The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts scripts/agent.py and scripts/process.py use subprocess.check_output and subprocess.run to interact with the Docker daemon. These calls are used to execute commands like docker ps, docker version, and docker inspect to gather container metadata for auditing. The commands are constructed using static lists and specific IDs gathered from the Docker daemon itself, minimizing the risk of arbitrary command injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local environment by auditing Docker container configurations, which could potentially contain malicious payloads in container names or labels.
Ingestion points: The scripts in the scripts/ directory ingest data via the JSON output of docker inspect.
Boundary markers: The scripts rely on json.loads() to parse structured data, but do not implement specific boundary markers for the strings within the generated reports.
Capability inventory: The scripts have the capability to execute Docker CLI commands and write findings to a local JSON report file.
Sanitization: There is no specific sanitization of the container metadata strings before they are included in the final audit report, presenting a low-risk surface for indirect injection if an agent interprets the resulting report.

hardening-docker-containers-for-production