The Agent Skills Directory

[SAFE]: No malicious patterns or behaviors were detected in the skill's code or metadata.
[COMMAND_EXECUTION]: The script scripts/health_check.py uses subprocess.run to execute the which command. This is used solely to verify the existence of configured MCP server executables on the system path, which is a safe and common administrative check.
[DATA_EXFILTRATION]: The skill accesses ~/.claude.json to manage MCP server settings. The analysis confirmed that this access is local-only and necessary for the skill's primary management functions, with no evidence of data being transmitted externally.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: User-provided task descriptions are processed by the recommend command in scripts/mcp_manager.py.
Boundary markers: No explicit boundary markers are present in the recommendation logic.
Capability inventory: Subprocess execution is limited to environmental checks (which).
Sanitization: The input is used for string-based matching against a static capability library, which does not present a path for command injection or agent subversion.

mcp-manager