Veto - Protect your agent spending

Operators can author policies in YAML that specify limits on transactions, merchants, time windows, and semantic categories. The system evaluates each spend request through multiple stages including prompt injection detection, sanctions screening, and behavioral analysis before issuing an allow, deny, or escalate decision in under 300 milliseconds. Every verdict is signed with Ed25519 and made publicly accessible for offline verification, creating an immutable audit trail that cites the exact policy version used. This approach works across different agents and payment systems without requiring changes to the underlying rails. The tool integrates directly with environments like Claude Code through a simple CLI installation, enabling live policy enforcement and rollback capabilities while maintaining an open-source approach to its core components.