.
Institutions: Florida International University, Indiana University
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7222912320