karpathy-education-first

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

• Third-party content exposure detected (high risk: 0.75). Outsider source: the workflow’s “B的第4步（karpathy-output-evolution）/education-first” is based on the SKILL.md “Source” URLs to public X/Twitter posts, and at runtime the agent would fetch/read that outsider-authored free text and include it in the LLM context for rewriting.