0xku/kon — reverse-engineered prompt
Reverse engineered prompt
Build me a small, opinionated terminal coding assistant called Kon in Python. I want it to feel minimal, fast, and easy to understand, with a simple interactive text UI plus a single prompt mode for scripts. It should let me pick a model and provider, pass an API key or base URL, continue my last session or resume a specific one, and optionally turn on web search and web fetch tools. Inside the app I should be able to chat with the agent, run shell commands, and also run shell commands with the model explaining the results.
Please make project context a first class thing, especially loading instructions from an AGENTS.md file, and keep the built in prompt and tool layer very small. I also want sensible config stored in the user config folder, session saving, permissions for tool use, and a few quality of life touches in the terminal UI like slash commands and themes. Make it work well on macOS and Linux, with tests and basic docs. Look up current docs online if you need to.
Want more depth? Deep Reverse