Anthropic launched a new modular system called “Skills” for its flagship AI model, Claude
Source: Saiyp | Date: 2025-12-14 20:47:00+08
The Skills system organizes task-specific instructions, scripts, and resources into themed folders that Claude can automatically select and load as needed.
This approach turns complex prompt engineering into reusable, stackable, and portable components.
Claude decides independently which skills to use for each task, enhancing its ability to handle complex workflows efficiently.
The feature is available across Claude’s app, Claude Code, and API, supporting broad cross-platform use.
Availability & User Access
In the Claude app, Skills are accessible to Pro, Max, Team, and Enterprise users.
Anthropic provides built-in skills for generating tables, presentations, PDFs, and more.
Users can create custom skills, and Claude clearly shows which skills are active during editing.
Team and Enterprise admins must enable the feature before their members can use it.
A special “Skill Creator” skill helps users build new skills using a folder structure, a SKILL.md config file, and required assets.
Developer Integration
For developers, Skills integrate with the Messages API and the /v1/skills endpoint when code execution is enabled (currently in beta).
In Claude Code, skills can be installed from Anthropic’s official marketplace at anthropics/skills or manually placed in ~/.claude/skills.
Claude automatically loads relevant skills and includes version control and team collaboration features.
The Claude Agent SDK also supports integration with custom agents.
Prebuilt skills for Excel, Word, and PowerPoint are already available.
Users can build and manage custom skills via the Claude Console.
Security Recommendations
Anthropic strongly advises installing skills only from trusted sources.
Users should carefully review all files—especially code dependencies and embedded resources like images or scripts—before using unfamiliar skills.
They should also watch out for any instructions that direct Claude to connect to untrusted external services.