deep-research

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime workflow explicitly performs WebSearch + WebFetch and then reads fetched pages’ body text into the agent for summarization/citation, which can include public web content authored by outsiders (indirect prompt injection risk via fetched free-form text).

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly requires runtime fetches from Context7 MCP (via resolve-library-id / get-library-docs) and arbitrary web URLs retrieved by WebSearch then loaded with WebFetch (http(s)://... pages), and those fetched documents are used directly to drive the agent's prompts and final report, so these external URLs (Context7 endpoints and the http(s) URLs returned by WebSearch/WebFetch) are runtime dependencies that control the agent.