The Agent Skills Directory

[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill instructs the user to install the mcporter package globally from the public NPM registry (npm install -g mcporter). This introduces an external dependency that is not from a pre-verified trusted source.
[COMMAND_EXECUTION]: The skill relies on executing shell commands via the mcporter CLI to interact with servers and the mkdir command to create directories for newly generated skills.
[INDIRECT_PROMPT_INJECTION]: The skill retrieves tool schemas from external MCP servers and uses that content to generate new SKILL.md files. This creates a vulnerability where a malicious MCP server could provide descriptions designed to inject instructions into the agent's persistent logic library.
Ingestion points: Data retrieved from external MCP servers via mcporter list --schema (as seen in SKILL.md).
Boundary markers: Absent. The schema content is interpolated directly into markdown templates without delimiters or warnings to ignore embedded instructions.
Capability inventory: The agent can write files to its own skills directory (~/skills/) and execute subprocesses via the mcporter command (documented in SKILL.md).
Sanitization: No sanitization or validation of the external schema data is performed before writing it to the file system.
[DYNAMIC_EXECUTION]: The skill dynamically creates and writes new operational logic in the form of SKILL.md files based on templates and external data inputs retrieved at runtime.

mcporter