Agent Skills documented — limits noted
Anthropic’s Agent Skills docs say skills are bundled with Claude Code, can be installed from a repository, and keep definitions plus execution data under the company’s standard retention policy — importantly, Agent Skills are explicitly not covered by ZDR arrangements (platform.claude.com). The page frames Agent Skills as reusable building blocks for automation while clarifying how data retention and contractual protections apply (platform.claude.com).
Anthropic’s new Agent Skills documentation says the feature stores both skill definitions and run data under standard retention rules, not Zero Data Retention arrangements. (platform.claude.com) Agent Skills are packaged instructions, metadata, and optional files such as scripts or templates that Claude can load when a task matches. Anthropic’s docs describe them as modular capabilities that extend Claude automatically when relevant. (platform.claude.com) Anthropic says pre-built Agent Skills are available on claude.ai and through the Claude Application Programming Interface, and custom skills can be uploaded and managed through a Skills Application Programming Interface. The API guide says both paths use the same code-execution container structure. (platform.claude.com, platform.claude.com) The current docs also tie Skills closely to Claude Code, Anthropic’s coding agent. Claude Code’s docs say skills can be created, managed, and shared there, and its command reference says bundled skills appear as slash commands inside sessions. (code.claude.com, code.claude.com) Anthropic’s public GitHub repository shows how the company is distributing examples and built-ins. The repository says skills are folders with a `SKILL.md` file plus resources, and includes document-related skills for Word, Excel, PowerPoint, and Portable Document Format workflows. (github.com, github.com) The API quickstart lists four Anthropic-managed skills by short name: `pptx`, `xlsx`, `docx`, and `pdf`. It says developers can list them through the Skills endpoint before calling them in requests. (platform.claude.com) Anthropic’s retention note is unusually direct for a feature page. The overview says the feature is “not eligible” for Zero Data Retention, and the mirrored markdown version says skill definitions and execution data follow Anthropic’s standard retention policy. (platform.claude.com, docs.claude.com) That puts a boundary around how enterprises can use the feature. Claude Code’s settings and directory docs say teams can share skills through repository-level `.claude` folders and git, but the retention terms mean the feature does not inherit the stricter contractual handling some customers use for other workflows. (code.claude.com, platform.claude.com) Anthropic is framing Skills as reusable building blocks for automation, but the documentation now makes the tradeoff plain: easier reuse, with standard retention attached. (platform.claude.com, code.claude.com)