hanria

Checks an agent's proposed action against rules you wrote, then keeps a tamper-evident log of the decision.

Ecosystem: Agent Skills
Skill category: Security
Creator / organization: Original creator not yet identified

What it is

hanria is published by HANRIA-AI. You first write a mandate that defines what an AI agent may and may not do. Before the agent takes a consequential action, the Skill checks that action against the mandate and returns one of three outcomes: permit, deny, or escalate to a human. Each decision is appended to a local log linked by a hash chain, so later tampering can be detected. It runs entirely on your machine, with no network access and no credentials. The project is candid about its limits: the Skill is advisory and cannot enforce anything, so an agent that ignores a denial can still act; credential detection is a heuristic that guards against accidents only; and the unsigned log supports local integrity checks but is not evidence.

Who it's for:developers and teams who let AI agents take actions and want an audit trail.

Before you start: install with npx skills add. It needs only the Python 3 standard library. Licensed under MIT.

Based on the project's public documentation. Not tested by AI Inspiration World.