The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes vulnerabilities based on reports generated by external auditing tools, which could allow malicious content in the scanned environment to influence the agent's reasoning.
Ingestion points: Processes scan results from results.xml, report.html, and /tmp/lynis-report.dat as mentioned in SKILL.md.
Boundary markers: No specific boundary markers or instructions to treat the report data as untrusted are defined.
Capability inventory: The skill uses apt for package installation, docker for image execution, and executes multiple auditing tools that access system configuration files.
Sanitization: There are no specified sanitization or validation steps for the audit reports before they are processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill uses the system package manager to install security tools (apt install openscap-scanner, apt install lynis) and downloads the aquasec/kube-bench container image from a public registry for Kubernetes auditing.
[COMMAND_EXECUTION]: The skill executes several system audit commands which require high privileges to review sensitive configuration files in directories like /etc and /var to verify benchmark compliance.

cis-benchmarks