**Workflow Toolbox is the proportionate agent-orchestration scale for Claude Code.** Pick the right rung for the task in front of you — a single agent, a team of agents, a pilot, an orchestrator, or a deterministic workflow. Most tasks want a low rung; the toolbox exists for the work that genuinely earns a higher one.

The orchestration scale

Rung	What it is	Reach for it when
Plain agent	One Claude Code subagent, one context.	The task is self-contained.
A team of agents	A few agents messaging each other.	The work splits into a couple of coordinated roles.
Pilot	One agent drives a whole unit of work end-to-end — intake → plan → do → verify → report — escalating only when it must.	You want one card or ticket carried autonomously.
Orchestrator	One agent spawns and arbitrates a wave of pilots.	Several coupled units must move at once.
Deterministic workflow	A JavaScript script owns the loops, branches, and fan-out; only the leaf agent() calls think, each in a fresh context.	The work fans out, needs independent verification, or is too big for one context window.

**You don't write any of this by hand.** You ask Claude in plain language, and the shipped skills compose the right rung for you — describe a process in a sentence and the `workflow-composer` skill writes the workflow; ask to drive a card end-to-end and the `pilot-wave` skill composes the briefed pilot. What comes back is plain JavaScript you can read, keep, and edit — a starting point, not a black box.

The first two rungs are Claude Code's own n