Farmslot
How I build
My personal software factory. Farmslot runs AI coding agents in isolated worktree “slots”, validates each change against recipes that collect traces, screenshots and logs, links related attempts so runs can be compared, and gives me a Command Center plus an Expo companion app to supervise and review. It's in private beta and part of my daily workflow, not a finished product. The question it explores: what structure makes agentic work safe enough to scale? Repo: deeeed/farmslot.
Why?
I started Farmslot because one agent in one terminal is not enough once the work becomes serious. I wanted a way to dispatch work into isolated worktrees, keep the runs observable, validate the result with recipes, and recover context when a session crashes or needs a second pass.
The goal is not to make a flashy agent demo. The goal is to make my own development process more repeatable: assign work, isolate it, collect evidence, review it, and decide what deserves to be merged.
What it does
- Slot-based execution: run agents in isolated worktrees with assigned ports, devices, and runtime directories.
- Recipe validation: describe what a change must prove, then collect traces, artifacts, screenshots, and logs from the run.
- Family observability: keep related attempts connected so I can compare a first run, fixes, reviews, and reruns.
- Command Center: a web surface for supervising slots, runs, artifacts, and review state.
- Mobile companion: an Expo app for checking runs and decisions when I am away from the main machine.
Current status
Farmslot is in private beta and already part of my own workflow. I am preparing the iOS companion app for TestFlight/App Store distribution, but I am not presenting it as a finished product yet. It is still very much a tool I am using, hardening, and reshaping as my process changes.
What I am trying to learn
The interesting part is not just “can an agent write code”. The question I keep coming back to is: what structure makes agentic work safe enough to scale? For me that means isolation, recipes, review loops, good defaults for Expo/native projects, and a way to inspect the actual evidence instead of trusting a summary.