The Agent Skills Directory

[COMMAND_EXECUTION]: The mcporter list --stdio and mcporter call --stdio commands allow for the execution of arbitrary shell strings to initialize and communicate with local MCP servers. While this is a standard method for interacting with stdio-based MCP servers, it represents a direct command execution vector if the command string is influenced by untrusted external data.
[DATA_EXFILTRATION]: The CLI supports a shorthand @filename syntax (e.g., body=@comment.md) to read the contents of local files and pass them as arguments to tools. This capability could be abused to read sensitive files (such as .env or SSH keys) if an agent is instructed to use a sensitive path as a tool argument.
[EXTERNAL_DOWNLOADS]: The skill facilitates connections to ad-hoc external MCP servers via URLs (e.g., https://mcp.linear.app/mcp). This allows the agent to communicate with arbitrary remote infrastructure, which is the intended purpose of the tool but remains an external communication channel.
[INDIRECT_PROMPT_INJECTION]: The skill functions by ingesting tool signatures, documentation, and data from local or remote MCP servers to populate the agent's context.
Ingestion points: mcporter list fetches tool schemas and docstrings from external servers.
Boundary markers: The skill does not implement explicit delimiters to separate external tool documentation from system instructions.
Capability inventory: The skill allows network operations (tool calls), shell command execution (stdio servers), and local file reading.
Sanitization: There is no evidence of sanitization or filtering of the metadata returned by MCP servers before it is presented to the agent. A malicious server could provide crafted docstrings designed to influence the agent's behavior.

mcporter