claude-skill

What This Skill Does

The claude-skill allows OpenClaw to directly leverage Claude Code, an advanced AI-powered CLI tool designed to assist with complex software engineering tasks. By integrating this skill, the AI agent can autonomously perform tasks such as refactoring code, implementing new features, debugging issues, and even reviewing pull requests directly within your terminal environment. It is designed for seamless, non-interactive automation, enabling hands-off task execution when you need to focus on higher-level system architecture while the agent handles implementation details.

Installation

To install this skill, run the following command via the ClawHub CLI: clawhub install openclaw/skills/skills/feiskyer/claude-skill

Ensure that you have claude installed globally (npm install -g @anthropic-ai/claude-code) and that tmux is available on your system, as it is required for the full interactive workflow. Verify installation by running claude --version.

Use Cases

• Automated Refactoring: Task the agent with updating legacy codebases to use modern syntax or library standards.
• PR Triage: Configure the agent to scan existing repositories, suggest fixes for failing tests, and draft code review summaries.
• Feature Implementation: Provide a high-level design specification and let the agent manage the worktree, create files, and write implementation code.
• Long-running Automation: Execute background tasks in non-interactive mode where the agent autonomously navigates directory structures and file edits.

Example Prompts

1. "Use Claude Code to refactor the authentication module in the current directory to use the latest SDK version."
2. "Review the open pull request in this project, generate a detailed summary of potential bugs, and suggest code improvements."
3. "Implement a new unit test suite for the utility functions found in /src/utils and automatically fix any uncovered edge cases."

Tips & Limitations

• Safety First: When running in --dangerously-skip-permissions mode, ensure you are operating within an isolated environment like a container or a dedicated VM.
• Output Handling: Always use --output-format stream-json to ensure progress can be tracked via OpenClaw logs. Avoid setting hard timeouts; utilize background execution and polling instead to allow the agent the time it needs for complex reasoning.
• Context awareness: Use the --append-system-prompt flag if you have specific company coding standards or architectural constraints that must be enforced during the agent's work session.