A professional-grade Agent execution framework extracted from core Claude Code patterns to ensure secure, efficient, and traceable AI operations.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install claude-code-framework
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 claude-code-framework 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 Claude Code Framework?
The Claude Code Framework is a robust execution layer designed for AI agents, modeled after the best practices found in high-performance coding assistants. It provides a structured environment where every agent action is governed by safety protocols, resource management, and customizable intervention points. By implementing this framework within Openclaw Skills, developers can ensure their agents operate with precision, maintaining a strict balance between autonomous action and human-supervised security.
At its core, the framework addresses the common pitfalls of LLM-based agents: runaway context usage and high-risk command execution. It introduces a sophisticated risk classification system that categorizes tools into automatic, approval-required, or blocked tiers. This architectural pattern ensures that Openclaw Skills remain production-ready, auditable, and context-aware throughout complex development tasks.
Claude Code Framework Use Cases
- Enforcing security boundaries by requiring human approval for destructive CLI commands.
- Preventing agent failures caused by context window overflow through proactive budget monitoring.
- Debugging complex agent workflows using lifecycle hooks for logging and state tracking.
- Implementing read-only modes for AI agents to explore codebases without risk of modification.
- Standardizing agent behavior across multiple projects using a unified configuration schema.
How Claude Code Framework Works
- Pre-Task Initialization: The framework triggers the pre_task hook to validate permissions and verify current context availability before any logic begins.
- Risk Assessment: Every tool call is intercepted by the on_tool_call handler, which evaluates the command against a tiered risk matrix (AUTO, APPROVE, BLOCK).
- Context Budgeting: The system continuously monitors token usage, transitioning through NORMAL, WARNING, and CRITICAL states to trigger automated compaction or blocking.
- Lifecycle Interception: Custom logic is injected at key stages (pre/post tool calls) to manage metadata, update memory, or handle errors gracefully.
- Task Finalization: Upon completion, the post_task hook summarizes results and updates the agent's long-term workspace memory.
Claude Code Framework Setup
To integrate this framework into your Openclaw Skills environment, configure your project directory and initialize the handler:
# Navigate to your skills directory
cd skills/claude-code-framework
# Install dependencies (if applicable)
npm install
# Enable the framework in your local config.json
# Set "enabled": true under the claudeCodeFramework key
Claude Code Framework Data Schema & Taxonomy
The framework organizes its operational data using a structured taxonomy to maintain transparency and control:
| Component | Data Managed | Description |
|---|---|---|
| Risk Rules | BUILTIN_RULES |
Defines which CLI tools require manual approval versus automatic execution. |
| Context Budget | ContextBudget |
Tracks percentage, used tokens, and threshold-based status (80%/90%/98%). |
| Execution Modes | ExecutionMode |
Switches between DEFAULT, READ_ONLY, AUTO, and BYPASS modes. |
| Hook Registry | HookManager |
Manages registration of custom logic for the 5 primary lifecycle events. |
Claude Code Framework Advanced Features
- Multi-stage context compaction (Micro-compact and Session-compact) based on defined token thresholds.
- Granular risk classifier for BASH commands, identifying dangerous patterns like disk formatting or registry deletion.
- Custom hook system allowing developers to bridge Openclaw Skills with external monitoring or approval systems.
- Support for specific execution modes to restrict AI agents to non-destructive analysis tasks.
SKILL.md
Loading
Related Openclaw Skills
Feishu Message Download
A professional utility to programmatically download attachments like videos, images, and documents from Feishu chat messages to local storage.
Agent Link
Agent Link is a secure communication bridge that enables distributed Openclaw Skills to exchange messages across different devices via a relay server.
LLMCities Web Host
A specialized web hosting skill that enables AI agents to publish content, manage static websites, and track visitor analytics on LLMCities.com.
Gold Price Fetcher
A high-performance utility to retrieve and cache real-time gold price data directly from JD Finance for AI agents.
AI-Integrated STEAM Lesson Generator
A specialized tool for Hong Kong educators to generate complete, AI-integrated lesson plans that meet EDB funding and curriculum requirements.
Agent Link
Agent Link is a secure communication framework that enables OpenClaw instances and agents to exchange messages across different computers via a dedicated relay server.
Featured*
