code-review

🦀 ClawedBack 🦀

Your Personal AI Assistant — Runs in Claude Code

A clean-room port of OpenClaw that runs inside Claude Code 17 skills · 5 agents · 3 install modes · ClawHub compatible

Multi-channel AI assistant with web chat, file sharing, voice messages, scheduled tasks, webhook ingestion, approval workflows, and a skill marketplace — all running inside Claude Code. Clone, setup, chat. 🦀

---

🦀 What Is This?

ClawedBack is a clean-room reimplementation of the OpenClaw personal AI assistant platform, rebuilt from the ground up to run inside Claude Code.

Important: ClawedBack is a port, not a fork. It resembles OpenClaw in spirit and functionality — a multi-channel, self-hosted AI assistant with tool execution, automation, and approval workflows — but is an entirely separate codebase with a different architecture. No OpenClaw code was used. 🦀

Where OpenClaw is a TypeScript monolith with a WebSocket gateway, ClawedBack flips the script: Claude Code IS the agent runtime. It's Claude Code skills.

🦀 Setup

Use of this project is subject to Anthropic's Terms of Service.

Prerequisites

• Python 3.11+
• tmux (sudo apt install tmux)

Step 1: Install Claude Code

curl -fsSL https://claude.ai/install.sh | bash

Step 2: Clone and enter the project

git clone https://github.com/reedmayhew18/ClawedBack.git
cd ClawedBack

Step 3: Start in tmux

tmux new -s clawedback

Step 4: Choose your install mode and launch Claude

Best for trying it out, development, or personal use. Safest option.

claude
# then: /oc-setup → choose Mode A

• Python venv inside the project folder
• Runs as your normal user
• Claude asks permission before risky operations (add --dangerously-skip-permissions to allow autonomous functionality)

Best for a dedicated server or VM where the assistant needs elevated access.

sudo -E "$(command -v claude || echo ~/.local/bin/claude)"
# then: /oc-setup → choose Mode B

• Python venv inside the project folder (same as A)
• Runs as root for full system access
• Commands for the regular user: sudo -u <username> <command>
• Still asks permission before destructive operations (add --dangerously-skip-permissions for autonomous mode — run the IS_SANDBOX=1 lines from Mode C first if using this flag)

Best for sandboxes, VMs, or containers designated for ClawedBack. Not for machines with important data. 🦀

# Set sandbox flag (both user and root)
echo 'export IS_SANDBOX=1' >> ~/.bashrc && source ~/.bashrc
sudo bash -c 'echo "export IS_SANDBOX=1" >> /root/.bashrc'

# Start Claude as root with no permission prompts
sudo -E "$(command -v claude || echo ~/.local/bin/claude)" --dangerously-skip-permissions
# then: /oc-setup → choose Mode C

• System Python directly (--break-system-packages), no venv
• Root with --dangerously-skip-permissions — Claude can do anything without asking
• Maximum capability, zero guardrails

WARNING: Only use Mode C on a system that is fully backed up, sandboxed, or designated for this purpose. Claude will have unrestricted root access with no confirmation prompts.

Step 5: Run the setup wizard

/oc-setup

The wizard handles everything: installs dependencies, starts the server, generates your auth token, offers voice (Whisper) setup, and begins polling. Open the URL it gives you, paste the token, and start chatting. 🦀

To detach tmux (keeps ClawedBack running): Ctrl+B, then D To reattach later: tmux attach -t clawedback

If the tmux session died (reboot, crash, etc.), start a new one and resume:

tmux new -s clawedback
cd /path/to/ClawedBack
claude    # (or sudo -E ... for root modes)

Then run:

/start

This ensures the server is up, creates the polling cron job, and runs one poll cycle. Run /start whenever you restart Claude Code. 🦀

If the config file is missing (deleted or corrupted), use /oc-resume instead — it auto-detects and rebuilds the config.

To pull the latest updates from GitHub:

/oc-update

This fetches upstream changes, shows what's new, and applies them — preserving your data/, auth token, config, and any custom skills you've installed. If server files changed, it automatically restarts the server. 🦀

---

🦀 Features

✅ Implemented

• 🌐 Web chat interface — dark theme, mobile-responsive, full markdown rendering
• 📎 File uploads — drag-and-drop, analyze any file type
• 📁 Organized file system — date-based storage with temporary share links
• 🎤 Voice messages — recorded in-browser, transcribed with Whisper (CPU or GPU)
• 🔐 Token auth — auto-generated or custom, single-user security
• 🗄️ Message queue — SQLite with WAL mode, CLI interface for skills
• 📡 SSE streaming — fetch-based with auth header, cross-device sync
• ✅ Read receipts — green checkmark when Claude reads your message
• 🔒 SSL/HTTPS — --ssl flag, custom cert paths, or free Let's Encrypt via /oc-ssl
• 🤖 17 custom skills — routing, sessions, approvals, automation, webhooks, file sharing, updates
• ⏰ Scheduled tasks — reminders, recurring jobs, cron expressions
• 🔗 Webhook ingestion — receive events from GitHub, CI, custom services
• ✋ Approval gates — single gates and multi-step workflow chains with crash recovery
• 🧩 Skill marketplace — browse ClawHub, import with security scanning
• 🔍 Security scanning — mandatory audit of imported skills (subagent-isolated, 5-category scan)
• 📺 Channel registry — extensible adapter system for future platforms
• 🔄 Self-updating — /oc-update pulls latest while preserving config/data
• 🩺 Self-healing resume — /oc-resume auto-detects setup if config is missing
• 🖥️ Three install modes — user, root, or full root with no guardrails

🔮 Planned

• 🦀 Browser automation (Playwright)
• 🦀 Device nodes (remote command execution)
• 🦀 TTS voice responses

---

🦀 Tool Compatibility (19/23 Matched)

OpenClaw has 23 built-in tools. Claude Code matches 19 of them out of the box — and the 4 missing are all API wrappers that can be built as skills. Full mapping in [.claude/skills/oc-hub/references/tool-mapping.md](.claude/s