Squad Plan Skill

Generate a detailed parallel execution plan for multi-agent work on a complex feature.

When to Use

• Task touches ≥3 independent components/files
• Components don't share mutable state during the task
• Estimated single-agent time > 15 minutes
• Each subtask can own a distinct file set (no overlap)

Protocol

Phase 1: Analyze the Task

1. Read CLAUDE.md and claude/architecture.md for project context
2. Break down the task into independent subtasks
3. Identify natural parallelism boundaries (frontend/backend, services, features, layers)
4. Map dependencies — what must complete before something else starts
5. Identify shared resources requiring coordination (DB schemas, API contracts, shared types)

Phase 2: Design Workstreams (2-5 workstreams)

For each workstream:

• Workstream ID & Title — e.g., "WS1: API Layer"
• Objective — what this workstream accomplishes
• Files/Components owned — specific parts of the codebase
• Tasks — numbered checklist items with expected outcomes and verification steps
• Dependencies — which tasks from other workstreams must complete first
• Deliverables — what this workstream produces for others to consume

Phase 3: Integration Strategy

1. Define integration points between workstreams
2. Create integration test checklist
3. Specify merge order (which branch merges first)
4. Define smoke test for the combined result

Phase 4: Write ACTION_PLAN.md

Save to claude/tasks/ACTION_PLAN.md:

# ACTION_PLAN — [Feature Name]

## Overall Objective
[Clear statement of the goal]

## Workstream Definitions

### WS1: [Title]
**Objective:** [Goal]
**Files:** [List]

#### Tasks
- [ ] Task 1: [Description]
- [ ] Task 2: ...

**Dependencies:** None (can start immediately)
**Delivers to:** WS2 needs [output] from Task 3

### WS2: [Title]
...

## Integration Points
- [ ] [Integration task 1]

## Testing Strategy
- [ ] Unit tests per workstream
- [ ] Integration tests after merge
- [ ] End-to-end smoke test

Usage with Multiple Claude Code Instances

After generating, assign instances:

"You are the Backend Developer. Execute Workstream WS1 from claude/tasks/ACTION_PLAN.md."
"You are the Frontend Developer. Execute WS2, noting you depend on WS1 Task 3."

Each instance reads CLAUDE.md for project standards and follows the Explore-Plan-Act workflow.