An automated coding agent for Claude Code that handles the end-to-end lifecycle of features, bugfixes, and refactors within isolated worktrees.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install claude-skill
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-skill 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 Agent Skill?
This skill enables developers to leverage Claude Code as a fully managed coding agent, providing a structured framework to operate across the entire development lifecycle. By integrating with Openclaw Skills, it transitions Claude from a simple chat interface into a production-grade automation tool capable of managing worktree setups, branch isolation, and pull request merges.
Built for reliability, the skill supports both quick one-shot tasks and complex, long-running features. It utilizes tmux for session persistence and steering, ensuring that developers can monitor progress or intervene without interrupting the agent's workflow. This approach makes it a core component of the Openclaw Skills ecosystem for teams seeking autonomous coding capabilities.
Claude Code Agent Skill Use Cases
- Automated feature implementation within isolated Git branches to maintain clean main environments.
- One-shot bug fixes and documentation updates using the non-interactive automation mode.
- Complex code refactors that require multi-file awareness and persistent execution.
- Autonomous security auditing and code analysis of existing repositories.
- Streamlining the PR lifecycle by automating branch creation, testing, and multi-model reviews.
How Claude Code Agent Skill Works
- The skill creates an isolated Git worktree and dedicated branch for the specific task to prevent environment contamination.
- It launches a Claude Code instance inside a managed tmux session, enabling interactive steering and background execution.
- Task metadata is registered in a centralized JSON registry with atomic file locking to track session IDs, PIDs, and progress.
- An adaptive monitoring loop polls the process status and analyzes log files for output changes or potential loops.
- Upon completion, the skill validates the work against a Definition of Done checklist, including CI passing and PR creation.
- A secondary AI model is often invoked to provide a cross-model code review before the task is marked as finished.
Claude Code Agent Skill Setup
To use this skill within the Openclaw Skills framework, ensure your environment meets the following requirements:
# Install the Claude Code CLI
npm install -g @anthropic-ai/claude-code
# Ensure tmux is installed for session management
tmux -V
# Verify GitHub CLI for PR automation
gh --version
Configure your Anthropic API key and ensure your local Git environment has the necessary permissions to create worktrees and push branches.
Claude Code Agent Skill Data Schema & Taxonomy
The skill maintains a persistent registry in .clawd/active-tasks.json to manage agent state. Data is organized as follows:
| Field | Type | Description |
|---|---|---|
| id | String | Unique task identifier used for tmux and worktree names. |
| status | String | Current execution state: running, done, or blocked. |
| worktree | String | Absolute path to the isolated task directory. |
| panePid | Integer | The process ID of the shell inside tmux for reliable monitoring. |
| checks | Object | Boolean flags for PR creation, CI status, and code reviews. |
| silentRounds | Integer | Counter for adaptive timeout logic to detect stalled processes. |
Claude Code Agent Skill Advanced Features
- Non-interactive mode using the dangerously-skip-permissions flag for headless automation.
- Real-time steering via tmux send-keys, allowing human intervention to redirect the agent mid-task.
- Smart retry strategy that automatically attaches failure logs and CI errors to subsequent attempts.
- Multi-model code review integration to catch logic errors using a different LLM than the authoring agent.
- ANSI-stripped logging for clean, grep-parseable history of all agent interactions.
- Adaptive timeout logic (Poll-and-Extend) that supports tasks ranging from minutes to 12+ hours.
SKILL.md
Loading
Related Openclaw Skills
Xiaohongshu Login
A specialized workflow for managing Xiaohongshu authentication and QR code login via Openclaw Skills.
Mail Parser
Automatically transform plain text emails into clean, actionable JSON objects for seamless automation.
Genviral Partner API Skill
A powerful CLI wrapper and automation suite for the Genviral Partner API that handles the entire content lifecycle from AI generation to performance tracking.
OpenClaw Genie
A comprehensive technical guide and operational assistant for managing the OpenClaw self-hosted AI agent gateway.
Deep Research Multi-Agent Workflow
An advanced multi-agent orchestration workflow that automates deep, systematic research by splitting complex objectives into parallel sub-tasks.
Nanobanana Image Generation
A versatile image generation and editing tool powered by Google Gemini API for seamless AI-driven visual creation.
Featured*
