Skip to main content

​
Security

Agntor ships three safety controls to reduce prompt injection, data leaks, and unsafe tool use.

​
Guard (Input Validation)

  • Scans incoming prompts for injection or malicious patterns.
  • Returns pass or block with violation types and CWE mappings.

​
Redact (Output Filtering)

  • Scans agent outputs for PII, secrets, and unsafe content.
  • Redacts matches before returning to the user.

​
Tool Guard (Function Control)

  • Enforces allow/deny lists for tool execution.
  • Useful for shell commands, database access, or external APIs.