Proprietary LLM APIs can have reasoning traces extracted, which may include sensitive information such as API keys and tokens. A method is proposed to sanitize repositories by searching for specific patterns related to sensitive data while avoiding modifications to non-sensitive files.