dispatching-parallel-agents

Purpose

When you have multiple independent failures or tasks that don't depend on each other, dispatch them in parallel. This skill defines how to identify, split, and recombine parallel work.

Step 1: Independence Check

Tasks are independent if:

• They don't read each other's outputs
• They don't write to the same files
• They don't share database transactions

task-A modifies src/auth.ts
task-B modifies src/payments.ts
task-C modifies src/notifications.ts
→ All independent → dispatch in parallel

task-A creates User model
task-B creates Auth that imports User
→ task-B blocked_by task-A → serial only

Step 2: Parallel Dispatch via Task Tool

Use the Task tool to spawn multiple agents simultaneously:

Spawn Task 1: {role: "developer", task: "implement auth module", files: ["src/auth.ts"]}
Spawn Task 2: {role: "developer", task: "implement payments module", files: ["src/payments.ts"]}
Spawn Task 3: {role: "tester",    task: "fix auth-abort.test.ts failures", files: ["tests/auth-abort.test.ts"]}

These run concurrently. Time to resolution: 1× instead of 3×.

Step 3: File Lock Protocol

To prevent race conditions, each parallel agent registers its file ownership:

Lock file: .jonggrang/locks/{agentId}.lock

{
  "agent_id": "developer-auth-001",
  "role": "developer",
  "files": ["src/auth.ts", "src/auth.test.ts"],
  "acquired_at": "2024-01-01T00:00:00Z"
}

Before writing any file:

1. Check if a lock exists for that file by another agent
2. If locked → add to your blocked list, skip for now
3. If unlocked → write your lock file, proceed

After completing task:

1. Remove your lock file

Step 4: Recombination

After all parallel agents complete:

1. Collect all outputs (JSON results from each agent)
2. Check for conflicts:
   • Same function modified by two agents → manual resolution needed
   • Imports added to same file → merge required
3. Run tests across the full set to verify no integration issues

Use This Skill For