marketing-customer-research

Pass

Audited by Gen Agent Trust Hub on Aug 25, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from external sources such as customer reviews, transcripts, and social media threads. This introduces a vulnerability surface where instructions embedded in processed data could potentially influence agent behavior.
  • Ingestion points: Processes raw research materials including transcripts, surveys, support tickets, and web content from platforms like Reddit and G2 as described in SKILL.md and references/source-guides.md.
  • Boundary markers: The skill lacks explicit boundary markers or delimiters to isolate user-provided research data from the agent's core instructions.
  • Capability inventory: The skill instructions involve reading local configuration files (.agents/product-marketing.md) and gathering data from external websites.
  • Sanitization: No explicit sanitization or validation of the external content is mentioned in the workflow instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 25, 2026, 12:30 PM
Security Audit — agent-trust-hub — marketing-customer-research