自主 AI 代理技能

Anthropic introduced a major expansion of its Claude Skills system, positioning it as an open standard for building and sharing AI agent skills. The Agent Skills specification lets developers and enterprises define repeatable tasks Claude can perform autonomously, using structured instructions tailored to real workflows. By opening the standard, Anthropic aimed to make these reusable skills easier to create, distribute, and adopt across different tools and teams.

The update added centralized management for organizations on Team and Enterprise plans, giving admins a single place to govern which skills are available and how they are used. Anthropic also partnered with platforms like Canva, Notion, Figma, and Atlassian to offer prebuilt skills, lowering the barrier to getting started. Skills can be authored from simple files and folders that describe step-by-step processes, such as completing online forms or navigating specific websites, enabling Claude to execute those sequences independently.

For consumers and business users, this shift highlighted how AI is moving from generic chat to practical, task-focused automation. Skills help teams encode institutional knowledge into reusable agents that reduce manual work and human handoffs. The open standard approach supports interoperability and governance, making it easier for enterprises to scale agentic AI while maintaining control, oversight, and security around what these agents are allowed to do.