The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses dynamic context injection (the ! syntax) in SKILL.md to automatically execute shell commands like grep and cat when the skill is loaded. These commands are used for environment discovery and file inspection without user intervention.
[DATA_EXFILTRATION]: The skill automatically reads the contents of docker-compose.override.yml and Dockerfile.user via shell commands during the load process. These files often contain sensitive information such as environment variables, local development credentials, or API tokens. This process silently harvests the file contents into the agent's prompt context.
[PROMPT_INJECTION]: The skill captures user-provided tool names from $ARGUMENTS and interpolates them directly into a Dockerfile RUN command block. The absence of sanitization or validation logic allows for indirect prompt injection, where a malicious user could provide input that adds arbitrary Dockerfile instructions or shell commands to the build process.

docker-extend