Claude Code Ultracode

What Claude Code ultracode is, how /effort ultracode relates to xhigh and dynamic workflows, when to use it, when to avoid it, and how to control cost.

Claude Code ultracode is an effort setting for serious coding sessions. It combines xhigh reasoning effort with automatic dynamic workflow orchestration, so Claude can decide when a task should be split into a workflow instead of staying in one ordinary chat loop.

Last checked on May 29, 2026. Official Claude Code docs describe ultracode as part of dynamic workflows, which are still a research preview and require Claude Code 2.1.154 or later. npm currently reports @anthropic-ai/claude-code latest and next as 2.1.156, and stable as 2.1.145.

Quick Answer

Use ultracode when the task is broad, hard to verify in one pass, or important enough to justify extra usage:

/effort ultracode

Then ask for a scoped task:

Audit the auth, billing, and admin routes for missing permission checks. Use a workflow only if it helps. Report confirmed findings with file paths, evidence, and minimal fixes.

Do not turn on ultracode as a permanent default. It can be slower, more expensive, and more autonomous than a normal Claude Code session.

Ultracode In One Table

Setting	What it does	Best for	Main risk
`high`	Strong default reasoning for normal coding.	Everyday multi-file edits, debugging, implementation.	May still miss broad repository patterns.
`xhigh`	More reasoning depth without automatic workflow fan-out.	Hard debugging, architecture review, delicate refactors.	Higher token use and slower turns.
`ultracode`	`xhigh` plus automatic dynamic workflow orchestration.	Large audits, migrations, research, and cross-checked decisions.	Much higher usage if the task fans out widely.
`max`	Maximum reasoning depth where available.	Rare, very difficult tasks.	Slow and expensive; not a daily driver.

The key distinction: xhigh changes how hard Claude thinks. ultracode also lets Claude decide when the work should become a workflow with background agents.

Why People Search For Ultracode After Opus 4.8

The Opus 4.8 rollout created a few new search intents that deserve direct answers:

Search phrase	Real question behind it	Short answer
`what is ultracode effort`	Is this a new model, a plan, or a mode?	It is a Claude Code effort setting, not a separate model or subscription.
`Claude Code ultracode vs xhigh`	Which one should I use?	Use `xhigh` for depth; use `ultracode` when workflow orchestration may help.
`Claude Code ultracode not showing`	Why is the option missing?	Check version, plan/provider, `/config`, and organization admin settings.
`Opus 4.8 ultracode slow`	Is slow behavior expected?	Often yes: higher effort and workflow orchestration can take longer.
`Claude Code ultracode cost`	Will it burn more usage?	It can. Start scoped and check cost/usage surfaces before long unattended use.
`Claude Code workflow agents can code`	Are workflow agents only for review?	They can support large coding workflows, but edits should be scoped and tested.
`Claude Code ultracode thinking blocks`	Is the Opus 4.8 API error related?	Update to at least `2.1.156` if you hit the thinking-block API error.

These are not official keyword volumes. They are observed community and creator discussion patterns from the Opus 4.8 launch window.

When Ultracode Is Worth Trying

Good fits:

Task	Why ultracode can help
Repository-wide permission audit	Multiple agents can inspect different route groups and cross-check results.
Large framework or API migration	The task can be split by package, file group, or migration step.
Security or hardening sweep	Findings benefit from independent verification before implementation.
SEO/content inventory for a large site	Agents can compare content files, routes, metadata, sitemap, and sources.
Complex root-cause investigation	Several hypotheses can be explored in parallel, then reconciled.
High-stakes plan review	One workflow can draft, challenge, and refine the plan before edits.

Poor fits:

Task	Better option
One-file typo, import, or style fix	Normal `high` effort session.
A small feature with clear files	Plan Mode, then ordinary implementation.
Repeated prompt you already know	Slash command or skill.
Cost-sensitive exploration	Narrow prompt, `high` effort, and manual checkpoints.
Work needing constant decisions	Staged interactive session instead of a background workflow.

How To Enable Ultracode

Inside Claude Code:

/effort ultracode

If it is missing, check in this order:

Run claude --version.
Update to a version that supports dynamic workflows.
Open /config and look for Dynamic workflows.
Check whether you are on a plan or provider surface where workflows are available.
If you use an organization account, ask whether admins disabled workflows.
Restart the session after changing settings.

If the option appears but workflows do not run, ask explicitly:

Create a workflow for this audit, but start read-only and ask before edits.

Cost And Speed Expectations

Ultracode can feel slow for three different reasons:

Reason	What is happening	What to do
Higher reasoning effort	Claude is spending more thought on each step.	Use `high` unless the task truly needs `xhigh`.
Workflow fan-out	Claude may launch a background workflow with many agents.	Scope the first run and watch `/workflows`.
Verification loops	Agents may check findings before returning the final answer.	Ask for read-only findings first, then run implementation separately.
Large context or noisy files	Claude spends effort parsing old logs, MCP output, or broad files.	Use `/clear`, `/compact`, and narrow file paths.
Fast mode misunderstanding	Fast mode speeds output but has different pricing.	Do not use Fast mode as the default for long unattended workflow runs.

If the goal is cost control, use Context management, Pricing and limits, and Opus 4.8 slow or expensive troubleshooting.

Prompt Patterns That Work

Read-Only Audit

/effort ultracode
Run a read-only workflow to audit content/pages for stale Claude Code version claims. Do not edit files. Return a table with file path, claim, evidence, and recommended change.

Large Migration Pilot

/effort ultracode
Create a workflow for a migration pilot in src/modules/billing only. First inventory the affected files, then propose the smallest safe sequence. Do not edit until I approve the plan.

Verification Workflow

/effort ultracode
Use a workflow to verify the previous implementation. Check tests, route generation, sitemap inclusion, and obvious regressions. Report only evidence-backed issues.

The pattern is deliberate: scope, mode, evidence, and stopping condition. Without those, ultracode can over-expand.

Ultracode vs Dynamic Workflows

Ultracode is the switch. Dynamic workflows are the mechanism.

Term	Meaning
Ultracode	A Claude Code effort setting that enables `xhigh` plus workflow choice.
Dynamic workflow	The background orchestration run that Claude may create for a task.
`/workflows`	The place to watch running or completed workflow jobs.
`/deep-research`	A bundled workflow for research when web search is available.
Subagent	A focused worker with separate context; workflows can coordinate many.

Read Dynamic workflows for the full mechanism and Subagents for the smaller building block.

Common Mistakes

Mistake	Why it hurts	Better approach
Using ultracode for every prompt	It spends more effort than most tasks need.	Use `high` by default, then escalate.
Asking for "fix the whole repo"	The workflow may fan out without useful boundary.	Name directories, modules, and excluded files.
Skipping read-only planning	Many agents may discover and edit at once.	Separate inventory, plan, implementation, verification.
Treating community reports as facts	Launch-day reports can be noisy and account-specific.	Verify with docs, version, plan, and local behavior.
Ignoring admin settings	Enterprise and organization accounts can differ.	Check `/config` and managed settings.
Saving the first workflow as a command	You may preserve a bad process.	Save only after the run succeeds on a real task.