Anthropic published its Agent Skills specification on December 18, 2025, introducing a standardized Markdown-based format for AI agent workflows. OpenAI shipped support within 48 hours, adopting the exact format rather than building a proprietary alternative.
By March 2026—approximately 12 weeks after release—around 30 agent products supported the specification. Adopters included Microsoft Copilot, Cursor, JetBrains AI tooling, Google's Gemini CLI, Block's Goose and Anthropic's Claude Code.
The SKILL.md format consists of a folder containing a Markdown file and a small YAML header with two required fields: name and description. Optional elements include bundled scripts, reference documents and additional Markdown files. The format requires no SDK, vendor library or proprietary loader.
The specification's rapid adoption has concrete business implications. For companies like Bex.co, standardization determines whether internal runbook packs are written as portable SKILL.md files or as bespoke prompt formats, directly affecting development efficiency and interoperability costs.
The same quarter saw another convergence event: Anthropic donated its Machine Comprehension Protocol to the newly formed Agentic AI Foundation, operated under the Linux Foundation and co-founded by Anthropic, OpenAI and Block. AWS, Google, Microsoft, Cloudflare, GitHub and Bloomberg signed on as supporters.
MCP had achieved substantial scale before its formal transition: 97 million monthly SDK downloads and roughly 10,000 active servers. The protocol launched in November 2024.
MCP and Agent Skills address distinct problems. MCP provides the protocol for how an AI agent calls a tool—reaching a Postgres instance or triggering a deploy API. Agent Skills defines what the agent should do once it interacts with that tool or learns a workflow.
The open standard nature of SKILL.md functions across Claude, OpenAI, Gemini, open models via Groq or Together and local models through Ollama. This broad compatibility reduces vendor lock-in and lowers switching costs for enterprises building multi-model agent systems.
