using-workflowslisted

# Using workflows Claude Code's `Workflow` tool runs a deterministic JavaScript script that fans out subagents (`agent()`/`parallel()`/`pipeline()`, structured `schema`, token `budget`, resume). It is a *preferred Claude-Code-only accelerator* for exactly one orchestrator layer today (code review) — not a hard dependency. The markdown flow stays canonical and portable. The tool contract is documented at [workflows](https://code.claude.com/docs/en/workflows). ## When a chosen fan-out should run on the Workflow tool Whether to fan out at all is `dispatching-parallel-agents`' decision. Once a fan-out is chosen, route it through a workflow only if all three hold — a fan-out has to earn its coordination cost, and multi-agent runs use 3–10× the tokens of a single agent on the same task ([Anthropic](https://claude.com/blog/building-multi-agent-systems-when-and-how-to-use-them)): 1. **Breadth-first** — the work splits into parts that can run at once. 2. **Independent** — the parts don't share context or depend on each other's output mid-flight. 3. **High enough value** — the result justifies the multiplier (review of a real diff, a real migration, a real research sweep — not a one-line edit). Prefer a plain subagent unless the workflow buys something a subagent cannot: orchestration that is itself repeatable, or a quality pattern one pass can't apply — independent agents adversarially checking each other. A handful of delegated tasks in one turn is subagent-shaped. This plug