geo-gap-fixer

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

• Third-party content exposure detected (high risk: 0.85). Outsider free text is ingested into the agent’s LLM context via the runtime LLM responses captured by scripts/probe_llms.py (the response field from OpenAI/Anthropic/Google/Perplexity), which may contain arbitrary third-party content and is then processed into data/raw_responses.json → data/analysis.json → report/geo_audit_report.md.