referrals
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill acknowledges the potential for indirect prompt injection when analyzing external customer feedback. Evidence chain:\n
- Ingestion points: External reviews, customer profiles, and web pages mentioned in
references/review-programs.md.\n - Boundary markers: Explicit instructions in
references/review-programs.mdstating that fetched reviews are untrusted data and should be used only for evidence extraction.\n - Capability inventory: No subprocess calls, file-system writes, or network operations are implemented across the skill's files.\n
- Sanitization: Guidance focuses on neutral evidence extraction and specifically warns against following instructions embedded within external text.\n- [EXTERNAL_DOWNLOADS]: The skill contains links to documentation and tool integration guides hosted on GitHub (
github.com/coreyhaines31/marketingskills). These are informational references to Markdown files and do not involve executable code or package installations.\n- [SAFE]: The core functionality of the skill involves designing business logic for referral and affiliate programs. It adheres to best practices and does not request or use sensitive system permissions.
Audit Metadata