karpathy-kb

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

• Third-party content exposure detected (medium risk: 0.30). In Procedure 2/3/5 and the lint routines, the LLM reads user-staged outsider free text that has been scraped or fetched into <topic>/raw/articles/ (via the firecrawl skill) and bookmark content in <topic>/raw/bookmarks/, then loads those raw sources fully into context to compile and answer.