Claude Ultracode: How Workflow Mode Works in Claude Code

Claude ultracode explained: the keyword and session setting that make Claude Code run tasks as multi-agent workflows, what it costs and when to use it.

Josh Pocock8 min read

Ultracode cover

AI Coding

Ultracode

Stride AI Academy

Claude ultracode is how you tell Claude Code to run work as a workflow: a script Claude writes that orchestrates many subagents in the background. You can use it two ways. Type the keyword ultracode in one prompt to run that task as a workflow, or turn on ultracode for the whole session with /effort ultracode, and Claude plans a workflow for every substantive task. Either way it uses more tokens than working through the task turn by turn.

This guide covers what ultracode does, the exact ways to turn it on and off, what happens during a run, what it costs and when it's worth it. It also explains how it differs from ultrathink, subagents and agent teams. Everything here was checked against Claude Code's workflows docs, the model configuration docs and the settings reference on November 20, 2026.

What ultracode is

Ultracode sits on top of Claude Code's workflows feature. A workflow is a JavaScript script that Claude writes for your task. A runtime executes it, spawning subagents to do the reading, editing and checking, while your session stays responsive.

The big change is who holds the plan. In a normal session, Claude decides turn by turn what to do next, and every result lands in its context window. In a workflow, the script holds the loop, the branching and the intermediate results. Claude's context only gets the final answer. That lets a workflow fan out across hundreds of files, or have independent agents check each other's findings before anything is reported.

Workflows are available on all paid plans and with Anthropic API access, plus Amazon Bedrock, Google Cloud's Agent Platform and Microsoft Foundry. On Pro, you first switch them on from the workflows row in /config.

How to turn on ultracode

There are two levels, and it helps to keep them apart.

For one prompt: the keyword

Put ultracode in a prompt you type:

ultracode: audit every API endpoint under src/routes/ for missing auth checks

Claude Code marks the word in your input and Claude writes a workflow for that task. Your session's settings don't change. Asking in plain words, such as "use a workflow", counts as the same opt-in.

The keyword only works in a prompt you type yourself: the terminal, an IDE extension panel, a Remote Control client, or an Agent SDK app that marks the input as human. It does not start a workflow from a claude -p prompt, a scheduled task, a webhook payload or a pull request comment.

Typed it by accident? Press Alt+W on Windows and Linux or Option+W on macOS to dismiss the highlight for that prompt, or press backspace right after the word. To stop the keyword triggering at all, turn off Ultracode keyword trigger in /config.

For a whole session: the setting

Turn ultracode on for the session and Claude decides when a task needs a workflow:

/effort ultracode
/effort ultracode off

You can also flip the Ultracode toggle in the /effort slider with Tab, then press Enter. To start a session with it on:

Terminal
claude --effort ultracode

That launch flag also sets the effort level to xhigh. Turning it on with /effort or the setting leaves your effort level where it was.

To have every session start with ultracode on, add this to a settings file:

JSON
{
  "ultracode": true
}

Note that the effortLevel setting and the CLAUDE_CODE_EFFORT_LEVEL variable don't accept ultracode as a value. Use the ultracode key. For where settings files live, see our Claude Code settings.json guide.

When it isn't available

Ultracode is off when workflows are turned off, or when your model doesn't support xhigh effort. In those cases claude --effort ultracode starts without it.

What happens when a workflow runs

An ultracode run

  1. 1You type ultracode or ask for a workflow
  2. 2Claude writes the workflow script
  3. 3You approve the planned phasesDepends on your permission mode
  4. 4Agents run in the background
  5. 5One report lands in your session

Before a run starts, the CLI shows the planned phases with options to run it, view the raw script or cancel. Whether you see that prompt depends on your permission mode. In auto mode you're asked on the first launch only, and not at all with ultracode on. In manual and accept edits mode you're asked every run.

While it runs, type /workflows to open the progress view. It shows each phase with its agent count, and you can open any agent to see its prompt, tool calls and result. Useful keys:

Key What it does
p Pause or resume the run
x Stop the selected agent, or the whole run
r Restart the selected running agent
s Save the run's script as a command

The agents' tool calls get the same permission checks and sandboxing as anything else in your session. The keyword changes how Claude structures the work, not what it's allowed to do.

By default the runtime runs up to 16 agents at once, fewer on machines with fewer CPUs, and caps a run at 1,000 agents. Runs can't take your input midway, so for sign-off between stages, run each stage as its own workflow.

