The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several system tools including docker, crane, curl, and trufflehog to interact with network services, pull external container images, and perform filesystem operations during secret scanning.
[CREDENTIALS_UNSAFE]: The instructions direct the agent to locate and extract highly sensitive data such as .env files, cloud provider keys, kubeconfig files, and other authentication tokens that may be baked into container image layers.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data ingested from external container registries. This creates an attack surface where a malicious registry or container image could provide crafted content designed to influence the agent's subsequent analysis or actions.
Ingestion points: Container registry catalogs, image tags, and image layers via curl, docker pull, and crane (specified in SKILL.md).
Boundary markers: None identified; instructions do not explicitly state to ignore embedded commands in the scanned data.
Capability inventory: Subprocess calls for docker, crane, curl, and trufflehog across all scanning steps.
Sanitization: No specific sanitization or filtering of external image content is mentioned before processing.

cloud-docker-registry