CodeClawed — Development skill for Claude Code

CodeClawed

A build prompt for a migration tool to move your [OpenClaw](https://github.com/openclaw) multi-agent deployment to [Claude Code](https://docs.anthropic.com/en/docs/claude-code).

Status: Work in Progress

This project is **not complete**. The core migration pipeline (analyze, generate, verify, chat) works end-to-end, but several (large!) components are stubs or partially implemented. See [Known Gaps](#known-gaps) below.

We are actively developing this tool against a live 16-agent OpenClaw deployment. The prompt and architecture are evolving as we resolve remaining issues.

What This Is

CodeClawed is two things:

1. **A CLI tool** that reads an OpenClaw installation (`~/.openclaw/`) and produces a fully functional Claude Code project — one directory per agent, with identity files, rules, memory, cron job wrappers, and a web chat frontend.

2. **A build prompt** (`codeclawed-migration-prompt.md`) that contains everything needed to reproduce the tool from scratch. Hand it to Claude Code and it will build the migration tool for your deployment.

How to Use the Prompt

The prompt is designed to be given to Claude Code (or any capable LLM) as a complete specification. It covers:

• OpenClaw's file formats and what each file does
• How each file maps to Claude Code's conventions
• The agent-runner abstraction (the single function that wraps claude -p)
• Cron-to-launchd schedule conversion
• The chat server with session resumption
• Memory system (files + optional pgvector)
• Build order and dependencies

For your own deployment

1. Read codeclawed-migration-prompt.md to understand the architecture
2. Point Claude Code at the prompt and your ~/.openclaw/ directory
3. Adjust for your deployment's specifics (number of agents, which cron categories you use, tool restrictions)
4. Run analyze → generate → verify → chat

To contribute

The prompt is the source of truth for the architecture. If you're fixing a bug or adding a feature,