blog-discourse

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

• Third-party content exposure detected (high risk: 0.85). Required workflow runs runtime WebSearch across public platforms (e.g., Reddit/X/Medium/GitHub) and captures the resulting free-text snippets into a JSON file that is then parsed by scripts/discourse_research.py, so outsider-authored prose is ingested into the agent’s LLM context via the untrusted snippet text pipeline.