The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill contains an indirect prompt injection surface due to its core function of ingesting and processing untrusted data from user goals and media references (images, video, audio) to construct generative video prompts. Ingestion points: User-provided goals, production phases, and media references identified in the intake step of SKILL.md. Boundary markers: Present; the skill provides instructions to 'preserve reference tags exactly' and 'keep prompts compact' to maintain focus and instruction integrity. Capability inventory: The skill aggregates data to build production prompts and routes tasks to specialized sub-skills for camera control, motion, and VFX. Sanitization: Present; the operating loop in SKILL.md includes a 'Safety gate' step that explicitly routes IP, likeness, and graphic wording through dedicated safety skills such as seedance-copyright and seedance-filter.

seedance-20