subagent-batching
Use when running multiple subagents. Provides patterns for parallel execution with 5-agent limit.
When & Why to Use This Skill
The subagent-batching skill provides optimized patterns and best practices for launching and managing multiple parallel subagents within Claude. It solves the bottleneck of sequential task execution by enabling efficient batch processing, robust error handling, and structured reporting for independent sub-tasks, significantly increasing throughput for complex agentic workflows.
Use Cases
- Bulk Code Analysis: Simultaneously processing multiple independent pull requests or source files to accelerate repository-wide reviews.
- Parallel Data Extraction: Running multiple subagents to extract information from a large set of independent documents or web pages concurrently.
- Automated Issue Triaging: Batch processing a high volume of independent bug reports or support tickets to categorize and route them in parallel.
- Distributed Content Generation: Orchestrating multiple agents to draft different sections of a large report or create social media posts for different platforms at once.
| name | subagent-batching |
|---|---|
| description | Patterns for launching and managing parallel subagents efficiently. |
| version | "2.0.0" |
| author | "JacobPEvans" |
Subagent Batching
Patterns for parallel subagent execution. Use your judgment on batch sizes based on task complexity.
When to Parallelize
Parallelize when:
- Processing multiple independent items (PRs, issues, files)
- Tasks have no dependencies between items
- Each item can succeed or fail independently
Don't parallelize when:
- Later items depend on earlier results
- Shared resources could cause conflicts
- You need to aggregate results before proceeding
Launch Pattern
Launch subagents in a single message with multiple Task tool calls. Don't launch sequentially.
# Correct: ONE message with multiple Task calls
Task 1: Process item A
Task 2: Process item B
Task 3: Process item C
# All launch simultaneously
Batching for Large Sets
For many items, batch them and validate between batches:
- Launch a batch of subagents in parallel (single message)
- Wait for all to complete using
TaskOutputwithblock=true - Validate results before proceeding
- Start next batch after validation
Subagent Prompts
Give each subagent focused context:
- Clear task description
- Relevant identifiers and locations
- Expected output format
- Success criteria
Error Handling
- One failure shouldn't block the rest
- Mark failed items for human review
- Continue processing remaining items
- Provide summary of successes and failures
Reporting
After completion, summarize:
- Total items processed
- Successes with brief descriptions
- Failures with reasons
- Any required follow-up actions