ultrawork

name	ultrawork
description	Parallel execution engine for high-throughput task completion
argument-hint	<task description with parallel work items>
level	4

Ultrawork is a parallel execution engine and execution protocol for independent work. It emphasizes intent grounding, parallel context gathering, dependency-aware task graphs for non-trivial work, and concise evidence-backed execution summaries. It is a component, not a standalone persistence mode -- it provides parallelism and routing guidance, but not persistence, verification loops, or long-lived state management.

<Use_When>

• Multiple independent tasks can run simultaneously
• User says "ulw", "ultrawork", or wants parallel execution
• You need to delegate work to multiple agents at once
• Task benefits from concurrent execution but the user will manage completion themselves </Use_When>

<Do_Not_Use_When>

• Task requires guaranteed completion with verification -- use ralph instead (ralph includes ultrawork)
• Task requires a full autonomous pipeline -- use autopilot instead (autopilot includes ralph which includes ultrawork)
• There is only one sequential task with no parallelism opportunity -- delegate directly to an executor agent
• User needs session persistence for resume -- use ralph which adds persistence on top of ultrawork </Do_Not_Use_When>

<Why_This_Exists> Sequential task execution wastes time when tasks are independent. Ultrawork enables firing multiple agents simultaneously and routing each to the right model tier, reducing total execution time while controlling token costs. It is designed as a composable component that ralph and autopilot layer on top of. </Why_This_Exists>

1. **Read agent reference**: Load `docs/shared/agent-tiers.md` for tier selection 2. **Ground intent first**: Confirm whether the request is implementation, investigation, evaluation, or research; do not code before that is clear 3. **Gather context in parallel**: - direct tools for quick reads/searches - exploration/docs agents for broad context 4. **Classify tasks by independence**: Identify which tasks can run in parallel vs which have dependencies 5. **Create a task graph for non-trivial work**: - Parallel Execution Waves - Dependency Matrix - acceptance criteria and verification steps per task 6. **Route to correct tiers**: - Simple lookups/definitions: LOW tier (Haiku) - Standard implementation: MEDIUM tier (Sonnet) - Complex analysis/refactoring: HIGH tier (Opus) 7. **Fire independent tasks simultaneously**: Launch all parallel-safe tasks at once 8. **Run dependent tasks sequentially**: Wait for prerequisites before launching dependent work 9. **Background long operations**: Builds, installs, and test suites use `run_in_background: true` 10. **Verify when all tasks complete** (lightweight): - Build/typecheck passes - Affected tests pass - Manual QA completed for implemented behavior - No new errors introduced Three independent tasks fired simultaneously: ``` Task(subagent_type="oh-my-claudecode:executor", model="haiku", prompt="Add missing type export for Config interface") Task(subagent_type="oh-my-claudecode:executor", model="sonnet", prompt="Implement the /api/users endpoint with validation") Task(subagent_type="oh-my-claudecode:executor", model="sonnet", prompt="Add integration tests for the auth middleware") ``` Why good: Independent tasks at appropriate tiers, all fired at once. Correct use of background execution: ``` Task(subagent_type="oh-my-claudecode:executor", model="sonnet", prompt="npm install && npm run build", run_in_background=true) Task(subagent_type="oh-my-claudecode:executor", model="haiku", prompt="Update the README with new API endpoints") ``` Why good: Long build runs in background while short task runs in foreground. Sequential execution of independent work: ``` result1 = Task(executor, "Add type export") # wait... result2 = Task(executor, "Implement endpoint") # wait... result3 = Task(executor, "Add tests") # wait... ``` Why bad: These tasks are independent. Running them sequentially wastes time. Wrong tier selection: ``` Task(subagent_type="oh-my-claudecode:executor", model="opus", prompt="Add a missing semicolon") ``` Why bad: Opus is expensive overkill for a trivial fix. Use executor with Haiku instead. ## Relationship to Other Modes