explainer-stage-ad

Pass

Audited by Gen Agent Trust Hub on Sep 25, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it fetches and processes external content (e.g., product pages, company websites) and maintains state in local files.
  • Ingestion points: Fetches external URLs for company research, positioning, and brand voice analysis. Reads from .agents/product-context.md.
  • Boundary markers: The instructions explicitly command the agent to treat untrusted content as data, never as instructions, and refer to references/agent-security.md for specific handling procedures.
  • Capability inventory: The skill generates prompts for external image and video models and writes findings/context to .agents/product-context.md.
  • Sanitization: The references/agent-security.md file mandates HTML escaping for all interpolated values, credential redaction, and strict data minimization practices.
  • [SAFE]: The skill incorporates highly defensive security protocols via references/agent-security.md and references/house-rules.md. These include mandatory rules against acting on instructions embedded in data, strict credential protection (e.g., flagging secrets without reproducing them), and ethical data usage constraints.
  • [SAFE]: Heuristic detections regarding concealment were investigated and determined to be false positives. Instructions to omit preambles or warnings from drafts are intended to ensure the final generated ad copy is clean and production-ready for the user, rather than to hide malicious activity.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 25, 2026, 04:06 PM