Hacker News Reader

I’ve build myself a framework how a group of agents can verify if it’s done a good enough job.

After around 30 days I’ve implemented around 90% of how you manage a team in emergency medicine by accident and then applied the rest of it to it.

I’m quite happy with what they produce.

Could you describe a bit more what this framework is and how you built it?

I keep one git repository as the shared memory for every agent session.

1. I got a set of rules how it’s allowed to do work, and more importantly: what’s forbidden.

2. One state file per repo. It’s getting recorded what was tried, its results and the learnings from the experiments.

3. Tasks as Org mode files, one file per task and who works on it, aka codex, Claude code and which model.

Agent Hooks hold this together, the task and state files get updated automatically.

Code work happens on one branch per task. A second agent reviews each plan before code gets written.

I’m still baffled how well this works.