AI company Anthropic has launched a new feature called “Skills” for its Claude series of models. According to the company, Skills are designed to equip its AI with specialized expertise for specific tasks, making it more useful for business and professional workflows. The feature is available for users of Claude’s Pro, Max, Team, and Enterprise plans.
In essence, a Skill is an organized folder containing instructions, documents, and even executable code scripts. The AI can access this folder when it recognizes that a user’s request pertains to that skill’s domain. Anthropic compares the process to giving an AI an onboarding guide for a new job. This allows users to package their procedural knowledge into reusable resources, moving beyond single-use prompts.
How ‘progressive disclosure’ works
The system is built on a principle Anthropic calls “progressive disclosure,” which manages how much information the AI processes at any given time. This method is designed to be efficient and scalable:
- Initially, Claude is only aware of the name and a brief description of each available Skill.
- When a user’s task matches a Skill’s description, Claude reads the main instruction file within that Skill’s folder to get detailed guidance.
- If needed for more complex operations, Claude can then access additional documents or run code scripts included in the Skill’s folder.
This layered approach prevents the AI’s context window (its short-term memory) from being overloaded with irrelevant information. Skills can also include code, which allows Claude to perform tasks that require precision and consistency, such as data extraction from a document, by running a script instead of trying to generate the outcome itself.
According to statements from Anthropic, this combination of structured instructions, tiered information access, and the ability to execute code distinguishes Skills from other AI customization tools like OpenAI’s Custom GPTs.
Given that Skills can execute code, Anthropic advises users to install them only from trusted sources to mitigate security risks. The company provides administrative controls for enterprise customers to manage access to the feature within their organizations.
Sources: Anthropic, VentureBeat, The Verge
