convex

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

• Third-party content exposure detected (high risk: 0.80). The skill includes examples that ingest untrusted, user-supplied content — e.g., HTTP webhook handlers (request.json() in convex/http.ts), file uploads (/upload), and actions/queries that read user posts/documents and send them to LLMs or vector/full-text search (e.g., summarize action, vectorSearch and searchMessages) — so the agent would read/interpret third-party/user-generated content as part of its workflow.