OpenAgentsControl/plugins/claude-code/skills/parallel-execution/SKILL.md

---

name: parallel-execution

description: Use when multiple subtasks have no shared files or dependencies and can be executed simultaneously.

Parallel Execution

Overview

Execute multiple independent tasks simultaneously using multiple subagent invocations in a single message. Reduces implementation time by 50-80% for multi-component features.

Announce at start: "I'm using parallel-execution to run [N] independent tasks simultaneously."

The Process

Step 1: Identify Parallel Tasks

Check subtask JSON files for parallel: true:

bash .opencode/skills/task-management/router.sh parallel {feature}

Output shows which tasks can run together:

Batch 1 (Ready now):
- subtask_01.json (parallel: true)
- subtask_02.json (parallel: true)

Batch 2 (After Batch 1):
- subtask_03.json (depends_on: ["01", "02"])

Step 2: Execute Parallel Batch

Make multiple task() calls in SINGLE message:

I'll execute Batch 1 tasks in parallel:

task(
  subagent_type="CoderAgent",
  description="Execute subtask 01",
  prompt="Read .tmp/tasks/feature/subtask_01.json and implement..."
)

task(
  subagent_type="CoderAgent",
  description="Execute subtask 02",
  prompt="Read .tmp/tasks/feature/subtask_02.json and implement..."
)

CRITICAL: Both task() calls in SAME message—not separate messages.

Step 3: Wait for All to Complete

All tasks in batch must complete before proceeding to next batch.

Step 4: Verify Batch Completion

bash .opencode/skills/task-management/router.sh status {feature}

Check all tasks in batch marked status: "completed".

Step 5: Execute Next Batch

Once Batch 1 complete, proceed to Batch 2:

task(
  subagent_type="CoderAgent",
  description="Execute subtask 03",
  prompt="Read .tmp/tasks/feature/subtask_03.json and integrate..."
)

Example: Converting Multiple Files

I'll convert these 5 subagent files in parallel:

task(subagent_type="CoderAgent", description="Convert auth-agent",
     prompt="Convert .opencode/agent/subagents/auth-agent.md to new format...")

task(subagent_type="CoderAgent", description="Convert user-agent",
     prompt="Convert .opencode/agent/subagents/user-agent.md to new format...")

task(subagent_type="CoderAgent", description="Convert payment-agent",
     prompt="Convert .opencode/agent/subagents/payment-agent.md to new format...")

task(subagent_type="CoderAgent", description="Convert notification-agent",
     prompt="Convert .opencode/agent/subagents/notification-agent.md to new format...")

task(subagent_type="CoderAgent", description="Convert analytics-agent",
     prompt="Convert .opencode/agent/subagents/analytics-agent.md to new format...")

Time savings: 5 tasks × 10 min = 50 min sequential → ~10 min parallel (80% faster)

Avoiding Conflicts

✅ GOOD (isolated files):

task(CoderAgent, "Create src/auth/service.ts")
task(CoderAgent, "Create src/user/service.ts")
task(CoderAgent, "Create src/payment/service.ts")

❌ BAD (same file):

task(CoderAgent, "Add auth function to src/utils/helpers.ts")
task(CoderAgent, "Add validation function to src/utils/helpers.ts")

This causes merge conflicts—run sequentially instead.

Pre-Load Shared Context

Load context ONCE before parallel execution:

# Load context BEFORE parallel tasks
Read: .opencode/context/core/standards/code-quality.md
Read: .opencode/context/core/standards/security-patterns.md

# Now all parallel tasks have same context
task(CoderAgent, "Implement auth service...")
task(CoderAgent, "Implement user service...")

DO NOT let each task discover context separately (slower).

Error Handling

Task fails during parallel execution:

• Fix failed task before proceeding to next batch
• Other tasks in batch can still be used

Tasks completing out of order:

• Don't start next batch until ALL current batch complete
• Use dependency tracking in subtask JSON

File conflicts:

• Review task breakdown—ensure true isolation
• If tasks MUST touch same file, mark parallel: false

Remember

• Make multiple task() calls in SINGLE message for parallel execution
• Tasks must be truly independent (no shared files/state)
• Pre-load shared context ONCE before parallel tasks
• Wait for ALL tasks in batch to complete before next batch
• Time savings: 50-80% reduction for multi-component features
• Optimal batch size: 2-4 tasks (up to 8 if truly independent)
• DO NOT parallelize tasks that modify same file
• DO NOT parallelize tasks with dependencies

Related

• task-breakdown
• context-discovery
• code-execution

---

Key Pattern: Multiple independent task() calls in SAME message = parallel execution.