How to use beads securely?

[abitrolly]

Running bd commands means access to Bash. It is not that I don't trust AI - I don't trust myself even, the scripts I write and commands I run in default brain mode. Backups are no joke, but maybe there is an easier way - to prevent disasters instead of dealing with them.

So what is the right setup to use beads? I want to use it as a human, but also with the help of aux intelligence. My capabilities as a human are limited, and I also suffer from the same problems described in https://steve-yegge.medium.com/introducing-beads-a-coding-agent-memory-system-637d7d92514a I find exciting that I can split things into smaller tasks and throw out too much context from my head. I would gladly let some NN create a tree of issues, so that I can review and import it into GitHub. I appreciate the tools to collaborate and do just that, but I don't want my PC to become a hazardous lab from the artifacts that had been left from future experiments.

So what is your rig setup? My best concentration is achieved when I meditate in offline, and I am quite GPU poor, so I think about llmfit + nono + llama.cpp + bd + (some local tracker), but I am not quite sure how to setup boundaries, sockets and request flow. I am not VS literate, so CLI is more comfortable for me.