character-consistency
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied prompts and character reference datasets to generate instructions for external image models, creating a surface where untrusted input is interpolated into prompts for external APIs.
- Ingestion points: User-provided character descriptions and reference image datasets as described in SKILL.md and references/approach-3-dataset-system.md.
- Boundary markers: The skill recommends delta-only prompting in references/approach-1-guidelines.md to minimize user-controlled text.
- Capability inventory: Ability to invoke image generation tools and external APIs (OpenAI, Google, fal, Replicate) as specified in the per-generation loop in SKILL.md.
- Sanitization: Instructions suggest trimming user prompts to focus on changes, though no programmatic sanitization is defined.
Audit Metadata