TL;DR: Claude Code Ultracode is an “xhigh effort” mode triggered by the
/ultraplancommand, which automatically creates dynamic workflow scripts in JavaScript. The system orchestrates up to 1000 parallel subagents according to MarkTechPost data, eliminating the need for manual control of each step. The whole thing runs on Claude Opus 4.8 and is available as a research preview.
How Does Claude Code Ultracode and Maximum Effort Mode Work?
Claude Code Ultracode is a configuration triggered by the /ultraplan command, which sets the effort level to “xhigh” — a value higher than the standard “high.” According to FindSkill.ai documentation, this mode automatically activates dynamic workflows when a task exceeds a certain complexity threshold. The system decides on its own whether a given problem requires subagent orchestration.
In practice, this means Claude Code stops being a simple chatbot. Instead, it generates JavaScript scripts that coordinate the work of multiple agents simultaneously. Each subagent receives a specific fragment of the task and operates independently. The results are then aggregated by the orchestration script.
The xhigh mode consumes more tokens than the standard approach. For this reason, Anthropic recommends using it only for tasks requiring deep analysis, such as migrating large codebases. For simpler tasks, the standard “high” level is sufficient.
What Are Dynamic Workflows and Why Do They Replace the Traditional Approach?
Dynamic workflows are a mechanism in which Claude Code generates JavaScript scripts that serve as the orchestration layer. Instead of relying on conversation context, the system creates standalone programs managing the workflow. TechTimes reports that this approach eliminates the problem of response quality degradation during long sessions.
The traditional approach required continuous human-model interaction. Each step required approval, and the conversation context grew with each turn. Dynamic workflows solve this problem — the script operates independently, launching subagents according to a defined plan.
The JavaScript script determines the execution order, dependencies between steps, and how results are aggregated. Subagents don’t share conversation context — each receives an isolated task, which prevents context window overflow.
How Many Subagents Can Run Simultaneously in Ultracode?
According to MarkTechPost, Anthropic set a limit of 1000 subagents running in parallel within a single dynamic workflow. This is significantly more than in previous implementations, where the number of simultaneously operating agents was limited to a few dozen.
| Parameter | Standard Mode | Ultracode (xhigh) |
|---|---|---|
| Maximum number of subagents | no data | 1000 |
| Effort level | high | xhigh |
| Automatic orchestration | no | yes |
| Generated JS scripts | no | yes |
| Context isolation | partial | full |
1000 subagents is the upper limit. In most cases, the system launches far fewer agents — the exact number depends on task complexity. For example, refactoring a single file requires just one agent, while migrating an entire repository may require several hundred.
What Commands Activate Ultracode in Claude Code?
The main command activating Ultracode is /ultraplan, described in detail on claude code /ultraplan documentation. This command sets the effort level to “xhigh” and automatically enables dynamic workflows for tasks meeting complexity criteria.
Additionally, there is the option to manually set the effort level through the API. FindSkill.ai documents the effort: "xhigh" parameter that can be passed in a request to Claude Opus 4.8. This gives greater control over model behavior.
Ways to activate Ultracode:
/ultraplancommand in the Claude Code CLI interfaceeffort: "xhigh"parameter in an API request to Opus 4.8- Automatic activation for high-complexity tasks
- Configuration in project settings file
- Mid-conversation system messages changing effort level during a session
- Command-line flag
- Environment variable configuring default mode
How Does Ultracode Handle the Context Window Problem?
One of the main challenges in working with language models is response quality degradation during long conversations. Ken Huang on his blog describes how the traditional “chat and wait” approach leads to context overflow and decreased precision of responses.
Dynamic workflows solve this problem through isolation. Each subagent operates in its own context space and doesn’t share conversation history with others. The orchestration script manages data flow between agents, passing only essential information.
As a result, even complex tasks don’t cause quality degradation. The main agent’s context remains clean because the heavy work is delegated to subagents. This approach resembles microservices architecture — each component does one thing well.
What Are the Costs and Limitations of Ultracode Mode?
The Ultracode mode consumes more tokens than the standard approach. Each subagent is a separate model invocation, and with a limit of 1000 agents, costs can quickly escalate. FindSkill.ai warns about the so-called “token-burn catch” — a situation where automatic orchestration generates more invocations than necessary.
Anthropic recommends using Ultracode only for tasks that genuinely require parallel processing. For simple tasks, the standard mode with “high” level is sufficient. It’s worth starting with smaller test tasks before launching full orchestration on a large project.
Additionally, Dynamic Workflows are still in the research preview phase, which means potential changes in behavior and limits.
What Tasks Is Ultracode Best Suited For?
MindStudio points to codebase migrations as the primary use case. Example: moving a project from JavaScript to TypeScript, where each file requires individual transformation, is ideally suited for parallel processing.
Other applications include:
- Mass refactoring of multiple files simultaneously
- Security analysis of large repositories
- Documentation generation for entire projects
- Large-scale test automation
- Dependency auditing in complex microservice architectures
- Code synchronization between multiple repositories
- Mass package version updates in monorepos
- Anti-pattern detection in large projects
FAQ
How many parallel subagents does Ultracode support? Anthropic set a limit of 1000 subagents running in parallel in a single dynamic workflow according to MarkTechPost, though in most cases the system launches far fewer agents depending on task complexity.
What command activates Ultracode mode? The /ultraplan command sets the effort level to xhigh and automatically enables dynamic workflows according to FindSkill.ai, or you can use the effort: "xhigh" parameter in an API request to Opus 4.8.
Are Dynamic Workflows a stable feature? Dynamic Workflows are in the research preview phase, meaning potential changes in behavior and limits — don’t rely on them in production without testing.
When should you not use Ultracode? For simple tasks, the standard “high” mode is sufficient, since Ultracode consumes more tokens through automatic subagent orchestration.
Summary
Claude Code Ultracode introduces xhigh effort mode with automatic orchestration of up to 1000 subagents. Dynamic workflows replace the traditional “chat and wait” approach with JavaScript scripts managing the workflow. Each subagent operates in an isolated context, eliminating quality degradation during long sessions.
Key takeaways:
- Ultracode sets effort level to xhigh, higher than standard high
- Dynamic workflows generate JavaScript scripts orchestrating subagents
- Limit of 1000 subagents running in parallel according to MarkTechPost
- Feature is in research preview phase — may be subject to changes
- Token consumption is higher, requiring caution for simple tasks
Test Ultracode on a small project using the /ultraplan command and compare results with standard mode.