Ultrawork Mode: Activate All Features with One Command
What You'll Learn
- Activate all advanced features of oh-my-opencode with a single command
- Make multiple AI agents work in parallel like a real team
- Avoid manually configuring multiple agents and background tasks
- Understand the design philosophy and best practices of Ultrawork mode
Your Current Challenges
You may have encountered these situations during development:
- Too many features, don't know how to use them: You have 10 specialized agents, background tasks, LSP tools, but don't know how to quickly activate them
- Manual configuration required: Every complex task requires manually specifying agents, background concurrency, and other settings
- Inefficient agent collaboration: Calling agents in series, wasting time and costs
- Tasks getting stuck midway: Agents don't have enough motivation and constraints to complete tasks
These are all affecting your ability to unleash the true power of oh-my-opencode.
Core Concept
Ultrawork Mode is oh-my-opencode's "one-click activate all" mechanism.
What is Ultrawork Mode?
Ultrawork Mode is a special working mode triggered by a keyword. When your prompt contains ultrawork or its abbreviation ulw, the system automatically activates all advanced features: parallel background tasks, deep exploration, forced completion, multi-agent collaboration, and more.
Design Philosophy
Ultrawork mode is based on the following core principles (from the Ultrawork Manifesto):
| Principle | Description |
|---|---|
| Human intervention is a failure signal | If you need to constantly correct AI output, it means there's a problem with the system design |
| Indistinguishable code | Code written by AI should be indistinguishable from code written by senior engineers |
| Minimize cognitive burden | You only need to say "what to do", agents are responsible for "how to do it" |
| Predictable, consistent, delegatable | Agents should be as stable and reliable as a compiler |
Activation Mechanism
When the system detects the ultrawork or ulw keyword:
- Set maximum precision mode:
message.variant = "max" - Display Toast notification: "Ultrawork Mode Activated - Maximum precision engaged. All agents at your disposal."
- Inject complete instructions: Inject 200+ lines of ULTRAWORK instructions to agents, including:
- Require 100% certainty before starting implementation
- Require parallel use of background tasks
- Force use of Category + Skills system
- Force completion verification (TDD workflow)
- Prohibit any "I can't do it" excuses
Follow Along
Step 1: Trigger Ultrawork Mode
Enter a prompt containing the ultrawork or ulw keyword in OpenCode:
ultrawork develop a REST APIOr more concisely:
ulw add user authenticationYou should see:
- A Toast notification pops up on the interface: "Ultrawork Mode Activated"
- Agent response starts with "ULTRAWORK MODE ENABLED!"
Step 2: Observe Agent Behavior Changes
After activating Ultrawork mode, agents will:
Parallel exploration of codebase
delegate_task(agent="explore", prompt="find existing API patterns", background=true) delegate_task(agent="explore", prompt="find test infrastructure", background=true) delegate_task(agent="librarian", prompt="find authentication best practices", background=true)Call Plan agent to create work plan
delegate_task(subagent_type="plan", prompt="create detailed plan based on gathered context")Use Category + Skills to execute tasks
delegate_task(category="visual-engineering", load_skills=["frontend-ui-ux", "playwright"], ...)
You should see:
- Multiple background tasks running simultaneously
- Agents actively calling specialized agents (Oracle, Librarian, Explore)
- Complete test plans and work breakdown
- Tasks continue executing until 100% complete
Step 3: Verify Task Completion
After agents complete, they will:
- Show verification evidence: Actual test run output, manual verification descriptions
- Confirm all TODOs complete: Won't declare completion early
- Summarize work done: List what was done and why
You should see:
- Clear test results (not "should work")
- All issues resolved
- No unfinished TODO list
Checkpoint ✅
After completing the above steps, confirm:
- [ ] You see a Toast notification after entering
ulw - [ ] Agent response starts with "ULTRAWORK MODE ENABLED!"
