The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill downloads and executes code from the npm registry at runtime using npx @smithery/cli to facilitate the installation of servers.
[DATA_EXFILTRATION]: Accesses sensitive application configuration files in protected directories (e.g., ~/Library/Application Support/Claude/claude_desktop_config.json, %APPDATA%/Claude/, and various VSCode extension storage paths). These files are known to contain API keys, authentication tokens, and private user data.
[COMMAND_EXECUTION]: Uses the rg (ripgrep) utility to perform recursive file discovery across configuration roots to locate MCP-related JSON files.
[CREDENTIALS_UNSAFE]: The logic explicitly targets and manipulates configuration fields containing high-value secrets, including apiKey, accessToken, token, and Authorization headers.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: Reads existing JSON configuration files from multiple third-party applications (Claude, Cursor, Cline, etc.).
Boundary markers: Absent; the skill merges external JSON data directly into configuration objects without sanitization or delimiters.
Capability inventory: Includes the ability to execute npx commands and write to arbitrary application configuration paths.
Sanitization: No evidence of validation or sanitization for the content of the JSON files being read or the names of the servers being merged.

mcp-installer