The Agent Skills Directory

[COMMAND_EXECUTION]: The skill dynamically resolves the path to an executable engine named tzai-image and executes it using bash. The search logic looks for the engine in several hidden agent directories within the user's home folder.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied intent and prompt data to construct shell commands, creating a potential injection surface.
Ingestion points: User-provided inputs for kind and prompt specified in SKILL.md arguments.
Boundary markers: None identified; user input is directly compiled into a visual brief.
Capability inventory: The skill executes shell commands via bash using the resolved $ENGINE path.
Sanitization: No explicit escaping or sanitization of the user-provided prompt is performed before it is interpolated into the shell command template.

tzai-marketing