codex

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow reads outsider-authored free text because it instructs taking the user’s request verbatim (“contents = the user's request, verbatim”) and passing it to Codex via stdin (- < /tmp/codex-prompt.md), so any user-provided prompt can be ingested by the LLM.