discourse

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

• Third-party content exposure detected (high risk: 0.90). The SKILL.md explicitly instructs the agent to fetch and parse community content from public, user-generated sources (Hacker News via hn.algolia.com, Reddit via the JSON API, Lobsters via web search, and tech blogs) using WebFetch/WebSearch, so the agent will ingest untrusted third‑party content that can influence its findings and actions.