$ultrawork
High-parallelism execution engine that classifies independent tasks by complexity and runs them simultaneously
$ultrawork is the parallelism layer of the OMX execution stack. It classifies tasks by complexity, routes them to the appropriate model tier, and runs independent tasks simultaneously without waiting for one to finish before starting the next.
ultrawork is a component. It handles parallel execution and model routing, while $ralph layers persistence and verification loops on top, and $autopilot layers the full lifecycle above that.
When to use it
- When you want multiple independent tasks done simultaneously
- When you need to delegate work to multiple agent tiers at once
- When you need pure throughput without ralph's persistence overhead
Avoid when: tasks have dependencies between them, or when it's hard to manage post-completion verification separately — $ralph is a better fit then.
Trigger keywords: "ulw", "ultrawork", "parallel", "run concurrently", "all at once"
How to invoke
codex
> ulw add missing type exports, update the README, and add integration tests for the auth middlewarecodex
> ultrawork implement the /api/users endpoint, add unit tests, and lint-fix the entire src/ directoryHow it works
Receives the task description and decomposes it into independent subtasks. Classifies by complexity into three tiers.
- Simple lookups and small edits → low-cost tier
- Standard implementation work → standard tier
- Deep analysis and complex refactoring → high-quality tier
All independent tasks run simultaneously. Only tasks with dependencies are serialized where needed. Long-running work like installs, builds, and tests runs via run_in_background so they don't block other tasks.
Once every task is done, a lightweight verification pass kicks in. The bar is concrete. Build and typecheck have to pass, the affected tests have to pass, and no new errors can sneak in. Anything that keeps failing gets reported rather than retried forever.
Outputs
- Completed artifacts from parallel workers (code changes, test results, docs)
- Lightweight build and test evidence confirming no regressions