- [ ] You observe parallel background tasks running
- [ ] Agents use the Category + Skills system
- [ ] There's verification evidence after task completion
If any item fails, check:
- Is the keyword spelled correctly (
ultraworkorulw) - Are you in the main session (background tasks won't trigger the mode)
- Is the configuration file enabled with relevant features
When to Use This Technique
| Scenario | Use Ultrawork | Normal Mode |
|---|---|---|
| Complex new features | ✅ Recommended (requires multi-agent collaboration) | ❌ May not be efficient enough |
| Urgent fixes | ✅ Recommended (needs quick diagnosis and exploration) | ❌ May miss context |
| Simple modifications | ❌ Overkill (wastes resources) | ✅ More suitable |
| Quick idea validation | ❌ Overkill | ✅ More suitable |
Rule of thumb:
- Task involves multiple modules or systems → Use
ulw - Need deep research of codebase → Use
ulw - Need to call multiple specialized agents → Use
ulw - Single file small change → Don't need
ulw
Common Pitfalls
Important Notes
1. Don't use ulw in every prompt
Ultrawork mode injects a large amount of instructions, which is overkill for simple tasks. Only use it for complex tasks that truly require multi-agent collaboration, parallel exploration, and deep analysis.
2. Background tasks won't trigger Ultrawork mode
The keyword detector skips background sessions to avoid incorrectly injecting the mode into sub-agents. Ultrawork mode only works in the main session.
3. Ensure Provider configuration is correct
Ultrawork mode relies on multiple AI models working in parallel. If certain Providers are not configured or unavailable, agents may not be able to call specialized agents.
Lesson Summary
Ultrawork mode achieves the design goal of "activate all features with one command" through keyword triggering:
- Easy to use: Just enter
ulwto activate - Automatic collaboration: Agents automatically call other agents and execute background tasks in parallel
- Forced completion: Complete verification mechanism ensures 100% completion
- Zero configuration: No need to manually set agent priorities, concurrency limits, etc.
Remember: Ultrawork mode is designed to make agents work like a real team. You just need to express intent, agents are responsible for execution.
Next Lesson Preview
In the next lesson, we'll learn Provider Configuration.
You'll learn:
- How to configure multiple Providers like Anthropic, OpenAI, Google
- How multi-model strategy automatically degrades and selects optimal models
- How to test Provider connections and quota usage
Appendix: Source Code Reference
Click to expand source code locations
Last updated: 2026-01-26
| Feature | File Path | Line Numbers |
|---|---|---|
| Ultrawork design philosophy | docs/ultrawork-manifesto.md | 1-198 |
| Keyword detector Hook | src/hooks/keyword-detector/index.ts | 12-100 |
| ULTRAWORK instruction template | src/hooks/keyword-detector/constants.ts | 54-280 |
| Keyword detection logic | src/hooks/keyword-detector/detector.ts | 26-53 |
Key constants:
KEYWORD_DETECTORS: Keyword detector configuration, including three modes: ultrawork, search, analyzeCODE_BLOCK_PATTERN: Code block regex pattern, used to filter keywords in code blocksINLINE_CODE_PATTERN: Inline code regex pattern, used to filter keywords in inline code
Key functions:
createKeywordDetectorHook(): Creates keyword detector Hook, listens for UserPromptSubmit eventdetectKeywordsWithType(): Detects keywords in text and returns type (ultrawork/search/analyze)getUltraworkMessage(): Generates complete ULTRAWORK mode instructions (selects Planner or normal mode based on agent type)removeCodeBlocks(): Removes code blocks from text to avoid triggering keywords in code blocks
Business rules:
| Rule ID | Rule Description | Tag |
|---|---|---|
| BR-4.8.4-1 | Activate Ultrawork mode when "ultrawork" or "ulw" is detected | [Fact] |
| BR-4.8.4-2 | Ultrawork mode sets message.variant = "max" | [Fact] |
| BR-4.8.4-3 | Ultrawork mode displays Toast notification: "Ultrawork Mode Activated" | [Fact] |
| BR-4.8.4-4 | Background task sessions skip keyword detection to avoid mode injection | [Fact] |
| BR-4.8.4-5 | Non-main sessions only allow ultrawork keyword, blocking other mode injection | [Fact] |