parallel-execution

To run multiple subagent tasks in parallel, bundle all independent runSubagent calls into a single function_calls block. This triggers true parallel execution, significantly increasing throughput for multi-task workflows.

Executing directory-level analyses concurrently is achieved by grouping all independent subagent invocations into one function_calls block. This parallel execution pattern speeds up results across cross-module reviews.

Yes, you can synthesize results after parallel execution. The skill supports synthesis of results post-completion, allowing you to aggregate and process the outputs from multiple subagents once they finish running.

Parallel execution works for adversarial and perspective-based workflow patterns. It supports task-based, directory-based, perspective-based, and adversarial patterns, enabling concurrent processing for diverse cross-cutting reviews.

Sequential subagent invocations run one by one because they are not bundled into a single function_calls block. To achieve true parallel execution, all independent runSubagent calls must appear together within one function_calls block.

You should avoid bundling function_calls for parallel processing when subagent tasks are dependent on each other. This approach requires all runSubagent calls to be independent to correctly execute and synthesize results in parallel.