alibabacloud-agentloop-evaluation

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow scripts/analyze_evaluation_results.py queries the SLS evaluation_detail Logstore (aliyun sls get-logs-v2 --query ...) and then parses/normalizes fields like explanation, evaluation_process, and (optionally) raw eval_info/custom_outputs, which originate as free text from evaluation runs; an outsider could influence the stored log content via evaluation inputs/outputs.