Handle-Large-Tasks

Breaking down large projects involves splitting them into smaller, independent tasks assigned to subagents, each with clear scope and tests. This approach overcomes context window limitations by delegating specific components—such as API layers, database schemas, or authentication modules—to focused agents that work in parallel, then integrating results.

Organize subagents by functional component: define precise responsibilities for each (e.g., API handler, data layer, auth service), write tests upfront to specify expected behavior, then execute each subagent independently with the Task tool. This test-driven approach ensures coherence when components integrate.

Yes. Subagent orchestration is designed specifically for this. By splitting a task into discrete, bounded units with clear inputs and tests, each subagent operates within its own context, making it possible to tackle projects—like building microservices or large features—that would otherwise exhaust a single context window.

After each subagent completes its task, review and iterate on outputs against the original test suite. Run integrated tests across all components to confirm coherence, then use restarts to reuse prior plans if adjustments are needed. This iterative feedback loop ensures the final result is unified and correct.

Define clear, testable requirements for each subagent: write tests that specify expected behavior upfront, then document precise task boundaries and inputs. This test-driven approach ensures subagents know exactly what to build and makes it straightforward to verify their outputs integrate properly.

Split projects when they span long horizons or require multi-component orchestration that exceeds your context window. Examples include microservice builds, multi-layer feature implementations, or any task requiring coordination across distinct, independently testable components.