The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided pet descriptions and names which are interpolated into prompts for image generation.
Ingestion points: pet_name, description, and pet_notes parameters in SKILL.md and scripts/prepare_pet_run.py.
Boundary markers: The skill lacks explicit boundary markers or "ignore instructions" warnings around the user-supplied strings when constructing prompts for $imagegen or subagents.
Capability inventory: The skill performs file system writes (atlas composition, packaging to the .codex directory), executes local Python scripts via subprocess, and performs network requests via the $imagegen skill or a fallback OpenAI API.
Sanitization: Filenames and slugs are sanitized using a slugify function, and scripts/generate_pet_images.py contains explicit path traversal validation for input and output paths to prevent escaping the run directory.
[COMMAND_EXECUTION]: The skill utilizes subprocess.run to coordinate its complex multi-step workflow by executing its own bundled Python scripts.
Evidence: scripts/finalize_pet_run.py executes extract_strip_frames.py, inspect_frames.py, compose_atlas.py, and other specialized scripts within the skill's package.
Evidence: scripts/render_animation_videos.py executes the ffmpeg CLI tool to generate preview videos of the pet animations.

hatch-pet