yeschef

Orchestrate multiple coding agents in parallel across git worktrees, using herdr (an "agent multiplexer") as the terminal layer.

yeschef lets one agent session (the head chef) dispatch and supervise a brigade of agents, each working on its own branch in its own git worktree inside its own workspace of a shared yeschef herdr session. You talk to the head chef; it spawns line cooks, steers them, reads their output, and reports back. It is agent-agnostic — a line cook is just a command launched in a herdr pane (claude, codex, aider, …), so nothing is tied to a particular vendor. yeschef attach hands the human herdr's native UI, where the whole brigade shows as live, status-coloured workspaces (herdr detects each agent's state).

Inspired by firstmate. Where firstmate is bash scripts + an AGENTS.md, yeschef is a single Rust CLI that is the toolbelt, plus an AGENTS.md that teaches the head chef to use it.

Requirements

git on your PATH, plus either Nix or Rust/Cargo to run yeschef itself from source (see Running yeschef). nix run bakes a bundled herdr onto yeschef's PATH, so you don't need one pre-installed; a Cargo build falls back to a herdr on your PATH. No containers, no macOS requirement.

Workflow

yeschef init                                  # ~/yeschef + AGENTS.md
yeschef project add <git-url> [name]          # bare clone + worktrees dir
yeschef refresh [<project>]                   # git fetch --prune (all projects if omitted)

# dispatch a line cook: worktree + herdr workspace + agent
yeschef spawn <project> <branch> -p "Implement X and summarize what changed"

yeschef status                                # who's running / gone + herdr's live state
yeschef peek  <project> <branch>              # read an agent's pane
yeschef send  <project> <branch> "use the helper in utils.rs"   # one-line steer
yeschef tui                                   # attach to herdr's native brigade UI
yeschef attach [<project> <branch>]           # watch the brigade live
yeschef restart                               # bounce the herdr server; workspaces restored, agents resumed
yeschef kill  <project> <branch> --rm-worktree
yeschef cleanup [<project>] [--yes]           # reap merged/gone + DONE tickets (dry run unless --yes)

spawn --agent <cmd> chooses the harness (default claude); -p/--prompt is passed as the agent's first argument.

Running yeschef

yeschef is never installed or on your PATH. The bare yeschef <args> written above (and throughout AGENTS.md) is shorthand — you always run it from the canonical source checkout at ~/yeschef/yeschef-src, which works from any directory and always runs the latest source there:

nix run ~/yeschef/yeschef-src -- <args>                          # reproducible default
cargo run --manifest-path ~/yeschef/yeschef-src/Cargo.toml -- <args>   # faster for tight loops

AGENTS.md is the head chef's manual — the dispatch → supervise → land → teardown loop. yeschef init writes a copy to ~/yeschef/, and it also ships in the source checkout.

Development

nix build              # or: cargo build
nix build .#clippy     # clippy -D warnings -D clippy::pedantic
nix build .#test       # or: cargo test  (unit tests, no external deps)

# e2e (real git + herdr; no containers/macOS needed)
cargo test --test e2e -- --ignored
nix run .#e2e          # puts git + herdr on PATH first

Status

The core orchestration loop (spawn / send / peek / status / kill / attach / restart). Heavier firstmate-style features — an event-driven supervision watcher, ship/scout ticket types, PR/merge automation, and persistent specialist agents — are not implemented yet.