A soft human-in-the-loop approval gate that pauses AI agents before they execute high-risk or destructive actions.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install human-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 human-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 Human Approval?
The human-approval skill is an essential safety layer for Openclaw Skills designed to provide human-in-the-loop (HITL) oversight. It acts as a cognitive speed bump, ensuring that AI agents do not perform irreversible operations—such as deleting code, sending external communications, or executing system-level commands—without explicit user consent.
By intercepting the agent's intent, this skill presents a structured confirmation prompt that details the action, target, and associated risks. This model-cooperative approach is ideal for interactive development sessions where developers need to maintain visibility and control over an agent's autonomous behavior without implementing complex hard-coded policies.
Human Approval Use Cases
- Preventing accidental file or directory deletion during automated refactoring sessions.
- Reviewing and approving external communications like emails or API posts before they are sent.
- Guarding against destructive shell commands or system configuration changes.
- Auditing financial transactions or invoice generation in automated workflows.
- Debugging agent behavior by forcing confirmation on every tool call using strict mode.
How Human Approval Works
- The skill monitors the agent's reasoning process and tool-call intent.
- It identifies high-risk triggers based on pre-defined categories like file operations, communications, or system changes.
- The agent pauses execution and generates a standardized Approval Required block for the user.
- The user reviews the action, target, and risk assessment provided in the prompt.
- Based on the user's response (yes/no/modify), the agent either proceeds, cancels, or adjusts the planned action.
Human Approval Setup
To integrate this safeguard into your agent, install it via the CLI:
openclaw skills install openauthority/human-approval
Once installed, the agent will automatically begin monitoring for high-risk actions based on the default rule set.
Human Approval Data Schema & Taxonomy
The human-approval skill organizes its logic through a configurable list of action categories. Users can interact with this taxonomy using the following schema:
| Command | Description |
|---|---|
list |
Displays all active approval categories (e.g., file.delete, sys.command). |
add <category> |
Appends a new operation type to the high-risk watch list. |
remove <category> |
Exempts a specific operation type from requiring approval. |
strict |
Sets the taxonomy to global, requiring approval for all tool interactions. |
off |
Suspends the approval logic for the current session state. |
Human Approval Advanced Features
- Custom Category Tagging: Add specific hooks like
git.pushto the approval list to customize the agent's safety boundaries. - Strict Mode Auditing: Enable a step-by-step audit trail where every single agent action requires a manual 'yes' to proceed.
- Modification Feedback Loop: Supports a 'modify' response, allowing users to provide corrective instructions instead of a simple binary approval.
- Session-Based Disabling: Quickly toggle the gate off for low-risk sessions without uninstalling the skill.
SKILL.md
Loading
Related Openclaw Skills
BOE Technology Group Assistant
A specialized knowledge skill providing deep technical and corporate insights into BOE Technology Group's semiconductor display leadership.
Flight Price Chart
A data visualization skill that embeds interactive 60-day flight price trend charts into AI responses for smarter travel planning.
Booking.com Skill
An AI-powered skill for automating hotel searches and managing global accommodation data through Booking.com.
whatdidyoudo
A specialized skill for reconstructing and presenting a plain-language summary of an AI agent's recent tool calls and autonomous actions.
Barbie Encyclopedia
A professional-grade knowledge skill for researching Barbie history, cinematic releases, and collectible value.
iCloud Calendar Management Skill
A sophisticated CalDAV-based skill for managing iCloud calendars through natural language, featuring smart duration inference and recurring event support.
Featured*
