Anthropic Opens Agent Skills Standard, Continuing Its Pattern of Building Industry Infrastructure

Anthropic published Agent Skills as an open standard on December 18, releasing the specification and SDK at agentskills.io for any AI platform to adopt. The move extends Anthropic’s strategy of building industry infrastructure rather than proprietary moats—the same approach that made Model Context Protocol (MCP) ubiquitous.

Microsoft, OpenAI, Atlassian, Figma, Cursor, and GitHub have already adopted the standard. Partner-built skills from Canva, Stripe, Notion, and Zapier are available at launch.

What Agent Skills Are

Skills are directories containing instructions, scripts, and resources that AI agents can discover and load dynamically. Each skill requires a SKILL.md file with metadata describing its capabilities. When a user’s request matches a skill’s domain, the agent loads only the relevant information—a design Anthropic calls “progressive disclosure.”

The architecture solves a practical problem. Context windows are limited; stuffing every possible instruction into every request wastes resources. Skills let agents access specialized knowledge on demand without carrying it constantly.

A skill for PDF handling might include preferred libraries, edge cases, and output formatting. A skill for database operations could specify safety checks and rollback procedures. The instructions load only when the agent encounters those specific tasks.

Skill file example (Anthropic)

Following the MCP Playbook

Agent Skills follows the template Anthropic established with Model Context Protocol. MCP launched as an open standard for connecting AI systems to external tools, gained rapid adoption across competing platforms, and was donated to the Linux Foundation on December 9. Google, Microsoft, and AWS joined as foundation members.

The pattern is deliberate. Anthropic builds specifications that solve genuine interoperability problems, releases them as open standards, and lets adoption create value that accrues to the ecosystem rather than to Anthropic alone. In exchange, Anthropic establishes itself as the company that defines how AI infrastructure works.

The strategic logic: if skills become standard, Claude doesn’t need to be the only AI that uses them—it just needs to be the best at using them. Competing on execution rather than lock-in aligns with Anthropic’s positioning as the responsible AI company.

What This Means for the Industry

Skills portability addresses a real friction point for enterprises. Companies investing in AI customization face vendor lock-in if those customizations only work with one model provider. Skills written for Claude Code can now work with OpenAI’s Codex, Cursor, or any other platform that adopts the standard.

The skills convergence we reported earlier has now been formalized. OpenAI had already implemented a structurally identical system; the open standard codifies that convergence and invites others to join.

For developers, this creates a new distribution channel. A well-built skill can reach users across multiple AI platforms simultaneously. Anthropic’s partner directory at launch—Atlassian, Figma, Canva, Stripe, Notion, Zapier—represents significant reach for skills that solve enterprise workflows.

Enterprise Management Tools

Alongside the open standard, Anthropic announced organization-wide management tools for enterprise customers. Administrators can now enforce policies on which skills are available, control access to sensitive capabilities, and monitor skill usage across deployments.

The enterprise features position skills as IT-manageable infrastructure rather than ad-hoc customizations. For companies concerned about AI governance—what capabilities their AI systems have, who controls them, what guardrails exist—centralized skill management provides visibility and control.

The Bigger Picture

Anthropic has now contributed two foundational standards to AI infrastructure: MCP for tool connectivity and Agent Skills for capability customization. Both follow the same playbook: solve a real problem, release openly, drive adoption through usefulness rather than lock-in.

The approach contrasts sharply with OpenAI’s platform strategy. While OpenAI builds proprietary ecosystems—the GPT Store, the Apps SDK, platform-specific integrations—Anthropic builds standards that work everywhere. Both strategies can succeed; they optimize for different outcomes.

For the industry, open standards reduce fragmentation. Developers can build once and deploy across platforms. Enterprises can switch providers without rebuilding customizations. The competitive pressure shifts from ecosystem control to model quality and execution.

Anthropic is betting that’s a competition it can win. The Agent Skills standard is another step in making sure that competition happens on terms Anthropic helped define.