The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the uv run mcp-scanner CLI tool to perform security audits. It includes a functional mode for scanning stdio servers where it executes commands to launch and analyze local MCP server processes.
[DATA_EXPOSURE]: To perform its auditing functions, the skill accesses well-known MCP configuration paths across various IDEs and applications, including Windsurf, Cursor, and Claude Desktop. This access is limited to reading configuration metadata for security analysis.
[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for processing untrusted data by scanning external MCP servers, prompts, and resources. While this is the tool's primary function (to detect threats within that data), the ingestion of external content for LLM-based semantic analysis represents a theoretical attack surface.
Ingestion points: Remote MCP server URLs, local configuration files, and external resource URIs.
Boundary markers: Not explicitly defined in the skill instructions.
Capability inventory: The skill has the capability to read local files, perform network requests to servers and the Cisco AI Defense API, and execute commands via uv.
Sanitization: The underlying mcp-scanner tool is specifically designed to identify and mitigate malicious patterns in these inputs.

mcp-security-scan