prompt-guard

You can defend group chats from prompt injection using multi-language detection and severity scoring. This approach evaluates messages across EN, KO, JA, and ZH to identify and block unsafe prompts.

Secret protection and redaction prevent API keys and credentials from leaking in chat prompts. The system scans for sensitive tokens and automatically redacts them while logging critical security events.

Prompt injection monitoring works across multiple languages including English, Korean, Japanese, and Chinese. Multi-language detection evaluates message intent and assigns severity scores regardless of the input language.

You can configure automated responses based on severity scoring by adjusting YAML-based configuration files. This setup allows you to define specific actions and notifications for each detected threat level.

You can restrict security commands and audit logs to group chat owners only. The system supports owner-only commands and sends critical event notifications specifically to authorized administrators.

Deterministic checks for prompt injection defense rely on predefined YAML configurations and script components. This approach may require manual updates to address novel or highly obfuscated multi-language attack vectors.