⚡ EXECUTION INSTRUCTIONS FOR CLAUDE

When this command is invoked, YOU (Claude) must execute these steps immediately: This is NOT documentation - these are COMMANDS to execute right now. Use TodoWrite to track progress through multi-phase workflows.

🚨 EXECUTION WORKFLOW

Phase 1: Execute Documented Workflow

Action Steps:

1. Review the reference documentation below and execute the detailed steps sequentially.

📋 REFERENCE DOCUMENTATION

Parallel Tasks vs Subagents Reference Guide

Purpose: Central reference for deciding between parallel tasks and subagents to optimize token usage

Quick Decision Guide

🚀 Use Parallel Tasks (0 extra tokens)

When: Simple, independent operations < 30 seconds each How: Background processes, GNU parallel, xargs, or batched tool calls

🤖 Use Subagents (~50k+ tokens each)

When: Complex workflows, different directories, or operations > 5 minutes Cost: 100x more expensive than parallel tasks

Implementation Methods

1. Background Processes (Bash &)

# Best for: 2-10 simple commands

(command1) &
(command2) &
(command3) &
wait

Pros: Simple, built-in, error handling possible Cons: Manual PID management, limited scaling

2. GNU Parallel

# Best for: Structured parallel execution

parallel -j 4 ::: "cmd1" "cmd2" "cmd3" "cmd4"

# With input files

parallel -a files.txt process_file {}

Pros: Advanced job control, progress bars, automatic load balancing Cons: Requires installation, learning curve

3. xargs Parallel

# Best for: Processing file lists

find . -name "*.py" | xargs -P 8 -I {} python lint.py {}

Pros: Standard Unix tool, efficient for file operations Cons: Limited error handling, basic job control

4. Batched Tool Calls

# Best for: Multiple Claude tool calls

# Execute multiple searches/reads in one response

Pros: Most efficient for Claude operations, atomic execution Cons: Limited to tool capabilities

Decision Criteria

Factor	Parallel Tasks	Subagents
Token Cost	~0 additional	~50k+ each
Execution Time	< 30 seconds	> 5 minutes
Complexity	Simple, independent	Complex, multi-step
Directory	Same working dir	Different branches/dirs
Failure Handling	Basic	Advanced isolation
Context Needed	Shared	Independent

Common Scenarios

✅ Perfect for Parallel Tasks

• Search codebase for patterns
• Run test suites
• Check multiple PR statuses
• Lint multiple files
• Aggregate data from APIs
• Build multiple components

✅ Requires Subagents

• Implement new features
• Fix complex bugs
• Create/update PRs
• Multi-branch operations
• Architectural changes
• Long-running migrations

Token Cost Examples

10 file searches:
- Parallel: 0 additional tokens
- Subagents: 500,000+ tokens (10 × 50k)
- Savings: 100%

Feature implementation:
- Parallel: Not suitable (complex workflow)
- Subagent: 50k tokens (justified by complexity)

Monitoring Success

# Before optimization

ccusage daily --instances

# After switching to parallel

ccusage daily --instances --since 2025-07-29

# Expected: 90%+ reduction in token usage

Integration with Commands

This guide is referenced by:

• /plan - Planning execution strategy
• /execute - Choosing execution method
• /arch - Architectural task distribution
• /test - Test execution strategy
• CLAUDE.md - Core operating guidelines

---

Remember: Default to parallel tasks. Only use subagents when truly needed for complexity or isolation.