The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The skill suggests running standard runtime version commands (e.g., node --version, python --version) to detect the appropriate base image. These are routine, safe operations used to match the production environment to the developer's local setup.
[PRIVILEGE_ESCALATION] (SAFE): The skill actively promotes security by including instructions to create and use a non-root user (USER appuser) within the generated Dockerfiles, reducing the attack surface of the resulting containers.
[DATA_EXFILTRATION] (SAFE): Analysis of the skill reveals no attempts to access sensitive file paths (like ~/.ssh or .env files for export) or make unauthorized network requests. It focuses strictly on project configuration files like package.json and go.mod.
[PROMPT_INJECTION] (SAFE): The instructions are task-oriented and do not contain patterns designed to bypass AI safety constraints, extract system prompts, or induce 'jailbroken' behavior.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill ingests untrusted data by reading project configuration and lockfiles. While this is a potential vector for indirect injection (e.g., malicious content in a package.json field), the skill's logic is constrained to specific functional keys (dependencies, engines), which significantly mitigates the risk.
[DYNAMIC_EXECUTION] (SAFE): The generated Dockerfiles include a HEALTHCHECK that uses node -e. This is a standard and safe method for verifying application health in minimal container images where utilities like curl are absent.

dockerize