orchestrating-parallel-worklisted

# Orchestrating Parallel Work ## Overview This skill provides patterns for coordinating parallel subagent execution in Claude Code workflows. Apply these patterns when tasks can run simultaneously without interdependencies — parallel fan-out saves wall-clock time and keeps large intermediate output out of the main context. ## When to Use - Launching multiple research or exploration agents - Implementing features with independent subtasks - Running diagnostics that check multiple layers - Processing batch operations with dependency graphs - Any workflow where "wait for A, then start B" isn't required ## Key Mechanics The **`Agent` tool** spawns a subagent with its own isolated context: - **Parallel launch**: to run agents concurrently, send multiple `Agent` calls in a **single message**. Each call takes `description`, `prompt`, and `subagent_type`. - **Results**: the agent's final message comes back as the tool result. It is not shown to the user — relay what matters. - **Background mode**: `run_in_background: true` returns immediately; the main conversation continues and completion arrives as an automatic task notification. No polling API exists — you are re-invoked when the agent finishes. - **Follow-ups**: `SendMessage` (addressed by the agent's ID or name) continues a previously spawned agent with its context intact. A fresh `Agent` call starts from zero. - **Isolation**: `isolation: "worktree"` gives an agent its own git worktree — required when parallel agents mut