grpo-rl-training

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). High-risk: the codebase includes unsafe execution of model-generated code (exec of completions) enabling remote code execution/backdoor abuse, plus default telemetry and remote model/dataset pulls that can leak sensitive data or be abused via supply‑chain attacks.

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

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime LLM context is fed only from the operating user’s chosen dataset and prompts; in the provided template it loads GSM8K via load_dataset('openai/gsm8k', ...) and maps it into prompt/answer, so the model ingests outsider-authored dataset question/answer text (public web/community dataset) through the training prompts.