# Mission: SwarmForge (unclebob/swarm-forge)

## Why
You want a **complete mental model** of [unclebob/swarm-forge](https://github.com/unclebob/swarm-forge) so you can **reimplement and customize** it — likely in **Rust** or **TypeScript with a TUI** — without cargo-culting shell and Babashka details. The outcome is not “run the original once”; it is **own the design**: what it does, how the pieces fit, and which decisions are load-bearing vs incidental.

## Success looks like
- Explain SwarmForge in one paragraph and in three layers (runtime, handoff protocol, discipline/prompts)
- Trace startup → worktree → agent launch → handoff → completion without notes
- Name which components a Rust/TS rewrite must keep vs can replace (tmux, daemon, terminal adapters, packs)
- Spot customization seams: packs, roles, constitution, receive mode, backends
- Sketch a migration boundary map (domain core vs OS integration vs UX shell)

## Constraints
- Teach from **primary sources**: the cloned repo under `/tmp` (not recollections)
- Prefer **deep architecture** over “how to install and run” tourism
- Lessons stay short; reference docs hold the dense cheat-sheets
- Migration language (Rust vs TS) is deferred until the domain model is solid
- Clone and experimental forks stay in `/tmp`; this teach folder is the learning workspace only

## Out of scope
- Shipping your rewrite in this repo (learning first)
- Full Uncle Bob methodology / Clean Architecture course beyond what SwarmForge encodes
- Operating production multi-agent farms or cloud orchestrators (k8s, Temporal, etc.) as primary focus
- Banking/token scams or anything related to the README warning about SWARM tokens