"Three humble cobblers surpass one Zhuge Liang" (三个臭皮匠,顶个诸葛亮) — A Chinese proverb reminding us that multiple ordinary agents, working together, can outperform a single genius. PRE Engineering is the embodiment of this wisdom.
The three agent roles — Plan, Review, Execute — form the acronym PRE. This also echoes "preparation": this skill provides the essential materials for a project to take flight.
When triggered, this skill detects the user's language (Chinese or English), confirms with the user, then guides through a 2-step interactive process to collect project requirements, generating 5 core documents:
- Project Goals Document — Requirements and constraints (only humans can modify)
- Collaboration Log — Agent communication medium with initial PLN_WAIT entry
- Planner Guide — Planner role behavior specification
- Executor Guide — Executor role behavior specification
- Reviewer Guide — Reviewer role behavior specification
Document names adapt to the selected language.
Recommended — use npx skills add:
npx skills add zhangzhanluo/pre-engineeringOr manually: place skills/pre-engineering/ into your Claude Code skills directory.
To update the skill to the latest version, simply re-run the install command — it will overwrite the skill files with the latest version:
npx skills add zhangzhanluo/pre-engineeringNote: Updating the skill does NOT affect existing .pre/{PROJECT_NAME}/ collaboration documents in your projects — those are separate from the skill templates and remain untouched.
- Initialize a multi-agent collaborative project
- Set up PRE system / Plan-Review-Execute workflow
- Create Planner/Executor/Reviewer guide documents
- Start a collaboration log-driven multi-agent workflow
- Express multi-agent collaboration intent (even without explicitly mentioning PRE)
skills/pre-engineering/
├── SKILL.md ← Skill definition
├── README.md ← This file (bilingual)
└── references/
├── en/ ← English templates
│ ├── planner-guide-template.md
│ ├── executor-guide-template.md
│ ├── reviewer-guide-template.md
│ ├── collaboration-log-format-reference.md
│ └── state-system-reference.md
└── zh/ ← Chinese templates (中文模板)
├── 规划者指导模板.md
├── 执行者指导模板.md
├── 审核者指导模板.md
├── 协作日志格式参考.md
└── 状态体系参考.md
- Install:
npx skills add zhangzhanluo/pre-engineering(recommended) or copy the directory manually - Trigger: Express multi-agent collaboration intent in Claude Code
- Language detection: Auto-detects your language and confirms with you
- Requirement collection: 2-step interactive process
- Document generation: 5
.pre/{PROJECT_NAME}/collaboration documents generated (with project-specific absolute paths) - Launch agents: Start three agents with
/loopcommands (using absolute paths — no directory switching needed)
- Language Detection: Auto-detects Chinese/English from user input
- Loop Prevention: 3 consecutive rejections trigger blockage → PLN_WAIT
- Loop Exit: On DONE, each agent self-cancels its /loop at runtime (CronList + CronDelete) — no job-ID registration in the log
- Version Recording: Reviewer manages git commit after execution reviews (no stash — direct commit)
- Code Conciseness: Reviewer enforces strict quality standards
Thanks to @import-wang for the creative inspiration and suggestions during development.
张战罗 (zhangzhanluo) — zhangzhanluo@outlook.com
"三个臭皮匠,顶个诸葛亮" — 多个平凡智能体协作,能胜过单一天才智能体。臭皮匠工程正是这一中国谚语的实践。
三个角色的英文名——Plan(规划)、Review(审核)、Execute(执行)——首字母组成了 PRE。同时,PRE 也是 preparation(准备)的缩写,暗含本 skill 的定位:为项目起飞提供必要的准备材料。
触发后,本 skill 自动检测用户语言(中文或英文)并与用户确认,通过2步交互流程收集项目需求,生成5份核心文档:
- 项目目标文档 — 功能需求与技术约束(仅人工可修改)
- 协作日志 — 智能体间通信媒介,含初始 PLN_WAIT 条目
- 规划者指导 — 规划者角色行为规范
- 执行者指导 — 执行者角色行为规范
- 审核者指导 — 审核者角色行为规范
文档名称根据所选语言自动适配。
推荐方式 — 使用 npx skills add 安装:
npx skills add zhangzhanluo/pre-engineering或手动将 skills/pre-engineering/ 目录放入 Claude Code skills 目录中。
重新运行安装命令即可更新至最新版本,已有 .pre/{PROJECT_NAME}/ 协作文档不受影响:
npx skills add zhangzhanluo/pre-engineering- 初始化多智能体协同项目
- 设置 PRE 系统 / 规划-审核-执行工作流
- 创建规划者/执行者/审核者指导文档
- 启动协作日志驱动的多智能体工作流
- 表达多智能体协同意图(即使未明确提到 PRE)
skills/pre-engineering/
├── SKILL.md ← Skill 定义
├── README.md ← 本文件(中英双语)
└── references/
├── en/ ← 英文模板
│ ├── planner-guide-template.md
│ ├── executor-guide-template.md
│ ├── reviewer-guide-template.md
│ ├── collaboration-log-format-reference.md
│ └── state-system-reference.md
└── zh/ ← 中文模板
├── 规划者指导模板.md
├── 执行者指导模板.md
├── 审核者指导模板.md
├── 协作日志格式参考.md
└── 状态体系参考.md
- 安装:
npx skills add zhangzhanluo/pre-engineering(推荐)或手动复制目录 - 触发:在 Claude Code 中输入多智能体协同意图
- 语言检测:自动检测你的语言并与你确认
- 需求收集:2步交互流程
- 文档生成:生成5份
.pre/{PROJECT_NAME}/协作文档(含项目专属绝对路径) - 启动智能体:用
/loop命令启动三个智能体(使用绝对路径,无需切换目录)
- 语言检测:从用户输入自动检测中/英文
- 循环阻断:连续驳回3次触发阻断 → PLN_WAIT
- 循环退出:DONE 后各角色运行时自主取消 /loop(CronList + CronDelete),不在日志登记 job ID
- 版本记录:审核者审核通过后直接 git commit(不使用 stash)
- 代码简洁性:审核者严格审核,大胆驳回冗余和膨胀
感谢 @import-wang 提供的创意启发和过程中的建议。
张战罗 (zhangzhanluo) — zhangzhanluo@outlook.com