The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-controlled data that is directly interpolated into prompts for the Gemini Image API, which presents an attack surface for indirect prompt injection.
Ingestion points: The scripts/generate_storyboard.py script loads untrusted data from script.json files (specifically the image_prompt, visual.description, and action fields) via the ProjectManager.load_script method.
Boundary markers: The prompt construction functions (build_grid_prompt, build_scene_prompt, and build_direct_scene_prompt) concatenate this data directly into the final prompt strings without using delimiters or instructional guards to prevent the model from following instructions embedded in the data.
Capability inventory: The skill can write files to the project directory (projects/{name}/storyboards/), update project script metadata via ProjectManager.update_scene_asset, and perform network operations to the Gemini API.
Sanitization: There is no evidence of sanitization, filtering, or validation of the text retrieved from the script files before it is passed to the LLM for image generation.

generate-storyboard