This site is written from current official or primary documentation wherever possible. The annotations below describe how each source is used. Machine-readable summaries are available at llms.txt and llms-full.txt.
| Source | Publisher | Used for |
|---|---|---|
| Agent Skills Overview | Agent Skills | Defines Agent Skills as a lightweight open format built around SKILL.md plus optional scripts, references, assets, and progressive disclosure. |
| Agent Skills Specification | Agent Skills | Documents the required directory structure, SKILL.md frontmatter, name and description constraints, optional directories, and validation guidance. |
| Best practices for skill creators | Agent Skills | Gives authoring guidance on scope, context budgeting, progressive disclosure, testing, and validation loops. |
| Agent Skills overview | Claude Platform Docs | Explains how Claude uses pre-built and custom skills, including metadata, instructions, resources, Claude API support, Claude Code support, and claude.ai behavior. |
| Skill authoring best practices | Claude Platform Docs | Provides Claude-specific guidance on concise SKILL.md files, discovery descriptions, progressive disclosure, structure, and testing. |
| Equipping agents for the real world with Agent Skills | Anthropic Engineering | Explains the architecture idea behind skills and progressive disclosure using bundled files and code execution. |
| Agent Skills - Codex | OpenAI Developers | Documents Codex support for skills, including Codex CLI, IDE extension, Codex app, progressive disclosure, and plugin distribution. |
| Use Agent Skills in VS Code | Visual Studio Code Docs | Documents GitHub Copilot support for the Agent Skills format in VS Code, Copilot CLI, and Copilot cloud agent. |
| What is the Model Context Protocol? | Model Context Protocol | Introduces MCP as an open-source standard connecting AI applications to external systems, data sources, tools, and workflows. |
| Architecture overview | Model Context Protocol | Defines MCP hosts, clients, servers, transports, JSON-RPC data layer, and the server primitives: tools, resources, and prompts. |
| Versioning | Model Context Protocol | Confirms string-based MCP protocol versions and the current protocol version referenced by this site. |
| MCP Specification 2025-11-25 | Model Context Protocol | Authoritative latest protocol specification for MCP as checked on July 6, 2026, including features and security principles. |
| MCP Tools | Model Context Protocol | Specifies model-controlled tools, tool discovery, invocation, schemas, output schemas, and security expectations. |
| MCP Resources | Model Context Protocol | Specifies resources as URI-addressed contextual data exposed by MCP servers to clients. |
| MCP Prompts | Model Context Protocol | Specifies prompts as reusable interaction templates provided by MCP servers. |
| Build with Agent Skills | Model Context Protocol | Shows Agent Skills being used as portable instruction sets for MCP server design and implementation workflows. |
| Client Best Practices | Model Context Protocol | Discusses progressive tool discovery for MCP hosts at large tool counts, including a documented token-use comparison. |
| Claude Has Skills | Claude Has Skills | Complementary Claude-specific directory and reference, linked where Claude-only examples are useful. |
Source map by topic
Agent Skills format
Use the Agent Skills overview and specification for the portable directory shape, required frontmatter, optional directories, progressive disclosure, validation, and file-reference rules. Anthropic's overview is used for Claude-specific product behavior and architecture framing.
Skill authoring
Use the Agent Skills and Anthropic best-practices guides for scope, context budgeting, description writing, workflow design, scripts, testing, and progressive-disclosure structure.
Ecosystem support
Use OpenAI Codex docs for Codex support, VS Code docs for GitHub Copilot support, and Anthropic docs for Claude support. Portability claims on this site are conservative: documented support does not mean every script, permission, or packaging convention works unchanged everywhere.
MCP
Use MCP's introduction, architecture docs, versioning page, latest specification, and server feature pages for protocol claims. The MCP client best-practices page is used for the progressive tool discovery comparison and the relationship between skill-style discovery and tool-scale context management.