Press s in /workflows to save a run you liked. It goes to .claude/workflows/ in your project or ~/.claude/workflows/ for you only, and runs as /<name> afterwards. Claude Code also ships one built-in workflow, /deep-research, which returns a cited report.

What ultracode costs

This is the part to take seriously. The workflows docs say a single run "can use meaningfully more tokens" than doing the same task in conversation, and runs count toward your plan's usage and rate limits.

With ultracode on for a session, it adds up faster. One request can become several workflows in a row: one to understand the code, one to make the change and one to verify it. That applies to every substantive task, so each request takes longer and uses more tokens. On a subscription, you hit a session or weekly limit sooner.

Turning ultracode on also switches off some guard rails, because you've already opted in to large runs:

  • The Large workflow warning, normally shown past 25 agents or a projected 1.5 million tokens, doesn't appear
  • The session's concurrent subagent limit isn't enforced for subagents Claude spawns
  • In auto mode, you aren't asked to approve the first workflow launch

Ways to keep the spend down:

  1. Try a small slice first. One directory instead of the whole repo.
  2. Watch /workflows. It shows each agent's token use, and you can stop the run there.
  3. Set a size guideline. /config workflowSizeGuideline=small asks Claude to aim for fewer than 5 agents. The default is medium (fewer than 10), or small on Pro.
  4. Use smaller models for easy stages. Ask for it when you describe the task.

For plan prices and API rates, see how much Claude Code costs.

When to use ultracode, and when not to

Use it when the task is bigger than one agent can hold in context, or the same step repeats across many items:

  • A codebase-wide audit, such as missing auth checks on every route
  • A large migration, such as every component from JavaScript to TypeScript
  • Fixing type errors until tsc passes
  • Research that needs sources checked against each other

Skip it for a one-file bug fix, a rename or anything you'll review line by line. Leave session-wide ultracode off for routine work, and use the keyword when a single task calls for it.

Ultracode vs ultrathink, subagents and agent teams

The names sound alike, but they control different things.

What it changes How long it lasts
Ultrathink How hard Claude reasons on one turn One prompt
Ultracode keyword Runs one task as a multi-agent workflow One prompt
Ultracode setting Claude plans workflows for every substantive task The session
Subagents Claude delegates a few tasks to workers, turn by turn Per task
Agent teams A lead agent supervises a handful of long-running peer sessions The team's life

The workflows docs put it in one line: with subagents and agent teams, Claude is the orchestrator, while a workflow "moves the plan into code". Subagents suit a few delegated tasks per turn. Agent teams suit a handful of peers that keep working if you interrupt. Workflows suit dozens to hundreds of agents and can be saved and rerun. Ultrathink is about depth of reasoning, not more agents.

How to turn workflows off

To block ultracode entirely, switch workflows off in /config, set "disableWorkflows": true in ~/.claude/settings.json, or set CLAUDE_CODE_DISABLE_WORKFLOWS=1. That stops the keyword triggering and removes the Ultracode toggle from /effort. Admins can set disableWorkflows in managed settings for a whole organization.

Learn to build with Claude Code

The Stride AI Coding course teaches Claude Code and Codex for building client tools, written for non-programmers, with weekly live calls to review your builds. Many of the free guides cover Claude Code setups, and the Claude Code community is where members compare theirs.

FAQ

What is ultracode in Claude Code?

It's the way to have Claude Code run work as a workflow, a script that orchestrates many subagents in the background. The ultracode keyword does it for one prompt, and /effort ultracode turns it on for every substantive task in the session.

How do I turn on ultracode?

Type ultracode in a prompt for one task. For a session, run /effort ultracode or launch with claude --effort ultracode. To start every session with it on, set "ultracode": true in a settings file.

Does ultracode use more usage?

Yes. A workflow spawns many agents, so a run can use meaningfully more tokens than the same task in conversation. With ultracode on for a session, every substantive request can turn into several workflows, so you reach usage limits sooner.

Is ultracode the same as ultrathink?

No. Ultrathink asks for deeper reasoning on one turn. Ultracode runs the task as a multi-agent workflow. One changes how hard Claude thinks, the other changes how the work is split up.

Why doesn't the ultracode keyword do anything?

The keyword only works in prompts you type, not in claude -p, scheduled tasks or relayed comments. It also stops working if workflows are disabled or the Ultracode keyword trigger is off in /config. On Pro, switch on workflows in /config first.

Build it with us in Stride AI Academy

Four courses, weekly live calls and a community of people building and selling AI. Or start with 320+ free guides.

Keep reading

Free AI Coding guides