brand-designer
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user content through markdown templates in the
templates/directory. This content is later interpolated into compiled brand books viascripts/brand-bookand into image generation prompts viascripts/generatewithout sanitization or boundary markers. While this creates a surface for indirect instructions to influence agent behavior, it is consistent with the skill's primary documentation purpose and the risk is considered minimal.\n - Ingestion points: User-filled markdown templates (e.g.,
strategy.md,visual-id.md).\n - Boundary markers: None identified in compiled outputs or generated prompts.\n
- Capability inventory: File read/write operations and prompt construction for external image generation.\n
- Sanitization: None; the scripts use direct text replacement and concatenation.\n- [SAFE]: All script operations are local and perform standard file management. The Python scripts use
yaml.safe_load(when PyYAML is present) and include a manual fallback parser that avoids unsafe dynamic execution methods likeeval().
Audit Metadata