RikyZ90/ShibaClaw — reverse-engineered prompt

Build me a self hosted AI agent called ShibaClaw that feels simple for a normal person to run, but is designed around privacy and safety from the start.

I want a local web interface where I can chat with the agent, connect my preferred AI provider like OpenAI, Anthropic, Gemini, DeepSeek, Ollama, and others, and use the same agent from places like Telegram, Discord, WhatsApp, terminal, or a Windows desktop launcher. It should have guided setup, token login, memory that learns useful things over time, scheduled tasks, automations, skills, and support for connecting MCP tools.

Please make security a core feature, not an afterthought. The agent should scan risky installs before running them, protect against prompt injection in tool results, block unsafe shell and network actions, and keep file access inside a workspace. It should be easy to run with Docker or pip, include a clean WebUI on localhost, and have good docs so I can get it working without babysitting it.