The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to use npx add-mcp from a well-known service's repository and various vendor-specific packages (e.g., @modelcontextprotocol/server-postgres, @github/mcp-server). These references target established organizations and well-known services.
[COMMAND_EXECUTION]: Provides instructions for executing CLI commands to add, list, and remove MCP servers using tools like claude, npx, opencode, and codex. It includes guidance on using environment variables and handling Windows-specific command wrappers.
[INDIRECT_PROMPT_INJECTION]: The skill manages tools (MCP servers) that ingest external data, creating an attack surface for indirect prompt injection. The documentation includes a troubleshooting section that warns users about the risks of third-party servers fetching untrusted content.
Ingestion points: MCP tools process data from various external sources such as APIs, files, and web pages (identified in references/search.md and references/transports.md).
Boundary markers: The skill does not define specific prompt boundary markers, as it relies on the underlying agent's implementation for safety.
Capability inventory: Capabilities include network requests, file system access, and database operations via the installed MCP servers (identified in SKILL.md and references/multi-agent.md).
Sanitization: No specific sanitization logic is provided within the skill instructions, as it focuses on configuration and management rather than data processing.
[CREDENTIALS_SAFE]: Mentions the use of environment variables and headers for authentication but correctly uses placeholders (e.g., ${API_KEY}, TOKEN) and recommends secure practices like using native agent authentication commands (claude mcp login).

mcp-management