The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py and scripts/process.py execute the docker binary using subprocess.run to perform security audits and container inspections.
[PRIVILEGE_ESCALATION]: The skill uses privileged container flags (--net host, --pid host, --cap-add audit_control) and mounts the Docker socket and host directories to allow the audit tool to access the host configuration. Documentation also includes sudo commands for hardening system files.
[EXTERNAL_DOWNLOADS]: The workflow involves pulling and running the docker/docker-bench-security image from the public Docker registry.
[INDIRECT_PROMPT_INJECTION]: The skill ingests potentially untrusted metadata from container configurations and external tool output (ingestion points in scripts/agent.py). It lacks explicit boundary markers or sanitization for this data. The script's capability inventory includes executing subprocess calls to perform the audit.

performing-docker-bench-security-assessment