The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (LOW): The skill instructs users to install gemini-cli globally via npm. Per [TRUST-SCOPE-RULE], this is downgraded to LOW/INFO as it targets a recognized tool within the trusted Gemini ecosystem (google-gemini/gemini-cli).
[INDIRECT_PROMPT_INJECTION] (LOW): The skill facilitates the execution of MCP tools based on user-provided strings. This is a standard functional surface for indirect prompt injection.
Ingestion points: User-provided task descriptions passed to the gemini command or scripts/cli.ts.
Boundary markers: None explicitly defined in the provided skill documentation; relies on underlying CLI tools.
Capability inventory: Execution of arbitrary tools defined in local MCP server configurations via npx tsx subprocesses.
Sanitization: Not present in the skill instructions; sanitization is deferred to the LLM and the MCP client implementation.
[COMMAND_EXECUTION] (SAFE): Uses standard CLI commands (mkdir, ln, npx tsx) to manage configurations and execute local helper scripts. These actions are consistent with the skill's stated purpose of managing MCP integrations.
[CREDENTIALS_UNSAFE] (SAFE): While the skill references .claude/.mcp.json (which typically contains API keys or server endpoints), it only performs local configuration tasks like symlinking and reading for client initialization. No exfiltration patterns were detected.

mcp-management