The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, potentially untrusted data such as Dockerfiles, container logs, and image manifests. This creates a surface where malicious content could attempt to influence the agent's logic.
Ingestion points: Dockerfile content during edits, docker logs output, and results from docker manifest inspect.
Boundary markers: The instructions do not define specific delimiters or warnings to isolate untrusted data during processing.
Capability inventory: The allowed-tools configuration strictly limits the agent to read-only and metadata-focused Docker subcommands (e.g., inspect, ps, logs, history, manifest inspect). It explicitly omits commands for container execution, image building, or network management.
Sanitization: No explicit sanitization or validation steps are provided for the processed data.
[COMMAND_EXECUTION]: The skill provides instructions to use various docker and docker compose commands to gather information about the daemon and container environment. These operations are restricted to a safe, informational subset in the skill's tool configuration.
[SAFE]: The skill promotes several security best practices, including avoiding recursive chmod on shared host checkouts, suppressing the logging of secrets during container startup, and using stable image tags. It references well-known services such as Homebrew and Amazon ECR for legitimate operational tasks. The skill also includes a post-completion step to invoke a vendor-specific tool, wk-learn.

wk-docker