The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-supplied text prompts and image data to generate or modify visuals. This is the primary function of the skill. Evidence chain:
Ingestion points: User prompts are ingested via the host's image_gen tool, CLI arguments (--prompt, --prompt-file), and JSONL batch files processed by scripts/image_gen.py.
Boundary markers: The skill uses a structured prompt schema (e.g., 'Primary request:', 'Constraints:') to organize inputs, but does not employ strict delimiters to isolate embedded instructions from the agent's context.
Capability inventory: The skill has file-write capabilities in the output/imagegen/ directory and performs network requests to the OpenAI API.
Sanitization: CLI inputs are validated for format, size, and type, but the text prompts themselves are not semantically filtered.
[EXTERNAL_DOWNLOADS]: The skill specifies requirements for the openai and pillow Python packages. These are standard, well-known libraries used for API communication and local image manipulation.
[COMMAND_EXECUTION]: The skill documents and uses several shell commands to execute its bundled Python scripts (scripts/image_gen.py and scripts/remove_chroma_key.py) for processing images and running batch generation tasks.

imagegen