ultrawork

To execute independent tasks concurrently, you need a parallel execution engine that runs multiple workflows simultaneously while maintaining structured context. This approach accelerates throughput by delegating tasks to specialized lanes instead of processing them sequentially.

Lightweight evidence in parallel task management is a validation mechanism that runs alongside execution to support rapid iteration. It uses direct-tool and background evidence lanes to collect pass/fail criteria without relying on heavy, long-running persistence loops.

To set up parallel task delegation, you must define explicit task intent, constraints, and pass/fail criteria before execution. Providing these structured inputs allows the system to decide locally or delegate to specialized lanes for scalable throughput.

You can use parallel task execution without a persistence loop, but it avoids guaranteed long-running completion. It provides parallelism and context discipline, requiring an external persistence mode like Ralph alongside it if you need architect sign-off and completion guarantees.

You should avoid lightweight parallel task management when your workflows require guaranteed long-running completion, architect sign-off, or strict persistence. It is designed for high-throughput, independent tasks like data fetches and checks, not for dependent or stateful transactional processes.

Ultrawork provides parallelism for software engineering tasks but does not provide built-in persistence. It is a component focused on context discipline and delegation guidance, designed to operate alongside Ralph when persistence and long-running completion guarantees are needed.