Anthropic makes agent Skills an open standard – News

Anthropic PBC said today it’s updating its Skills feature, which enables companies to teach its powerful large language model how to perform very specific, work-related tasks.

The update comes as the artificial intelligence industry doubles down on so-called AI agents, which are AI systems designed to work autonomously with minimal human supervision. As part of today’s update, Anthropic said, it’s making the Agent Skills specification an open standard, making it easier for anyone to share and deploy new skills.

The company launched Skills for Claude in October. They’re a new method for developing specialized AI agents using files and folders. Those folders include instructions, resources and scripts that Claude and other LLMs can leverage to perform specific tasks. For instance, if workers want Claude to create an Excel spreadsheet or PowerPoint presentation, they can direct it to open a folder with special skills for those tools, so it will know exactly how to use them to complete the objective.

Users can also make their own changes to these skills folders, adding files that describe processes such as browsing a website or filling out a specific kind of form, in order to teach agents new things they haven’t come across before. Anthropic’s Skills are there to make AI agents more capable, and can even teach them how to run new software without having to talk to another tool or agent.

With today’s update, Anthropic is adding a new organization-wide management capability, which makes it possible for enterprises subscribed to Claude’s Team and Enterprise plans to manage Skills from a central location.

The decision to make Skills an open standard is perhaps the most significant change, though. This move follows the success of Anthropic’s Model Context Protocol, which is an open standard for AI agents to communicate with each other and utilize third-party tools. MCP has become the de facto communication standard for the agentic AI industry, and earlier this month, control of the project was officially handed over to the Linux Foundation.

Anthropic certainly made a huge contribution to the emergence of AI agents when it released MCP, but that doesn’t mean it will be able to do so again, said Holger Mueller of Constellation Research. The analyst said he’s skeptical that the company can be the flag-bearer of new standards in AI, because agentic skills can be taught and accessed in various ways. “Single vendors are rarely able to set standards by themselves, and while MCP was endorsed by practically all AI vendors, it doesn’t mean history will repeat itself,” Mueller said. “The merits of Agent Skills doesn’t guarantee adoption, especially considering it’s very similar to APIs, which have already standardized AI’s communications and signature protocols, but time will tell.”

A simpler update makes Skills more accessible to Claude subscribers within the Tools sidebar. Anthropic described this as “quick-create flow – just describe what you want, and Claude builds it.”

Finally, the company has created a directory of “partner-built skills” that includes contributions from the likes of Atlassian Corp., Canva Pty. Ltd., Notion Labs Inc., Figma Inc., Cloudflare Inc., Stripe Inc. and Zapier Inc. Anthropic said the new partner skills can be used in various ways. For instance, they can help AI agents to apply brand style guidelines to marketing materials created in Figma, create tasks in Atlassian’s Jira and Trello platforms, and so on. This library of third-party skills is expected to grow significantly over time as more partners experiment with the new, open standard, Anthropic said.

Image: Anthropic