run-deep-research

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 via the codex/Claude research tasks that run run-research and perform runtime web/Reddit scraping (e.g., raw-scrape-links / smart-scrape-links returning arbitrary page/Reddit comment markdown), which is then included in the task’s prompt context for synthesis and written into the corpus.