crawford-method

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

• Third-party content exposure detected (high risk: 0.75). SKILL.md describes sourcing heuristics from “public data via ColdIQ” endpoints (e.g., /v1/limadata/research/ai-search for filings/regulation/events and /v1/limadata/enrich/company for growth/firmographics), which at runtime implies ingestion of outsider-authored public text into the agent’s LLM context when the endpoints return readable prose.