suede-campaign-in-a-box
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-provided creative assets and textual descriptions, which creates an indirect prompt injection surface.
- Ingestion points:
references/lanes.mdindicates the skill ingests artist lyrics, notes, timestamps, and fan comments (Lanes 4, 6, and 10). - Boundary markers: The skill includes an 'Evidence boundaries' section in
SKILL.mdthat defines non-negotiable operational limits and explicitly warns against following unverified claims. - Capability inventory: The skill can spawn sub-agents (multi-agent mode) and generate text-based campaign assets such as social media captions, emails, and site copy.
- Sanitization: A 'Public Copy Gate' is implemented in
SKILL.mdto review and refine generated copy, filtering out generic or unsupported AI-generated claims. - [SAFE]: No malicious code, obfuscation, or unauthorized data access patterns were detected. The skill follows security best practices by implementing clear 'Evidence boundaries' and 'Red flags' that restrict its autonomy over critical legal and financial decisions.
Audit Metadata