The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates a complex workflow by executing internal Python scripts using subprocess.run(). For example, finalize_pet_run.py calls scripts for frame extraction, inspection, and atlas composition. These operations are performed using secure argument lists and are limited to the skill's own directory.
[EXTERNAL_DOWNLOADS]: A fallback script (generate_pet_images.py) is included to generate pet images via OpenAI's official API when system-level tools are unavailable. The script handles API communication securely and relies on environment-stored API keys rather than hardcoded credentials.
[INDIRECT_PROMPT_INJECTION]: The skill processes external visual data provided by the user to ground the pet generation process. This creates a surface for potential prompt injection via embedded instructions in design references.
Ingestion points: User-provided images are passed as command-line arguments to prepare_pet_run.py and subsequently processed by visual generation tools.
Boundary markers: The skill defines strict 'Identity Lock' and 'Transparency and Artifact' rules within the prompts passed to generation agents to ensure consistent behavior.
Capability inventory: The skill maintains the ability to write to the local filesystem for asset packaging and can execute sub-processes for media rendering.
Sanitization: The skill implements comprehensive path traversal checks in its image processing scripts (validated by test_generate_pet_images.py) to prevent unauthorized file access outside the designated run directory.

hatch-pet