research-and-analyze

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

• Third-party content exposure detected (high risk: 0.85). Outsider free text can enter the LLM context during the gather-sources phase when scripts/fetch-sources.py fetches public web pages (URLs from research/{subject}/manifest.md) and writes their extracted article text to research/{subject}/sources/*.md, which is then read by the analyze-claims workflow into the agent prompt.