memrise

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

• Third-party content exposure detected (high risk: 0.80). The SKILL.md runbook explicitly instructs the agent to run memcli read commands (e.g., courses, course-levels, words, get-pool, search-pool) that fetch community Memrise course data (user-generated/open-web content) and to parse/validate that JSON output before taking write actions, so untrusted third‑party content can influence tool use and decisions.