Usage

Run Brunch end to end, and understand what it's doing


This section covers working with Brunch in practice: writing a spec, compiling it into a plan, and building it with one AI coding agent per step — each in its own git worktree, streamed live into Petrinaut.

What's in this section

  • Run guide — the eight steps from an empty project to a finished build: set up your environment, write a spec, generate a plan, and cook it live. Includes the full command and flag reference.
  • How it works — the concepts behind the run: how a spec becomes a plan, how the plan compiles to a Petri net, the two-lane subnet each step runs through, and how a run streams to Petrinaut.

The core loop

Brunch follows a short, repeatable loop:

  1. Spec — describe the small app you want; Brunch interviews you and writes a reviewable specification.
  2. Plan — turn the spec into a plan.yaml of steps, test targets, and dependencies.
  3. Netcook compiles the plan's dependencies into a Petri net that knows what waits for what.
  4. Build — each step fires one coding agent in an isolated worktree, streamed live into Petrinaut.

🧪 Brunch is an experimental prototype. It works best for building small, self-contained CLI tools in TypeScript, and may struggle with existing codebases or large projects.

Ready to run it? Start with the Run guide.