The Agent Skills Directory

[COMMAND_EXECUTION]: The skill searches for a local executable script named tzai-image within several agent-specific directories (e.g., ~/.agents/skills/tzai-image/scripts/tzai-image) and executes it using bash to generate images.
[COMMAND_EXECUTION]: User-provided input is interpolated into a shell command: bash "$ENGINE" cover --prompt "<user subject>". While wrapped in double quotes, this pattern relies on the AI agent's internal sanitization to prevent potential shell command injection if the user input contains backticks or subshells.
[PROMPT_INJECTION]: The skill processes untrusted user data (the image subject) and incorporates it directly into a functional command execution path without explicit boundary markers or sanitization logic within the script itself.
Ingestion points: User subject text provided via the /tzai-cover slash command or natural language request (found in SKILL.md).
Boundary markers: Absent; the user input is placed directly into the --prompt argument of the shell command.
Capability inventory: Executes shell scripts via bash and calls the tzai-image CLI tool (found in SKILL.md).
Sanitization: Absent; the skill does not perform escaping or validation on the <user subject> placeholder.

tzai-cover