Claude Code Agent Skill for OpenClaw 🤖

An OpenClaw skill that turns Claude Code into a background developer agent — delegated, managed, and reported on automatically.

Concept

You (Cody/OpenClaw) = PM / Architect
Claude Code (on server) = Developer

You assign a task → Claude Code codes, commits, pushes → you get an auto-review summary.

What It Does

• Delegates coding tasks to Claude Code running on the server
• Manages Claude Code sessions with persistence (resume across restarts)
• Auto-reviews work via a stop hook: git log, build/lint, code quality
• Two-way notifications: Telegram alerts when Claude Code finishes
• Tracks project history in memory/projects/<project>.md

Requirements

• OpenClaw agent running on a server
• Claude Code installed (npm install -g @anthropic-ai/claude-code)
• MCP server at /opt/cody-mcp/ (see references/project-setup.md)
• Telegram bot for notifications (optional but recommended)

Structure

claude-code-agent/
├── SKILL.md              # Main skill instructions for the OpenClaw agent
└── references/
    ├── memory-system.md  # How project memory and decision logging works
    └── project-setup.md  # MCP server setup, stop hooks, session persistence

Installation

# Via ClawhHub (when published)
clawhub install claude-code-agent

# Or manually — clone into your OpenClaw workspace skills folder
git clone https://github.com/starychenko/claude-code-agent \
  ~/.openclaw/workspace/skills/claude-code-agent

Core Workflow

1. Tell your OpenClaw agent about a coding task
2. Agent writes the task to MCP, launches Claude Code in the background
3. Claude Code works autonomously — codes, runs tests, commits
4. Stop hook fires → wakes the OpenClaw agent for auto-review
5. OpenClaw agent reviews git log, build output, code quality
6. You get a Telegram summary + git pull instruction

References

• OpenClaw Docs
• Claude Code Docs
• ClawhHub

---

Part of the Hal Stack ecosystem.