🐾 Introducing ShibaClaw — My OpenClaw-Based Personal AI Agent Framework

[RikyZ90]

🐾 Introducing ShibaClaw — My OpenClaw-Based Personal AI Agent Framework

The only AI agent framework combining multi-layer security with minimal token consumption — keeping your costs low without sacrificing power.

🤔 What is ShibaClaw?

ShibaClaw is a self-hosted, containerized AI agent runtime built on top of the NanoBot architecture.
It's designed to be lightweight, modular, and easy to deploy — whether on your local machine or a home server via Docker.

The goal? A personal agent that actually does things — not just chats.

✨ Key Features

• 🐳 Full Docker support — lightweight image, fast startup, minimal dependencies
• 🔌 Multi-LLM compatible — native SDKs (OpenAI, Anthropic), no LiteLLM overhead
• 🧠 Persistent memory — your agent remembers context across sessions
• 🛠️ Extensible skill system — drop in new capabilities easily
• 🛡️ Randomized Tool Output Wrapping — dynamic per-session nonces block Indirect Prompt Injection attacks before they can hijack the agent
• 🔍 Smart Install Guard — every pip install, npm install, apt install is intercepted and scanned for CVEs via pip-audit and npm audit before execution; only critical/high severity packages get blocked, clean ones go through freely
• 🔒 Self-hosted & private — your data never leaves your infrastructure, zero cloud lock-in
• 🔑 GitHub Copilot OAuth — rewritten with raw async device flow for stable background token refresh, no proxy dependencies

🚀 Why ShibaClaw?

I wanted something between "too simple" and "enterprise overkill".
ShibaClaw hits that sweet spot — practical, hackable, and secure for real personal automation workflows.

📢 Latest News — v0.0.7

• 🔍 Smart Install Guard — Package installations are no longer blindly blocked. They are intercepted and audited for CVEs using pip-audit and npm audit before execution. Only critical/high severity vulnerabilities get blocked; clean packages install freely. Destructive operations (uninstall, remove, purge) remain blocked.
• 🛡️ Security Hardening — Enhanced Indirect Prompt Injection protection via Randomized Tool Output Wrapping with dynamic nonces per-session.
• 🐾 LiteLLM Removed — Architecture modernized to native SDKs (openai, anthropic), dramatically reducing image sizes, startup times, and dependency risks.
• 🔐 GitHub Copilot OAuth rewritten — Raw async device flow for highly stable background token refresh, no proxy dependencies.

📂 Try it out

👉 GitHub: https://github.com/RikyZ90/ShibaClaw

Feedback, issues, PRs, and ⭐ stars are all super welcome!

---

🐕 Yes, it's named after my Shiba Inu. No, she does not contribute code. Yet.