Batch Large-Scale Changes

WorkflowAdvanced

Claude Code's official bundled skill /batch automates large-scale code changes. It researches the codebase to understand change scope, decomposes work into 5-30 independent units, processes each in a separate git worktree with an independent agent, and automatically creates PRs upon completion.

Trigger/batch

FrequencyFor large tasks

Team needing API migration across hundreds of files? Use /batch to auto-decompose and parallelize changes

Tech lead enforcing code standards across the codebase? Run /batch to split by module and generate PRs at once

BatchParallelWorktreeBundled SkillLarge-Scale

How It Works

Run /batch [description] → Research codebase

↓

Decompose into 5-30 independent units + execution plan

↓

Phase 2: Parallel agents per worktree

worktree-1

Process unit 1 changes

worktree-2

Process unit 2 changes

worktree-N

Process unit N changes

↓

✓ Auto-generate PRs per unit + change summary

Skill Code

# Batch Orchestrator — Built-in Bundled Skill
## Trigger: /batch [change description]

Claude Code's official bundled skill for large-scale codebase changes.

### Workflow:

1. Research phase:
  - Scan codebase for affected files
  - Identify dependencies and impact scope
  - Map change boundaries

2. Decomposition:
  - Split into 5-30 independent units
  - Ensure each unit is self-contained
  - Generate execution plan with ordering

3. Parallel execution (git worktrees):
```bash
# Each unit runs in its own worktree
.claude/worktrees/batch-unit-1/
.claude/worktrees/batch-unit-2/
.claude/worktrees/batch-unit-N/

# Each agent:
# 1. Creates a branch
# 2. Applies changes
# 3. Runs tests
# 4. Creates a PR
```

4. Post-processing:
  - Collect results from all worktrees
  - Report success/failure per unit
  - Generate summary with PR links

### Key features:
- Automatic conflict detection between units
- Shared memory across worktrees (v2.1.63+)
- Resume capability for failed units
- Dry-run mode for preview

### SKILL.md structure:
```yaml
---
name: batch
description: Research codebase, decompose, and parallelize changes
```

Copy and paste into your CLAUDE.md to start using immediately.

How Batch Large-Scale Changes Works

Batch scans the codebase to map change scope, decomposes work into 5-30 independent units, processes each in a separate git worktree, and auto-generates PRs after tests pass.