Balanov/shim ? reverse-engineered prompt

Reverse engineered prompt

Build me a tiny terminal based AI assistant that works locally first and feels simple to use.

I want one small binary with no extra app window, no database, and no weird setup. It should let me talk in plain language or use normal commands, connect to different AI providers, and work with local models or hosted ones. It also needs to support MCP servers and reusable skills, so I can add tools and instructions from common folders and load them when needed.

Please make it safe by default, so anything that changes files, runs shell commands, or calls tools asks for confirmation unless I explicitly allow it. Keep config in plain JSON, keep sessions in JSONL, and never store API keys in the config, only read them from environment variables. It should work on Windows, macOS, and Linux, and have a simple setup command plus helpful commands like model management, skills, MCP, sessions, and status. If you need to check current docs for provider or MCP details, look them up online.

Are you gonna build this?

make sure you review the code using coderabbit

Try freeSponsored — opens CodeRabbit in a new tab