film-style-library
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests external, untrusted data to generate prompts and configuration files.
- Ingestion points: Processes user-supplied images and project-specific references from the workspace to infer visual directions.
- Boundary markers: Uses a 'Style lock file' (
style.json) as a stable reference point, which acts as a logical boundary for downstream tool invocations. - Capability inventory: Performs local file writes to
films/<slug>/style.jsonand invokes external image/video generation tools available in the environment. - Sanitization: Implements a 'Guard later references' step that validates new inputs against the locked style ID and silhouette/palette rules, stopping execution if a conflict is detected.
Audit Metadata