email-generation
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data that could contain malicious instructions designed to subvert the agent's behavior.
- Ingestion points: Data is ingested from
campaigns/{slug}/accounts/*.jsonand contact CSV files in Step 1. - Boundary markers: There is no evidence of boundary markers (like XML tags) or "ignore embedded instructions" warnings being used when interpolating the contact context into the prompt template.
- Capability inventory: The skill performs file reads, file writes to
campaigns/{campaign-slug}/emails/, and invokes LLM text generation. - Sanitization: The skill performs quality validation (banned words, generic content detection), but these are designed for marketing quality rather than security sanitization against prompt injection.
Audit Metadata