A policy enforcement layer for Openclaw Skills that routes agent decisions through human-in-the-loop approvals and runtime guards.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install loop-engine-governance
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install loop-engine-governance using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Loop Engine Governance?
Loop Engine Governance provides a robust framework for managing decision-making within agentic workflows. By integrating with Openclaw Skills, it ensures that every action proposed by an AI agent is evaluated against predefined policies, including human approval gates and structured evidence requirements. This skill allows developers to move beyond prompt-based instructions into runtime enforcement, providing a reliable audit trail and immutable transition history for every automated step.
Whether operating in local-only mode for maximum privacy or using LLM-augmented adapters for complex decision logic, this skill ensures that agents operate within safe boundaries. It effectively bridges the gap between autonomous AI capabilities and organizational compliance requirements by treating governance as a first-class citizen in the development of Openclaw Skills.
Loop Engine Governance Use Cases
- Requiring human approval for high-stakes agent actions like infrastructure changes or financial transactions.
- Blocking AI-generated recommendations that fall below a specific confidence score threshold.
- Ensuring structured evidence is captured and sanitized before transitioning workflow states.
- Creating immutable audit logs for all agent and human interactions within a distributed system.
- Sanitizing sensitive data before it is sent to external LLM providers during the reasoning process.
How Loop Engine Governance Works
- The OpenClaw agent proposes a specific action or state transition based on its current task.
- Loop Engine intercepts the proposal and evaluates it against configured guards such as confidence thresholds and actor authorization.
- If the policy requires human intervention, the workflow pauses until an authorized human actor triggers the approved signal.
- The system validates the presence and structure of required evidence fields before allowing the transition to proceed.
- Upon successful validation, the transition is recorded in the immutable audit trail, and the OpenClaw agent executes the approved action.
Loop Engine Governance Setup
Install the core Loop Engine SDK and the necessary OpenClaw adapter using npm:
npm install @loop-engine/sdk @loop-engine/adapter-memory @loop-engine/adapter-openclaw
If you intend to use provider-backed governance (e.g., Anthropic or OpenAI), install the corresponding adapters:
npm install @loop-engine/adapter-anthropic @anthropic-ai/sdk
npm install @loop-engine/adapter-openai openai
Configure your environment variables for any external providers used (e.g., ANTHROPIC_API_KEY or OPENAI_API_KEY) and initialize your loop definition via YAML.
Loop Engine Governance Data Schema & Taxonomy
The skill manages governance data using a structured schema that ensures auditability and security:
| Component | Description |
|---|---|
| Loop Definition | YAML configuration defining states, transitions, allowed actors, and guard logic. |
| Evidence Object | A structured JSON object containing context for a decision, sanitized by guardEvidence. |
| Audit Trail | An immutable log entry containing actor IDs, timestamps, signal IDs, and state results. |
| Guard Registry | A collection of logic gates (Confidence, Human-Only, Evidence-Required) evaluated during transitions. |
Loop Engine Governance Advanced Features
- Multi-provider support allowing governance decisions to be backed by various LLMs including Claude, GPT-4o, and Grok.
- Local governance mode which enables policy enforcement and audit trails without making any external network calls.
- Advanced evidence sanitization that automatically strips PII, prevents prompt injection, and caps value lengths at the skill boundary.
- Sequential guard evaluation logic ensuring authorization, evidence completeness, and confidence thresholds are met in the correct order.
- WebSocket gateway integration for seamless event forwarding between OpenClaw and the Loop Engine runtime.
SKILL.md
Loading
Related Openclaw Skills
FreeRide
FreeRide automates the selection and configuration of high-quality free AI models from OpenRouter for the OpenClaw ecosystem.
Academic Engineering Literature Survey
A high-precision AI research agent that automates systematic engineering literature surveys and daily technical reading using formal academic methodologies.
RunningHub
A professional cloud platform for running ComfyUI workflows to generate AI images and videos via a robust API.
Data Analysis Workflow
A standardized six-stage framework for transforming raw data into publication-ready APA reports and scientific visualizations.
DaE Persona Context Injector
A specialized upstream profiling skill that builds high-fidelity, reusable persona profiles to eliminate generic AI responses and improve downstream agent collaboration.
Skill Discovery & Installation
A specialized skill designed to browse, discover, and instantly install new capabilities from the open agent skills ecosystem.
Featured*
