AgentSkillsCN

parallel-development

基于Git Worktree的并行开发流程，专为BMAD项目量身打造。适用于并行故事开发的搭建、多条故事的Worktree创建、多智能体Claude Code会话的管理，以及并行执行任务的统筹安排。当用户提到“并行开发”“Worktree”“并行故事”“多智能体开发”“并行任务集”，或提出同时开发多条故事的需求时，这一流程将助您高效推进项目进度。

SKILL.md
--- frontmatter
name: parallel-development
description: Git worktree-based parallel development workflow for BMAD projects. Use when setting up parallel story development, creating worktrees for multiple stories, managing multi-agent Claude Code sessions, or organizing stories for parallel execution. Triggers include "parallel development", "worktree", "parallel stories", "multi-agent development", "parallel set", or requests to develop multiple stories simultaneously.

Parallel Development Skill

Enables parallel story implementation using Git worktrees with multiple Claude Code sessions. Each story runs in an isolated worktree with its own terminal, preventing merge conflicts through module boundary separation.

Configuration

Check for CONFIG.md in this skill directory for project-specific settings. If not present, prompt for:

  • •PROJECT_PREFIX: Worktree naming prefix (e.g., "myproject")
  • •REPO_PATH: Absolute path to main repository
  • •STORY_FORMAT: Story ID format (default: E{epic}-S{story})

Quick Start

1. Generate Parallel Development Plan

Use references/create-stories-prompt.md with the SM agent to analyze PRD/Architecture and generate a parallelization plan.

2. Create Workplan

Copy references/workplan-template.md to project's _bmad-output/parallel-development-plan.md and populate with generated stories.

3. Create Workflow

Copy references/workflow.md to project's _bmad-output/parallel-development-workflow.md and customize paths.

4. Execute Parallel Set

bash
# Setup worktrees for a parallel set
./scripts/parallel-dev.sh setup <set-number>

# Open separate terminals, one per worktree
# Run Claude Code in each: claude

# After completion, cleanup
./scripts/parallel-dev.sh cleanup <set-number>

Core Workflow (8 Phases)

Phase 0: Prerequisites - Foundation stories complete on main

Phase 1: Prepare Stories - Create story files using BMAD create-story workflow on main branch

Phase 2: Create Worktrees - Branch and worktree per story using parallel-dev.sh setup

Phase 3: Parallel Implementation - One Claude Code terminal per worktree, one story per agent

Phase 4: Per-Story Code Review - Run code-review in each worktree

Phase 5: Integration Branch - Merge all feature branches to integration/parallel-set-N

Phase 6: Integration Review - Combined review, full test suite

Phase 7: Pull Request - PR from integration branch to main

Phase 8: Cleanup - Remove worktrees, delete branches, update sprint status

Critical Rules

  1. •ONE AGENT = ONE STORY - Never work on multiple stories in a single session
  2. •PHASES ARE SEQUENTIAL - Complete each phase before proceeding
  3. •PHASE 1 ON MAIN - Story files must be created on main before worktrees
  4. •CLEANUP IS MANDATORY - Phase 8 must complete before next parallel set

References

  • •Full Workflow: references/workflow.md - Complete 8-phase process with commands
  • •Workplan Template: references/workplan-template.md - Parallel set organization template
  • •Story Prompt: references/create-stories-prompt.md - Prompt for generating parallelization plan

Script Usage

bash
# View available commands
./scripts/parallel-dev.sh help

# Setup parallel set (creates branches + worktrees)
./scripts/parallel-dev.sh setup 1

# List current worktrees
./scripts/parallel-dev.sh status

# Cleanup after merge (removes worktrees + branches)
./scripts/parallel-dev.sh cleanup 1

# Create integration branch and merge features
./scripts/parallel-dev.sh integrate 1

Story ID Format

Default format: E{epic}-S{story} (e.g., E2-S1, E3-S4)

  • •E = Epic number
  • •S = Story number within epic

Branch naming: feature/{story-slug} (e.g., feature/brand-model)

Worktree naming: ../{PROJECT_PREFIX}-{story-id} (e.g., ../myproject-E2-S1)