A pre-execution safety gate that uses persona-weighted consensus to govern high-risk AI agent actions before they cause irreversible side effects.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install consensus-agent-action-guard
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 consensus-agent-action-guard 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 Consensus Agent Action Guard?
The Consensus Agent Action Guard is a specialized tool within the Openclaw Skills ecosystem designed to act as the final safety gate before an autonomous agent executes an action. By inserting an explicit governance layer, it prevents catastrophic automation failures that typically occur at the point of execution. This skill evaluates proposed actions based on risk, irreversibility, and potential side effects, ensuring that every move an agent makes is validated against a robust policy framework.
Utilizing the same underlying consensus stack as communication and merge guards, this skill provides a unified policy language across all agent operations. It functions by applying weighted consensus logic from various personas to reach a definitive decision, ensuring that high-stakes operations are never performed without proper authorization or alignment with organizational protocols.
Consensus Agent Action Guard Use Cases
- Gating destructive system operations such as mass file deletions or database schema changes.
- Controlling external communication channels, including automated messaging and social media posting.
- Enforcing human-in-the-loop requirements for irreversible financial transactions or high-risk administrative tasks.
- Validating complex multi-agent workflows where cross-persona agreement is mandatory before proceeding.
How Consensus Agent Action Guard Works
- The agent proposes an action and sends the input data to the guard interface.
- The skill analyzes the action metadata for risk levels and potential impact.
- Weighted consensus logic is applied based on the current persona set or external votes.
- The system calculates the outcome, resulting in a status of ALLOW, BLOCK, or REQUIRE_REWRITE.
- Decision artifacts and persona updates are written to the board for auditability and transparency.
Consensus Agent Action Guard Setup
To integrate this component of Openclaw Skills into your environment, follow these steps:
# Install the package
npm install consensus-agent-action-guard
# Configure the required environment variables
export CONSENSUS_STATE_FILE=./state/governance.json
export CONSENSUS_STATE_ROOT=./state/root
# Run the guard with a sample input
node --import tsx run.js --input ./examples/input.json
Consensus Agent Action Guard Data Schema & Taxonomy
The skill manages its governance data through structured board artifacts and state files. Below is the primary metadata taxonomy:
| Attribute | Type | Description |
|---|---|---|
CONSENSUS_STATE_FILE |
Path | Location of the JSON file tracking decision history and active states. |
persona_set_id |
String | Identifier for the group of personas involved in the consensus process. |
external_votes |
Array | A collection of votes provided by external agents or humans in external_agent mode. |
decision |
String | The final output: ALLOW, BLOCK, or REQUIRE_REWRITE. |
Consensus Agent Action Guard Advanced Features
- Support for external_agent mode, allowing the integration of manual human votes or third-party model evaluations for deterministic aggregation.
- Persona-based execution which leverages existing persona_set_id definitions to maintain strict governance consistency.
- Seamless integration with the broader Openclaw Skills framework via shared consensus-guard-core wrappers.
- Idempotent decision lookup capabilities to prevent duplicate evaluations of the same action trigger.
SKILL.md
Loading
Related Openclaw Skills
Sutrena
Deploy websites, forms, and dashboards instantly through a simple API with no build steps or hosting configuration required.
OpenClaw Traditional Chinese Dictionary Skill
A professional-grade skill for querying official Traditional Chinese dictionaries from Taiwan's Ministry of Education with automated version management.
Pi Workflow Orchestration
A comprehensive orchestration framework for AI agents to manage complex tasks, ensure high code quality, and implement a continuous self-improvement loop.
Claude Batch API for Anthropic
A high-throughput skill to process massive Claude API request volumes asynchronously while slashing token costs by 50%.
Consensus Code Merge Guard
A persona-weighted governance tool for AI-assisted engineering that evaluates PR risks and automates merge decisions.
consensus-deployment-guard
A deterministic safety gate for deployment execution that enforces governance through schema validation and persona-weighted voting.
METADATA
Featured*
