I built a template repository to govern your AI chats with more grip than usual. It leverages PI.dev coding agent and an aggressive isolation done with Docker + VS Code
So far works on Linux and MacOS, and should also works on WSL2.
The project uses a container to run your preferred AI harnesses CLI. The container is optional, but protect your disk from bad things (I noticed rm -rf is always the preferred removal command of Deepseek :) a bit scary)
I am slowly extending it, aiming at
- reduce token usage - ability to install extra skill via pi.dev - experimenting with variants (like AWS AI-SDLC v1).
Please feel free to comment and give me hints.