research-keywords

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly instructs the agent to fetch and crawl open web content — e.g., SKILL.md Phase 1: "Visit the user's website using WebFetch" and Phase 2's "Reddit & Community Mining" and the provided scripts (scripts/keyword-explorer.mjs and serp-analyzer.mjs) that pull Google autocomplete, PAA, and SERP results — which are untrusted, user-generated or public third-party sources whose contents the agent must read and use to drive keyword selection and subsequent actions.