arjraw10-cell/Kairos2 — reverse-engineered prompt

Reverse engineered prompt

Build me a minimal personal coding agent in Python that runs in the terminal and feels fast and polished. I want a clean chat style CLI with streaming responses, a thinking indicator, visible tool activity, and token usage shown after each turn. It should work with any OpenAI compatible endpoint from env settings, save chats automatically so I can resume them later, and compact old conversation when the context gets too full.

The agent should be able to read, write, edit, and search files, run shell commands in managed terminals, do basic git actions, and spin up child agents to help with tasks in parallel. I also want a skill system where it can load and create reusable skills stored as markdown files. Please include strong browser automation with Playwright style controls like opening pages, clicking, typing, selecting, scrolling, waiting, going back and forward, reloading, and capturing screenshots and snapshots. It should also support pasting text and images into the chat. Make it easy to run in the current folder or a chosen project. Look up current docs online if you need to.

Want more depth? Deep Reverse