style-clone
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill uses a 'copy-then-refine' workflow where it duplicates existing slide layouts (
duplicate_scene) and updates specific text elements using JSON Pointer paths (patch_stage). This approach limits the agent's ability to execute arbitrary code or commands by scoping modifications to defined JSON structures. - [SAFE]: The skill incorporates specific safety instructions regarding identity, stating that it should clone 'style, not identity' and must not fabricate biographies or endorsements attributed to the original author. This aligns with AI safety guidelines regarding impersonation.
- [SAFE]: The skill manages external data (PowerPoint files) via an
import_pptxtool. While it processes untrusted input, the instructions prioritize identifying and replacing all original text ('residue') with new content. This content-replacement focus, combined with the platform's 'fails loud' behavior for unsafe HTML, significantly mitigates the surface area for indirect prompt injection.
Audit Metadata