parallel-dispatch

# Parallel Dispatch Skill

> **Core Principle**: Parallelize only when tasks are truly independent.

## Decision Framework

```
┌─────────────────────────────────────────┐
│ Are tasks independent?                  │
│ (fixing one won't affect others)        │
└─────────────────┬───────────────────────┘
                  │
        ┌─────────┴─────────┐
        │                   │
       YES                  NO
        │                   │
        ▼                   ▼
┌───────────────┐    ┌──────────────────┐
│ Will they     │    │   SEQUENTIAL     │
│ edit the same │    │   Execute one    │
│ files?        │    │   at a time      │
└───────┬───────┘    └──────────────────┘
        │
    ┌───┴───┐
    │       │
   NO      YES
    │       │
    ▼       ▼
┌────────┐ ┌──────────────────┐
│PARALLEL│ │   SEQUENTIAL     │
│  OK    │ │   Avoid conflicts│
└────────┘ └──────────────────┘
```

---

## When to Parallelize

✅ **Safe for Parallel:**
- Independent test failures in different files
- Unrelated bugs in separate subsystems
- Documentation updates to different sections
- Adding features to isolated modules

❌ **Must be Sequential:**
- Failures that share root causes
- Tasks editing the same files
- Changes with dependencies between them
- Shared resource access (database, API limits)

---

## Parallel Dispatch Contract

Each parallel agent receives:

| Requirement | Why |
|-------------|-----|
| Isolated scope | One file or subsystem only |
| No shared resources | Prevent race conditions |
| Independent success criteria | Can be verified alone |
| Merge review after | Catch any conflicts |

---

## Dispatch Template

For each parallel task:

```yaml
task_id: [unique identifier]
scope: [specific file or subsystem]
isolation:
  files_to_edit: [list]
  files_read_only: [list]
  shared_resources: none
success_criteria:
  - [criterion 1]
  - [criterion 2]
```

---

## Post-Parallel Merge Protocol

After all parallel agents complete:

1. **Collect** all outputs
2. **Check