The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides installation commands for various MCP servers from trusted organizations and well-known services. Examples include official reference implementations from github.com/modelcontextprotocol/servers, and tools from Microsoft, GitHub, Cloudflare, Supabase, Neon, Google, and Stripe. These tools are distributed through official registries using standard package managers like npx and uvx.
[COMMAND_EXECUTION]: The skill instructs the agent on using the claude mcp add command to configure the environment. This is a legitimate administrative feature of the platform for extending agent capabilities and requires explicit configuration scope.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a task router, mapping user workflow needs to specific MCP servers.
Ingestion points: The skill processes user descriptions of tasks in SKILL.md and reads local configuration files (~/.claude.json, .mcp.json).
Boundary markers: None explicitly defined in the instructional text.
Capability inventory: File system access, network operations (via installed servers), and tool registration capabilities.
Sanitization: Not applicable as the skill primarily provides a static catalog and guidance.
This represents an architectural surface area common to routing skills and is managed by the platform's core safety guardrails.

mcp-mastery