loop

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

• Third-party content exposure detected (medium risk: 0.30). The $loop harness ingests outsider-authored free text via the human-supplied --objective/--objective-file into loop.py and passes the resulting objective string inside the JSON prompts to the explorer, coder, and code-reviewer workers (build_prompt(...)), so an outsider can poison that input without first selecting a specific item.