Micro-drama Video
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests raw user-provided content such as premises, treatments, or scripts to drive media generation tools, which creates a potential surface for indirect prompt injection attacks where the input content attempts to influence agent behavior or the media generation process.
- Ingestion points: User-supplied narrative content including premises, treatments, and approved scripts as described in the 'Required inputs' section of SKILL.md.
- Boundary markers: The skill incorporates human-in-the-loop review stages (e.g., 'approve the storyboard', 'acceptance checks') which mitigate accidental obedience, but it lacks specific instructions to delimit or quote user input when interpolating it into prompts for media models.
- Capability inventory: The skill possesses significant capabilities including image and video generation (media.generate_image, media.generate_video), file uploads (media.upload_file), and the ability to write manifests and receipts to the local workspace (workspace-write).
- Sanitization: There is no evidence of sanitization, escaping, or filtering of the user-provided narrative text before it is used by the agent or passed to external media generation services.
Audit Metadata