AgentSkillsCN

harness-engineering

在任意代码仓库中快速构建以智能体为核心的知识库。本技能适用于初始化新项目，或为现有仓库添加结构化的知识框架时使用。支持可配置的基础目录、新建或已有的代码仓库、Claude 设置与计划目录，以及内置的智能体能力。可通过 /harness-init 命令调用。

SKILL.md
--- frontmatter
name: git-worktree
description: >
  This skill manages Git worktrees for isolated parallel development. It handles creating,
  listing, switching, and cleaning up worktrees with automatic .env file copying and
  shared-context configuration. Use when spinning up parallel feature branches, code
  review in isolation, or multi-agent development.

Git Worktree Manager

Manage Git worktrees for isolated parallel development with environment file management and shared-context support for multi-agent workflows.

When to Use

  • •Spinning up a parallel feature branch without disrupting current work
  • •Isolated code review of a PR
  • •Multi-agent development where each agent works in its own worktree
  • •Any time git stash / branch-switching friction is slowing things down

CRITICAL: Always Use the Manager Script

Never call git worktree add directly. Always use the manager script — it handles env file copying, .gitignore management, and writes .worktree-config for agent use.

bash
# Correct
bash .claude/skills/git-worktree/scripts/worktree-manager.sh create feature-name

# Wrong
git worktree add .worktrees/feature-name -b feature-name main

Commands

bash
SCRIPT=".claude/skills/git-worktree/scripts/worktree-manager.sh"

# Create new worktree (copies .env files, writes .worktree-config)
bash $SCRIPT create <branch-name> [from-branch]

# List all worktrees
bash $SCRIPT list

# Switch to a worktree
bash $SCRIPT switch <name>

# Refresh .env files from main repo
bash $SCRIPT copy-env [name]

# Clean up inactive worktrees
bash $SCRIPT cleanup

What Happens on Create

  1. •Updates the base branch from remote
  2. •Creates worktree at .worktrees/<branch-name>/
  3. •Copies all .env* files from main repo (except .env.example)
  4. •Writes .worktree-config with MAIN_REPO path
  5. •Adds .worktrees to .gitignore if missing

Shared Context Pattern

Worktrees are isolated file trees — they do NOT share docs/, plans/, or other non-git content with the main repo. To avoid drift and duplication:

For Agents Working in a Worktree

  1. •Read .worktree-config to get MAIN_REPO path
  2. •Reference shared docs from $MAIN_REPO/docs/ — do not duplicate them into the worktree
  3. •When creating or editing plans, write to $MAIN_REPO/docs/plans/
  4. •The worktree is for code changes only; docs live in the main repo
bash
# Example: read MAIN_REPO from config
source .worktree-config
cat "$MAIN_REPO/docs/plans/my-plan.md"

Why Not Symlinks?

Symlinks point one direction. If worktree/docs → main/docs, edits in the worktree modify main's files directly. With multiple agents in multiple worktrees, this creates race conditions. The config-based approach is explicit and safe.

Directory Structure

code
project-root/
├── .worktrees/                    # gitignored
│   ├── feat/new-feature/          # worktree 1
│   │   ├── .git
│   │   ├── .env                   # copied from main
│   │   ├── .worktree-config       # MAIN_REPO pointer
│   │   └── src/
│   └── fix/bug-123/               # worktree 2
│       ├── .git
│       ├── .env
│       ├── .worktree-config
│       └── src/
├── docs/                          # single source of truth
│   ├── plans/
│   └── references/
└── .gitignore                     # includes .worktrees

Integration Notes

  • •Worktrees share git objects with the main repo (lightweight, no duplication)
  • •Each worktree has its own branch — commits from one don't affect others
  • •Push from any worktree normally
  • •.worktree-config is auto-generated and should not be committed