The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to use external web search results to determine connection details and command-line arguments for MCP servers if the user does not provide them. This creates a surface where the agent could ingest and apply malicious configuration strings from untrusted web sources.
Ingestion points: Server connection details are gathered from user input and potentially WebSearch results (SKILL.md, Step 2).
Boundary markers: Absent. The skill does not provide instructions to validate the integrity or safety of the configuration data retrieved from the web.
Capability inventory: The skill has the capability to write configuration files (.mcp.json) to the local filesystem, which define commands that the terminal environment will execute.
Sanitization: Absent. The skill focuses on preserving existing configuration structure but does not include steps to sanitize or verify the commands or URLs being added to the configuration.
[COMMAND_EXECUTION]: The skill guides the agent to write configurations that involve executable commands (e.g., npx, node) which are automatically run by the Warp terminal upon detection.
Evidence: Step 4 provides templates for "Command-based server (stdio transport)" that include command and args fields for defining executable processes.
[EXTERNAL_DOWNLOADS]: The configuration examples encouraged by the skill often use npx -y, a tool that automatically downloads and executes packages from the npm registry at runtime without further confirmation.
Evidence: The command-based server example in Step 4 uses "command": "npx" and "args": ["-y", "@scope/package-name"].

agent-add-mcp