Anthropic ships Agent Skills
Anthropic released Agent Skills, a package that bundles instructions, metadata and optional resources so Claude can invoke modular capabilities automatically when appropriate. The company also documents that file-creation and editing features run in a sandbox with network access disabled to limit data exfiltration. The documentation positions these capabilities as reusable, contained workflow units for tasks such as templates or scripts rather than a single monolithic chat interface. (platform.claude.com) (support.claude.com)
Anthropic has rolled out Agent Skills, a way to package reusable capabilities so Claude can call them automatically during a task. (platform.claude.com) Anthropic’s documentation says each Skill bundles instructions, metadata, and optional resources such as scripts or templates. Claude is meant to discover and use those packages when a request matches the Skill. (platform.claude.com) The company says Skills can be Anthropic-built or custom-built by customers, and the same system is available in the Claude application and through the Claude application programming interface. Anthropic’s API guide says developers can invoke them with code execution containers and specify a `skill_id` for custom setups. (platform.claude.com 1) (platform.claude.com 2) Anthropic is framing the feature around small, contained workflow units rather than one all-purpose chat prompt. Its best-practices guide tells developers to keep Skills concise, structured, and focused on repeatable tasks Claude can reliably recognize. (platform.claude.com) That packaging model is showing up alongside Anthropic’s broader push into agent products. Last week, the company launched Managed Agents, and its documentation says those agents can automatically invoke both Anthropic pre-built skills and organization-specific custom skills. (platform.claude.com) (wired.com) Anthropic is also pairing the feature with tighter controls around what Claude can do when it writes files. Its support documentation says file creation and editing run inside a sandbox, with network access disabled by default to block data from leaving Anthropic’s infrastructure even if another safeguard fails. (support.claude.com) That safeguard is not absolute. Anthropic says Team and Enterprise customers can turn on network egress for code execution and file creation, and mobile users can enable or disable code execution and file creation in settings. (support.claude.com) The company’s docs list pre-built Skills for document work including PowerPoint, Excel, Word, and Portable Document Format handling. Anthropic also maintains a public GitHub repository for Skills, where it describes them as folders of instructions, scripts, and resources loaded dynamically for specialized tasks. (platform.claude.com) (github.com) Anthropic’s release notes from April 2026 show the same file-creation tools expanding across paid plans, including Pro and mobile access. Agent Skills fit that same pattern: more of Claude’s work is being broken into named tools, gated controls, and reusable building blocks. (support.claude.com)
