A powerful tool interception framework to inject custom logic before and after AI tool execution.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install hook-system
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 hook-system 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 Hook System?
The Hook System is a specialized framework designed to provide granular control over the lifecycle of AI tool calls. By integrating this skill, developers can inject custom scripts into the execution flow, allowing for real-time monitoring, input validation, and output modification. It serves as a middleware layer that enhances the reliability and security of Openclaw Skills by ensuring every tool interaction follows predefined logic.
This system is essential for developers who need to build guardrails around their AI agents. It allows for the interruption of potentially harmful commands before they run (PreToolUse) and the sanitization of results before they are returned to the model (PostToolUse), making it a cornerstone for production-grade agentic workflows.
Hook System Use Cases
- Preventing path traversal attacks by validating file paths in tool arguments before execution.
- Automatically redacting sensitive data such as API keys or PII from tool outputs to maintain privacy.
- Generating detailed audit logs of every tool interaction for compliance and debugging purposes.
- Implementing conditional tool execution where an action is only allowed if specific environment criteria are met.
How Hook System Works
- When a user message triggers a tool call, the system identifies the associated Openclaw Skills hooks.
- The HookRunner executes PreToolUse scripts, passing tool metadata via environment variables.
- Based on the script exit code (0 for allow, 2 for deny), the system either proceeds to the Tool Executor or blocks the call.
- If allowed, the tool runs and its raw output is captured.
- The PostToolUse scripts are then executed to process, log, or modify the tool result before the final response is generated.
Hook System Setup
To get started with the Hook System, ensure you have Node.js 18 or higher installed. You can initialize the runner within your local Openclaw Skills environment using the following structure:
# Navigate to your skill directory
cd hook-system
# Configure your hooks in a JSON format
# { "hooks": { "preToolUse": ["node ./hooks/validate.mjs"] } }
To programmatically invoke the runner in your JavaScript environment:
const { HookRunner } = require('./scripts/hook-runner.mjs');
const runner = new HookRunner({
preToolUse: ['echo "Validating $HOOK_TOOL_NAME"'],
postToolUse: ['node ./hooks/filter-secrets.mjs']
});
Hook System Data Schema & Taxonomy
The Hook System leverages standard environment variables to communicate context to your scripts. This allows for language-agnostic hook implementation using Openclaw Skills.
| Environment Variable | Description |
|---|---|
HOOK_EVENT |
Specifies the hook type: PreToolUse or PostToolUse. |
HOOK_TOOL_NAME |
The name of the tool being intercepted. |
HOOK_TOOL_INPUT |
The raw JSON string of the tool arguments. |
HOOK_TOOL_OUTPUT |
The result of the tool execution (available in PostToolUse). |
HOOK_TOOL_IS_ERROR |
A flag (0 or 1) indicating if the tool execution failed. |
Hook System Advanced Features
- Granular Exit Code Logic: Distinguish between allowing execution, denying it, or continuing with a warning based on script return values.
- Environment Injection: Automatically provides parsed and beautified JSON inputs to simplify script logic for developers.
- Secret Scrubbing: Built-in patterns for filtering sensitive information from tool responses using PostToolUse hooks.
- Extensible Middleware: Support for multiple sequential hooks, enabling complex multi-stage validation pipelines for any agentic tool.
SKILL.md
Loading
Related Openclaw Skills
DeepRead Insurance Document Processing
DeepRead Insurance is a high-accuracy AI skill designed to extract structured data from complex insurance documents and redact sensitive PII.
Parasite SEO
A strategic framework for publishing content on high-authority platforms to leverage their domain strength for immediate search rankings and AI citations.
Session Compactor
An automated context management tool that prevents token overflows by summarizing long conversation histories into concise, structured snapshots.
Programmatic SEO
A professional framework for automating the generation of high-quality, data-driven landing pages at scale using AI-enhanced templates.
Tool Registry
A sophisticated tool discovery and registration system featuring token-based routing, granular permissions, and sub-agent whitelisting.
Usage Tracker
A comprehensive token monitoring and budget management system designed to track AI agent consumption and operational costs.
Featured*
