The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the uv package manager from its official source at astral.sh. This is a well-known development tool used to manage the mcp-scan dependency from Invariant Labs. Both sources are recognized for providing security tooling and follow standard installation patterns.
[COMMAND_EXECUTION]: The script scripts/agent.py uses subprocess.run() to execute the mcp-scan utility. This is the primary function of the skill—orchestrating a security scan of local configuration files. The input is restricted to user-provided configuration paths.
[PROMPT_INJECTION]: While the skill contains regex patterns designed to detect prompt injection (e.g., POISON_PATTERNS), these are used for defensive analysis of third-party tools and do not represent an attempt to inject the agent itself. The skill includes explicit warnings about the risks of indirect prompt injection in MCP environments.
[DATA_EXFILTRATION]: The skill mentions sensitive file paths (e.g., .ssh, .env) and external URL targets exclusively within the context of security testing (SSRF probing and forensic detection of exfiltration patterns in other tools). It does not perform exfiltration of the user's data.

auditing-mcp-servers-for-tool-poisoning