research

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

• Third-party content exposure detected (high risk: 1.00). This skill explicitly instructs the agent to fetch and scrape public web sources (see references/web-research-collection.md and references/exa-advanced-search-categories.md) and to load social discourse (see references/social-signal-pass.md) — including Reddit, X, HN, YouTube, and GitHub — as required evidence that the agent must read and synthesize to drive recommendations and next actions, so untrusted user-generated content can materially influence behavior.