A critical safety tool that pauses AI agent execution to request human authorization before performing high-risk or destructive operations.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install request-approval
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 request-approval 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 Request Approval?
The Request Approval skill provides a robust human-in-the-loop mechanism designed to prevent AI agents from performing unintended or dangerous actions. By integrating with the Preloop MCP server, this skill allows agents to pause their workflow and notify a human operator via Slack, email, or mobile apps whenever a sensitive operation is triggered. This ensures that autonomous agents remain under human supervision when it matters most, making it an essential component for production-ready Openclaw Skills deployments.
This skill is particularly valuable for maintaining system integrity and security. It forces the agent to document its reasoning and provide technical context before any changes are made, transforming blind automation into a transparent and accountable process. Whether you are managing cloud infrastructure or sensitive databases, this tool provides the oversight necessary to build trust in AI-driven workflows.
Request Approval Use Cases
- Authorizing the deletion of cloud resources, database records, or archived files to prevent data loss.
- Reviewing production deployments, environment variable changes, or CI/CD modifications.
- Verifying financial transactions such as purchases, charges, or refunds processed by an agent.
- Controlling external modifications like merging pull requests, posting to APIs, or sending bulk emails.
- Managing security-critical changes including permission updates, access control modifications, or API key rotations.
How Request Approval Works
- The AI agent identifies a task that meets the risk criteria, such as a system-level modification or a permanent deletion.
- Before execution, the agent invokes the request_approval tool with specific parameters: operation, context, and reasoning.
- Preloop triggers a notification to the designated human user through their preferred communication channel.
- The agent waits for a response, effectively pausing the specific workflow branch until authorization is granted.
- Upon receiving an 'Approved' response, the agent proceeds with the operation; if 'Denied', it immediately stops and informs the user.
Request Approval Setup
To enable this safety layer, you must have the Preloop MCP server configured in your environment. Detailed configuration can be found in the associated setup documentation for Openclaw Skills.
# Ensure your MCP server configuration includes the Preloop provider
# Update your agent's tool definitions to include request-approval
Once configured, the agent will automatically have the capability to call the request_approval function before risky steps.
Request Approval Data Schema & Taxonomy
The Request Approval skill organizes its authorization requests using a structured metadata schema to provide maximum clarity to the human reviewer.
| Parameter | Type | Required | Description |
|---|---|---|---|
| operation | string | Yes | A clear, specific description of the action (e.g., 'Delete logs/2024-01/'). |
| context | string | Yes | Technical details about what is affected (e.g., '3,421 files, 1.8GB'). |
| reasoning | string | Yes | The justification for why the operation is necessary (e.g., 'Disk usage at 92%'). |
| approval_policy | string | No | The specific policy name to govern the notification routing. |
Request Approval Advanced Features
- Multi-channel alerts: Receive approval requests across Slack, Email, and mobile platforms simultaneously for faster turnaround.
- Granular decision framework: Uses a predefined logic set to determine when an approval is strictly necessary versus optional.
- Detailed audit trails: Every request, including the reasoning and the human response, is logged for compliance and debugging within Openclaw Skills.
- Conditional logic: Seamlessly handles denials by allowing the agent to look for alternative, safer solutions when an initial request is rejected.
SKILL.md
Loading
Related Openclaw Skills
Nano Banana Pro Fallback
An advanced AI image generation and editing skill for OpenClaw that leverages the Gemini Image API with automated model failover for maximum reliability.
clawpet
An interactive virtual pet companion for AI agents that manages adoption, real-time status tracking, and AI image generation.
catime
A CLI-based utility that enables AI agents to retrieve and share hourly cat artwork generated by Google Gemini.
Claude Code Orchestrator (tmux-first)
A robust orchestration framework for running Claude Code in observable tmux sessions with automated progress tracking and completion callbacks.
WBS Planner
A hierarchical project management framework that decomposes complex goals into executable tasks for AI agents and human teams.
SuperDoc
A modern document editor skill for programmatically creating, editing, and formatting DOCX files with full control.
Featured*
