deep-research

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

• Third-party content exposure detected (high risk: 1.00). The skill's Step 3 explicitly instructs launched research agents to perform WebSearch and WebFetch on public web queries (including "community discussion", "practitioner blogs", and optional social media) and return sourced findings that are then ingested and used to modify the playbook, so it clearly consumes untrusted third‑party content (public websites, blogs, and social media) that can influence agent actions.