benchmark-methodology

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.85). SKILL.md instructs the runtime workflow to collect evidence from outsider-authored sources (e.g., “Competitor’s own site,” “Review directories,” “LinkedIn,” “Portfolio / craft platforms”) and to include “source link/screenshot” and “one-line justification” in the LLM-visible competitor profile cards, which would require ingesting scraped/free-form text from those pages into the agent context.