What problem does it solve? Sequential execution wastes turns when independent work slices could run at the same time, and unstructured delegation produces subagents that guess at intent without clear pass/fail targets. Ultrawork provides an in-session parallel execution discipline: ground context first, define acceptance criteria before launching lanes, fan out independent work to subagents, and close with concrete evidence instead of unverified claims. ## Core Features & Use Cases - Parallel lane orchestration: Fire independent subagent tasks in a single parallel call with explicit concurrency, model selection, and worktree isolation for colliding writes. - Agent tier guidance: A bundled reference maps work slices to builtin agents (scout, researcher, worker, reviewer, planner, oracle) by LOW/STANDARD/THOROUGH tiers. - Acceptance-first execution: Requires pass/fail criteria, diagnostics, and affected tests before completion is claimed, keeping verification lightweight but real. - Use Case: You need to fix a bug in a core module while also mapping regression coverage. Ultrawork keeps the coupled edit on the parent thread while a reviewer subagent runs as a parallel evidence lane, then closes with diagnostics and test output. ## Quick Start Ask the agent to run the ultrawork skill on your task, for example by saying "use ultrawork to fix the parser bug and review the changes in parallel".
