prompt-guard

Detect prompt injection attacks by analyzing text for pattern anomalies, normalizing inputs, and evaluating severity to neutralize malicious instructions before processing.

Prompt injection is an adversarial input technique that hijacks an AI agent's intended actions by manipulating its context. It compromises AI safety by overriding core operational directives.

Prevent hidden malicious instructions by running a guardrail check that scans skill content and external data, flagging unauthorized commands to stop context manipulation and instruction hijacking.

Yes, you can use a prompt guardrail to scan new skill files and external data. It ensures AI agent robustness by identifying and neutralizing harmful adversarial inputs prior to execution.

Yes, prompt injection detection works across various text sources including skill files, user inputs, and external data. It employs pattern detection to identify and flag hidden instructions.