deep-search

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 ingest public web and community content (e.g., mcp__brave-search__brave_web_search, mcp__exa__web_search_exa, mcp__tavily__tavily_search / _crawl, "community discussions (including Reddit)"), and to read public GitHub/docs as evidence, so untrusted third‑party content is consumed and can influence decisions and tool use.