The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions require the agent to ingest external, untrusted data such as scripts, briefs, and brand treatments to generate detailed prompt plans for other AI models. This creates an attack surface where hidden instructions in the user input could be propagated to downstream systems.
Ingestion points: Documented in SKILL.md within the "Intake checklist" section, which identifies inputs like scripts, briefs, and references.
Boundary markers: The skill lacks instructions for using delimiters or explicit warnings to the agent to treat information inside user-provided scripts as data and not as instructions.
Capability inventory: The skill produces structured "AI-generation prompt planning" and "shot-board" data (SKILL.md) which are directly used as inputs for image and video generation tools.
Sanitization: There are no sanitization or validation steps mentioned to filter the content of user-provided scripts before they are incorporated into generated prompts.
[EXTERNAL_DOWNLOADS]: The skill references multiple external URLs for professional filmmaking guidelines and platform-specific technical data.
Evidence: The skill fetches and cites official specifications from well-known organizations including Autodesk, Adobe, Unreal Engine, TikTok, Google (YouTube), and Meta (Facebook) in the "Source ledger" of SKILL.md to ensure production accuracy.

storyboard-previsualization