The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands to inventory existing MCP servers.
Evidence: Instructions to run claude mcp list and other client-documented subcommands in SKILL.md (Step 1).
Mitigations: The skill requires explicit informed consent from the user before execution, forbids inventing subcommands or flags, and warns about potential process side effects during health checks.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local discovery tools, which could contain instructions intended to influence the agent.
Ingestion points: Output from client listing commands or user-provided manual inventories in SKILL.md (Step 1).
Boundary markers: The agent is instructed to present all discovery findings to the user for review and confirmation before proceeding to registration in SKILL.md (Step 3).
Capability inventory: The skill can register remote MCP servers and configure network access settings via platform tools like register_catalog_mcp and set_mcp_network_access.
Sanitization: Strict rules are defined in SKILL.md (Step 2 and Step 9) to strip credentials, environment variables, and sensitive URL parameters from all ingested data and diagnostic logs.

add-existing-mcp-servers