worker-agents

To run parallel tasks without consuming your main context window, you can spawn isolated worker sessions. These ephemeral workers execute long-running research or data processing independently, preserving the primary agent's context for collaboration.

The best way to isolate long-running tasks is spawning ephemeral worker sessions. This isolates complex research or design work from the primary agent, preserving context and enabling scalable experimentation without blocking your main session.

You monitor progress and collect deliverables from parallel worker sessions via a shared communication bus. The bus handles two-way communication, reporting task progress to the parent agent until results are returned and the worker is cleanly torn down.

Yes, you can spawn one or more ephemeral worker sessions to handle independent research, data processing, or design tasks in parallel. Each worker operates in isolation and reports back through the shared bus upon completion.

Spawning isolated worker sessions requires no additional dependencies or components. The skill manages the worker lifecycle, per-task prompts, progress reporting, logging, and clean teardown natively within Claude Code.

You should avoid using parallel worker sessions for tasks that are not independent or long-running. Since workers operate in isolation with bus communication, tightly coupled tasks requiring continuous context sharing are better handled within the main session.