lov-riso-portrait
Pass
Audited by Gen Agent Trust Hub on Sep 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided photographs to serve as the identity reference for generative image editing, creating a potential surface for indirect injection via image content.
- Ingestion points: User-uploaded images are parsed during the source inspection step (SKILL.md).
- Boundary markers: Strict prompt construction logic in
references/riso-art-direction.mdenforces identity and style constraints, serving as a boundary for model behavior. - Capability inventory: The skill utilizes
gpt-image-2for raster image editing and provides scripts for local profile persistence (scripts/profile_store.py). - Sanitization: The workflow includes a mandatory 'Quality Gate' (references/quality-gate.md) to manually verify anatomy, identity, and framing facts against the original source.
Audit Metadata