Image by: Anthropic
TL;DR: Anthropic has launched Agent Skills, a new system that lets Claude load specialized task modules for more precise, efficient work across apps, code, and API.
What’s new: Claude can now use modular “Skills” to perform specialized tasks with improved accuracy and efficiency.
Technical details:
Agent Skills are structured folders containing instructions, scripts, and resources that Claude selectively loads when a task requires them. Each skill acts as a self-contained unit of expertise, like knowing how to build spreadsheets or apply a company’s brand voice. Claude will now dynamically scan available skills, activate the relevant ones and load only what’s needed to minimize resource overhead.
Skills are composable, meaning multiple can stack to support complex workflows; portable, working seamlessly across Claude apps, Claude Code, and API; and efficient, since only the minimal data is accessed at runtime. They can even include executable code, allowing Claude to handle technical processes traditionally outside natural language models’ comfort zones.
Practical impact:
For users, this means Claude gets smarter without slowing down as it automatically draws out the right expertise for each task. Teams can now package institutional knowledge, standard operating procedures, or domain-specific workflows into reusable skills. Developers can manage custom skill versions via the new /v1/skills API endpoint or the Claude Console. Meanwhile, Enterprise admins can centrally enable and distribute skills across organizations, keeping expertise standardized and secure.
Availability:
Skills are rolling out now to Pro, Max, Team, and Enterprise plans. The feature requires the Code Execution Tool beta. Anthropic provides built-in skills for common use cases, with example repositories and documentation available through the Claude Console and Anthropic Academy.
Agent Skills turn Claude from a general-purpose assistant into a truly modular AI platform to one that can actually learn how you work instead of just pretending to.