The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on executing a local binary (typically named ms) to perform its core functions. It provides a Python script, scripts/mcp-search.py, which uses the subprocess module to manage this execution.
Evidence: The scripts/mcp-search.py file uses subprocess.Popen to launch the binary path specified in the MS_BIN environment variable with the mcp serve arguments.
Context: This execution is the intended primary function of the skill, serving as a gateway to the external meta_skill search engine.
[PROMPT_INJECTION]: As a retrieval-focused skill, it possesses an attack surface for indirect prompt injection, as it loads and processes content from external repositories.
Ingestion points: The skill ingests data through ms search and ms load operations which read from local skill directories and potentially remote repositories.
Boundary markers: SKILL.md explicitly warns that ms is "retrieval-only" and that the calling agent owns all validation and subsequent decisions, emphasizing that retrieved results are advisory.
Capability inventory: The skill has the capability to execute the ms binary and write feedback and index state to the local filesystem (e.g., ~/Library/Application Support/ms/).
Sanitization: The skill serves as a transport layer and does not perform internal sanitization of the skill content it retrieves, relying instead on the downstream agent's own safety protocols.

ms