Agent Skills FAQ

Q: What is an Agent Skill?

A: An Agent Skill is a modular "knowledge package" designed to teach an AI agent how to perform a specialized task in a reliable, repeatable way. Unlike a generic prompt, a skill is a folder containing a SKILL.md file with Markdown instructions and metadata, along with optional scripts, templates, or reference documentation. These packages codify domain expertise—such as team coding standards or specific document workflows—allowing the agent to reuse that knowledge automatically.

Q: How do Skills work technically?

A: Skills utilize a mechanism called progressive disclosure (or lazy loading) to maintain efficiency. At startup, the agent loads only the name and description of the skill. The agent monitors the user's conversation, and if a request matches a skill's description, the agent dynamically loads the full instructions and context into the conversation. This approach prevents the agent's context window from being overwhelmed by unused instructions.

Q: How are Skills different from the Model Context Protocol (MCP)?

A: While they can work together, they serve different purposes:

• Skills (Know-How): Skills are Markdown files that provide "knowledge sharing," teaching the agent how to use tools or follow a process (e.g., "Review this PR using our specific security guidelines").
• MCP (Tools): MCP connects the agent to external systems and data (e.g., connecting to a database or executing code).

In short, MCP provides the "hands" (tools), while Skills provide the "brain" (methodology) for using those tools effectively.

Q: Is there a unified standard for Skills?

A: Yes. Capabilities across different platforms are built on the Agent Skills Open Specification (also referred to as the open agent skills standard). This specification defines the required structure, such as the SKILL.md file format and the YAML frontmatter (metadata like name and description) required for the skill to function. It also specifies the metadata formatted as XML tags that gets sent to the LLM or agent to pick a skill.

Q: How did Skills become a cross-platform standard?

A: The standard originated with Anthropic, which introduced Agent Skills (around October 2025) to extend Claude's capabilities. It became a de facto industry standard through broad adoption and compatibility:

• Open Specification: The explicit publication of the "Agent Skills Open Specification" allowed other developers to build compatible systems.
• Cross-Compatibility: Major platforms adopted the file structure defined by Anthropic. For example, GitHub Copilot and OpenCode were engineered to automatically discover and use skills located in .claude/skills directories, ensuring that a skill created for one agent often works for another without modification.

Q: Which AI agents and platforms support Skills?

A: As of late 2025, support for Agent Skills is available across several major developer tools:

• Claude Code: Supports skills via the .claude/skills directory, plugins, and the anthropics/skills repository.
• GitHub Copilot: Added support for Agent Skills on December 18, 2025. Copilot works with skills in the CLI, IDE, and coding agent, and it can read skills directly from .claude/skills repositories.
• OpenAI Codex: Supports skills via the CLI and IDE extensions, utilizing the .codex/skills directory or system-wide locations.
• OpenCode: Supports skills via .opencode/skill and maintains compatibility with .claude/skills.
• Skilz: With the Skilz installer, you can use Agent Skills with 14+ different platforms, including Gemini CLI, Aider, Cursor, Windsurf, Qwen Code, KIMI Code, and many more. You can even configure the Skilz installer to support tools that it does not currently support. The Skilz installer works with skillzwave.ai but it can also work standalone and can install from any git repo or GitHub repo URL or from the file system. It supports project installs as well as user home installs.

Q: Where can I find pre-made Skills?

A: Users can download curated skills from community repositories. Notable sources include the anthropics/skills repository on GitHub and the github/awesome-copilot collection. You can also browse thousands of skills here on Skillzwave. Platforms like Codex also include a $skill-installer tool to help users download skills from these repositories.