The Agent Skills Directory

[SAFE]: The skill serves as an architectural and procedural manual for developing Model Context Protocol (MCP) servers. It provides detailed instructions on implementing OAuth, managing API contracts, and following directory listing requirements for services like Anthropic's Claude and OpenAI's ChatGPT.
[EXTERNAL_DOWNLOADS]: The skill mentions standard developer tools and dependency management practices. It references installing packages via npm or npx (e.g., @anthropic-ai/mcpb, mcp-publisher, mcp-remote) and downloading release assets from GitHub repositories. These are documented as part of the standard deployment and distribution workflow for MCP servers.
[COMMAND_EXECUTION]: Instructions include standard shell commands for development workflows, such as gh api user, npm publish, and git tag. These are used for legitimate repository management and publishing tasks by the developer.
[CREDENTIALS_UNSAFE]: The skill follows security best practices for credential management. It explicitly instructs the user to store tokens in environment variables (e.g., Authorization: Bearer ${TOKEN}) rather than hardcoding them. It also mandates that tokens must never appear in application logs or shell history and that sensitive error text must be redacted.
[INDIRECT_PROMPT_INJECTION]: The skill identifies the surface area for indirect prompt injection when processing user-written text and mandates that such content be treated as untrusted. It recommends mitigations such as strict schema validation using zod and ajv to ensure that tool outputs conform to expected structures, preventing the model from misinterpreting data as instructions.

mcp-server