Dokumentationerreichbar
Anthropic, Orchestrate subagents at scale with dynamic workflows
code.claude.com (externe Seite)
Die Herstellerdokumentation zu den Dynamic Workflows in Claude Code: ein Skript in Java
geprüft 10.11.2026
Worauf sich diese Seite beruft, wörtlich, abgerufen am 12.10.2026:
The difference is who holds the plan
bestätigt 10.11.2026A workflow moves the plan into code.
bestätigt 10.11.2026A workflow script holds the loop, the branching, and the intermediate results itself, so Claude's context holds only the final answer.
bestätigt 10.11.2026Reach for a workflow when a task needs more agents than one conversation can coordinate, or when you want the orchestration codified as a script you can read and rerun.
bestätigt 10.11.2026it can have independent agents adversarially review each other's findings before they're reported
bestätigt 10.11.2026A workflow spawns many agents, so a single run can use meaningfully more tokens than working through the same task in conversation.
bestätigt 10.11.2026To gauge the spend before committing to a large task, run the workflow on a small slice first
bestätigt 10.11.2026The keyword only chooses how Claude structures the work
bestätigt 10.11.2026Ultracode is a Claude Code setting that turns on automatic workflow orchestration for the session
bestätigt 10.11.2026A single request can turn into several workflows in a row: one to understand the code, one to make the change, and one to verify it.
bestätigt 10.11.2026This applies to every task in the session, so each request uses more tokens and takes longer than the same request without a workflow.
bestätigt 10.11.2026a session with ultracode on reaches a session or weekly limit sooner than the same work with it off
bestätigt 10.11.2026Turning ultracode on already opts you in to large runs
bestätigt 10.11.2026when you return to routine work
bestätigt 10.11.2026Claude Code sends the guideline to Claude as advice, not a cap
bestätigt 10.11.2026When a workflow schedules more than 25 agents, or its projected token total passes 1.5 million
bestätigt 10.11.2026The warning is advisory: it doesn't pause or limit the run.
bestätigt 10.11.2026Up to 16 concurrent agents by default
bestätigt 10.11.20261,000 agents total per run
bestätigt 10.11.2026Stopping the whole run doesn't count any agent as failed.
bestätigt 10.11.2026That last case means a failure in the middle of a fan-out reruns work that already finished.
bestätigt 10.11.2026The session's concurrent subagent limit isn't enforced for the subagents Claude spawns with the Agent tool
bestätigt 10.11.2026When nothing else assigns one, the agent runs on your session's model.
bestätigt 10.11.2026The first agent whose prompt differs from the previous run, because you edited the script or an earlier agent returned something different, runs again, and so does every agent after it, even ones that completed.
bestätigt 10.11.2026The run pauses only when all of these hold
bestätigt 10.11.2026On Pro, turn them on from the Dynamic workflows row in /
bestätigt 10.11.2026config. In the CLI, the per-run prompt shows the planned phases and these options:
bestätigt 10.11.2026To let the workflow start in these runs, use one of these:
bestätigt 10.11.2026In the Desktop app, an approval card shows the workflow name, the phase list, and a token-usage caution, with Once, Always, and Deny actions.
bestätigt 10.11.2026