The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data which is then interpolated directly into prompts used for image and video generation tools across multiple phases.
Ingestion points: In SKILL.md, the user-provided inputs character_description, environment_description, action_script, and style_direction are used as placeholders within the prompts for each generation phase.
Boundary markers: Absent. User input is placed directly into template strings (e.g., Character reference sheet of {{character_description}}...) without delimiters or instructions for the model to ignore embedded commands.
Capability inventory: The skill orchestrates multiple AI models (gpt-image-2, nano-banana-2, seedance-v2.0-i2v) via the muapi CLI and external API requests to api.muapi.ai.
Sanitization: Absent. There are no mechanisms described in the instructions to validate, filter, or escape the user-provided content before it is processed by the downstream generation models.

muapi-ai-fight-scene