The Agent Skills Directory

[COMMAND_EXECUTION]: The nmap_custom_scan tool allows users to supply an arbitrary string of flags to the nmap binary. Although the skill documentation states that shell metacharacters and specific flags are blocked, this approach creates a substantial attack surface where sophisticated bypasses of the blocklist could lead to arbitrary command execution or system compromise.
[DATA_EXFILTRATION]: The nmap_get_scan tool retrieves files from disk based on a scan_id provided by the user. This functionality is inherently vulnerable to path traversal attacks (e.g., using ../ sequences) if the input is not strictly validated and restricted to the intended directory, potentially exposing sensitive system files.
[PROMPT_INJECTION]: The skill processes network scan data, which is an external and untrusted source of information, making it susceptible to indirect prompt injection attacks. \n
Ingestion points: Raw network scan output (banners, hostnames) and historical scan results retrieved from disk by nmap_get_scan. \n
Boundary markers: Not identified in the skill's documentation. \n
Capability inventory: Execution of the nmap binary via subprocess and file system read access via the scan retrieval tools. \n
Sanitization: The skill claims to filter shell metacharacters and specific nmap flags in the custom scan input tool.

nmap-scan-management