deep-learning-agent

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

• Third-party content exposure detected (high risk: 0.90). The SKILL.md and references/research-protocol.md explicitly require using web_search/web_fetch and ingesting content from public sites including Reddit and X/Twitter (public social media and web pages) during the core RESEARCH waves, meaning the agent will read untrusted, user-generated third‑party content that can influence its decisions and tool use.