claude-code-cli

What This Skill Does

The claude-code-cli skill provides a powerful interface between your OpenClaw agent and the Claude Code CLI, enabling autonomous coding, refactoring, and project exploration. By integrating directly into your development workflow, the skill allows you to delegate complex coding tasks—such as implementing new features, auditing pull requests, or performing large-scale refactors—directly to the terminal. It handles both interactive, human-in-the-loop sessions through PTY (pseudo-terminal) support and automated, non-interactive workflows via headless pipe mode, ensuring flexibility for every stage of your software development lifecycle.

Installation

To integrate this skill into your OpenClaw environment, execute the following command in your terminal: clawhub install openclaw/skills/skills/atelypham/claude-code-cli Once installed, ensure your environment has the Claude Code CLI binary available and configured with the necessary API credentials to authorize the agent's actions.

Use Cases

• Feature Implementation: Offload the boilerplate and structural scaffolding of new API endpoints or UI components to Claude Code.
• PR Reviewing: Automate the initial pass of code reviews, allowing Claude to flag potential bugs, security vulnerabilities, or styling inconsistencies before human intervention.
• Refactoring: Safely apply large-scale codebase changes, such as renaming modules, updating dependency versions, or standardizing error handling patterns across multiple files.
• Headless CI Automation: Utilize headless mode to perform automated linting, documentation generation, or code analysis tasks where input is pre-defined and confirmation prompts are suppressed via flags.

Example Prompts

1. "Use Claude Code to refactor the authentication middleware in the src/auth folder to support OIDC providers, running it as a background process."
2. "Please initiate a Claude Code session to review the open pull request #42 in this repository and provide a summary of suggested changes."
3. "Run Claude Code in headless mode to generate a JSON summary of all class definitions in the backend directory for our documentation portal."

Tips & Limitations

• PTY is Mandatory: Always enable pty:true when interacting with Claude Code in its default mode. Without it, the agent will fail to render the terminal interface correctly and may hang indefinitely.
• Avoid Sensitive Paths: Never execute this skill on configuration files or restricted directories (like ~/.openclaw/). Restrict its scope to your active project workspace.
• Use Headless for Scripts: When automating tasks, leverage the -p flag and structured output options (like stream-json) to ensure the agent receives clean, machine-readable data rather than raw terminal output.