guardian-angel

To enforce an ethical gate before tool calls, you need a runtime layer that evaluates actions using disposition-based virtue checks and blocks harmful commands. Guardian Angel applies intrinsic-evil checks and virtue scoring before execution.

Nonce-based escalation for ambiguous AI actions is a safety mechanism that pauses execution and requests one-time user approval. Guardian Angel uses this to balance safety and usability when virtue scores indicate unclear intent.

To implement virtue ethics in an OpenClaw workflow, integrate a moral conscience plugin that gates runtime actions. Guardian Angel applies Thomistic ethics to evaluate gateway configurations, data handling, and cross-service communications.

You can block harmful tool execution without hardcoding rules by using a disposition-based evaluation layer that continuously scores actions. Guardian Angel applies Clarity × Stakes scoring and intrinsic-evil checks to automatically gate tool calls.

Guardian Angel works with cross-service communications by running continuous disposition-based evaluations alongside deliberative processes. It enforces its ethical gate across all OpenClaw-enabled workflows including data handling and gateway changes.

To approve an action during an escalation, present a GUARDIAN_ANGEL_ESCALATE block containing a nonce and await user approval via the ga_approve tool. This one-time approval allows the gated action to execute.