Skills, in artificial intelligence systems, refer to standardized, modular execution units that encapsulate specific task capabilities and can be invoked on demand by intelligent agents. This concept was first proposed by Anthropic in Claude Code in October 2025 and developed into an open standard on December 18, 2025. Related research and engineering practices have been published in the paper "...".Agent Skills for Large Language Models: Architecture, Acquisition, Security, and the Path Forward"and"Agent Skills: A Data-Driven Analysis of Claude Skills for Extending Large Language Model Functionality"

Skills are essentially reusable capability packages composed of instructions, scripts, and resources. They typically encapsulate domain knowledge, task flows, and execution logic, possessing clear functional boundaries, stable input/output interfaces, and can be selectively invoked by the model according to task requirements to improve its performance and stability in specific tasks. This mechanism enables general-purpose large models to be transformed into task-oriented intelligent agents with specialized knowledge, and is an important capability abstraction layer following MCP and Tools.