spawn-ralph

To run parallel subtasks with isolated Claude agents, you can spawn multiple Ralph instances that execute independent work concurrently. Each agent operates with a fresh context to prevent interference, facilitating a fan-out pattern for distributed computing.

The fan-out and fan-in pattern for distributed agent execution involves spawning isolated Claude agents to process independent subtasks concurrently, then collecting and aggregating their results. This approach improves workflow efficiency when subtasks do not require shared context.

To check the status and collect results from background agents, use the TaskOutput mechanism. This feature allows you to monitor spawned Ralph agents during background execution and retrieve their outputs once the parallel subtask processing is complete.

Yes, you can restrict tool access for spawned agents during parallel processing. The Skill supports tool restriction, allowing you to limit the capabilities of each isolated Ralph agent to match the specific requirements and security context of its assigned subtask.

You should use isolated agents for parallel execution when your workflow contains independent subtasks that do not require shared context. Spawning separate Ralph agents prevents context interference and allows concurrent processing for complex software engineering workflows.

Context isolation ensures each spawned agent operates with a fresh context, meaning agents do not share data during subtask execution. This prevents interference but requires you to aggregate independent results manually after the parallel processing completes.