The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the 'paramiko' library to execute remote shell commands (e.g., dpkg-query, rpm, uname, ss) on Linux targets and 'winrm' for PowerShell commands on Windows targets to gather system inventory and telemetry.
[EXTERNAL_DOWNLOADS]: Fetches vulnerability definitions and CVE data from the official NIST National Vulnerability Database API (services.nvd.nist.gov). This is a well-known and trusted governmental source.
[INDIRECT_PROMPT_INJECTION]: The skill ingests system metadata and package lists from external remote hosts, which constitutes an ingestion surface for untrusted data. 1. Ingestion points: Remote command output captured in scripts/process.py and scripts/agent.py. 2. Boundary markers: Data is processed into structured JSON and CSV formats for reporting, though no explicit prompt isolation delimiters are used for the displayed output. 3. Capability inventory: Remote command execution via SSH/WinRM and network access to cloud APIs. 4. Sanitization: System responses are parsed as strings but are not specifically sanitized against potential embedded prompt instructions.
[SAFE]: AWS operations use the official boto3 SDK to interact with Inspector2 and EC2 APIs for snapshot-based analysis. These interactions are consistent with the skill's intended functionality for cloud-native vulnerability scanning.

performing-agentless-vulnerability-scanning