The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts scripts/agent.py and scripts/process.py utilize the subprocess module to perform system calls to docker, trivy, and kubectl. These commands are used to inspect image metadata, scan for vulnerabilities, and query container statuses from a Kubernetes cluster.
Evidence: subprocess.run() is used throughout both scripts to invoke CLI tools.
Security Evaluation: The implementation consistently uses argument lists rather than shell strings and avoids shell=True, effectively mitigating risks of command injection from untrusted image names.
[INDIRECT_PROMPT_INJECTION]: The scripts/process.py script identifies and processes container image names directly from Kubernetes cluster state, which constitutes an ingestion point for potentially untrusted data.
Ingestion points: scripts/process.py parses the output of kubectl get pods -o json to extract image strings.
Boundary markers: None present for the extracted strings.
Capability inventory: The extracted strings are passed as arguments to docker inspect and trivy image via subprocess.run.
Sanitization: While the strings are not sanitized, the use of list-based command execution prevents typical shell injection vectors. The risk is limited to potential LLM confusion if a malicious image name contains adversarial instructions intended for the agent's context.

implementing-container-image-minimal-base-with-distroless