The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use local system utilities such as sips (on macOS) or python3 (with PIL) to verify the actual pixel dimensions of generated images. This is a functional requirement for its quality assurance step.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, including user-provided images (photos, background frames) and text (topics, video scripts). It mitigates risk by referencing a central safety document (家族公约.md) and instructing the agent to validate cover text against safety "red lines."
Ingestion points: User-provided photos, background images, and video topics or scripts (Step 1).
Boundary markers: Relies on external protocol files (aikey/references/家族公约.md) to define safety boundaries.
Capability inventory: File system access within a designated vault, execution of image generation tools, and shell-based image metadata reading.
Sanitization: Explicitly instructs the agent to remove characters like / from the user-provided "topic" before using it as a directory name to prevent path traversal.
[DATA_EXPOSURE]: The skill explicitly mandates that API credentials must be read from environment variables and never stored in files, adhering to secret management best practices.

aikey-cover