guardian-angel

What problem does it solve?

Guardian Angel enforces a virtue-based moral conscience as an enforcement layer for AI agents, preventing harmful tool use by gating actions at runtime according to Thomistic ethics.

Core Features & Use Cases

• Continuous disposition-based evaluation that runs alongside deliberative processes
• Intrinsic evil checks, virtue scoring (Clarity × Stakes), nonce-based escalation
• Plugin-based enforcement that blocks or escalates ambiguous actions before execution
• Supports escalation with one-time approvals to balance safety and usability
• Integrates with OpenClaw workflow for actions like gateway changes, data handling, and tool calls

Quick Start

Invoke Guardian Angel on an action to trigger the before_tool_call gate; for escalations, present a GUARDIAN_ANGEL_ESCALATE block with a nonce and await user approval via the ga_approve tool.