Parallel Execution Command

Parse the provided arguments to extract:

1. N: The total number of parallel executions requested
2. Prompt: The task to execute in parallel

Batch Processing for N > 10

If N exceeds 10:

1. Calculate batches: Total batches = ceiling(N / 10)
2. Process in sequential batches of up to 10 parallel tasks
3. Maintain a running accumulator of all results across batches
4. Apply aggregation across all batches at the end

Example: N=25 creates:

• Batch 1: Tasks 1-10 (parallel)
• Batch 2: Tasks 11-20 (parallel)
• Batch 3: Tasks 21-25 (parallel)

Execution Strategy

Analyze the prompt to determine:

1. What operations are being requested for the parallel workers
2. What post-processing or aggregation is requested for the results
3. What context from the current conversation is relevant
4. Whether file modifications are involved
5. How to partition work safely

Prompt Decomposition

Split the prompt into two parts:

1. Worker Instructions: What each parallel task should do
2. Aggregation Instructions: How to process the combined results

Look for aggregation keywords like:

• "return a report with..."
• "combine into..."
• "no duplicates"
• "rank by...", "with severity", "prioritize"
• "summarize as..."
• "merge the results..."

Context Preparation

Based on the prompt analysis:

• If file operations are mentioned, identify all target files using Glob or Grep
• If the prompt references "current", "this", or "the" items, gather that context
• Extract any relevant findings, decisions, or state from the conversation

Work Partitioning

If the prompt involves file modifications:

1. Distribute files across N tasks evenly (no task gets the same file)
2. For single-file operations, partition by line ranges or sections
3. For creation operations, assign unique namespaces or indices

If the prompt is analysis or read-only:

• Safe to run identical prompts with shared context
• Consider partitioning by search space or analysis scope

Parallel Execution

For each batch (up to 10 tasks per batch): Launch tasks simultaneously using the Task tool with:

• subagent_type: 'general-purpose'
• Each task receives:
  • The worker instructions portion of the prompt
  • Its specific partition of work
  • Necessary context from the conversation
  • Clear boundaries to prevent conflicts
  • Instructions to return structured results for aggregation

For file modifications, include in each task prompt:

• "You are task X of N total (batch Y of Z)"
• "Your assigned files/scope: [specific partition]"
• "Do not modify files outside your assignment"
• "Return your findings in a structured format for aggregation"

Between batches:

• Collect results from completed batch
• Update the work partition for the next batch
• Ensure no work overlap between batches

Safety Constraints

• No limit on total N (will batch automatically)
• Maximum 10 concurrent tasks per batch
• If file conflicts are unavoidable, reduce batch size or serialize critical sections
• For database or API operations, include rate limiting instructions
• Prevent duplicate work by clearly defining boundaries

Result Aggregation

After all batches complete, process combined results based on the aggregation instructions:

If deduplication is requested:

• Merge identical findings from different workers
• Keep one instance of each unique result

If ranking/severity is requested:

• Apply the specified ranking criteria
• Sort results by priority, severity, or importance
• Group by categories if appropriate

If reporting format is specified:

• Format results according to the requested structure
• Combine worker outputs into the specified report format
• Include statistics (total found, by category, etc.)

Default aggregation:

• Summarize the collective results
• Report any failures or conflicts
• Provide a unified view of what was accomplished

Always follow any specific post-processing instructions from the original prompt.

$ARGUMENTS