ClaudeDevelopment

blueprint

将单行目标转化为多会话、多代理工程项目的分步构建计划。每个步骤包含独立的上下文简介，以便新代理能直接执行。包括对抗性审查门、依赖图、并行步骤检测、反模式目录和计划突变协议。触发条件：当用户请求复杂多PR任务的计划、蓝图或路线图，或描述需要多个会话的工作时。不触发条件：任务可在单个PR或少于3个工具调用中完成，或用户说“直接执行”时。origin: community

144,923 stars

byaffaan-m

Complexity: easy

View on GitHub Installation ↓

About this skill

The Blueprint skill converts a high-level, single-line objective into a comprehensive, step-by-step construction plan suitable for any coding agent to execute from a cold start. It's designed to manage complex engineering projects that span multiple sessions or require coordination among several agents. Each step in the plan is self-contained with an independent context brief, enabling new agents to execute tasks directly without needing to recall previous session context. The planning process is robust, incorporating advanced features like adversarial review gates (using the strongest available model), dependency graphs, parallel step detection, an anti-pattern directory, and a plan mutation protocol. The skill operates through a 5-stage pipeline: 1. **Research:** Gathers context by pre-checking git, GitHub authentication, remote repositories, default branches, project structure, existing plans, and memory files. 2. **Design:** Decomposes the objective into manageable, single-PR-sized steps (typically 3-12), assigning dependency edges, parallel/serial order, model tiers (strongest vs. default), and rollback strategies. 3. **Drafting:** Writes self-contained Markdown plan files to a `plans/` directory. Each step includes a context summary, task list, validation commands, and clear exit criteria. 4. **Review:** Delegates an adversarial review to the strongest model sub-agent (e.g., Opus) against a checklist and anti-pattern directory, fixing critical findings before finalization. 5. **Registration:** Saves the finalized plan, updates memory indexes, and presents a summary of step count and parallelization to the user. This skill is invaluable for tasks where inter-session context loss often leads to rework.

Best use case

Breaking down large features into multiple, clearly ordered pull requests; planning complex refactoring or migration projects that span across several sessions; coordinating parallel workflows among sub-agents; and any task prone to rework due to context loss between AI agent sessions.

A self-contained, multi-step engineering plan (written to a `plans/` directory) with each step clearly defined. Each step includes a context summary, task list, validation commands, and exit criteria. The plan will also outline dependencies, parallel/serial order, model tiers for execution, and rollback strategies, ensuring any agent can pick up and execute a step independently and efficiently.

Practical example

Example input

Create a robust API for user management with authentication, authorization, and CRUD operations, ensuring it's production-ready and scalable.

Example output

```json
{
  "plan_summary": "Generated a 7-step blueprint for a user management API. This plan includes 2 parallel steps and incorporates adversarial review to ensure robustness.",
  "plan_location": "plans/user-management-api-blueprint.md",
  "first_step_preview": "Step 1: Project Setup and Basic Authentication (Serial)\nContext: Initialize the project, set up basic dependencies, and implement a foundational user authentication mechanism.\nTasks:\n- Initialize a new project directory.\n- Install necessary frameworks (e.g., Express.js, Flask).\n- Configure basic authentication middleware (e.g., JWT).\n- Create a simple user model and registration endpoint.\nValidation: Run unit tests for registration and login. Verify API endpoints return expected responses.\nExit Criteria: Basic user registration and login endpoints are functional and tested."
}
```

When to use this skill

When the user requests a plan, blueprint, or roadmap for a complex, multi-PR task, or describes work that requires multiple sessions to complete. It is especially useful for strategic planning of software development projects.

When not to use this skill

Do not use when the task can be completed within a single pull request, requires fewer than 3 tool calls, or when the user explicitly states directives like "just execute" or "直接做" (do it directly).

Installation

Claude Code / Cursor / Codex

$curl -o ~/.claude/skills/blueprint/SKILL.md --create-dirs "https://raw.githubusercontent.com/affaan-m/everything-claude-code/main/docs/zh-CN/skills/blueprint/SKILL.md"

Manual Installation

Download SKILL.md from GitHub
Place it in .claude/skills/blueprint/SKILL.md inside your project
Restart your AI agent — it will auto-discover the skill

How blueprint Compares

Feature / Agent	blueprint	Standard Approach
Platform Support	Claude	Limited / Varies
Context Awareness	High	Baseline
Installation Complexity	easy	N/A

Frequently Asked Questions

What does this skill do?

Which AI agents support this skill?

This skill is designed for Claude.

How difficult is it to install?

The installation complexity is rated as easy. You can find the installation instructions above.

Where can I find the source code?

You can find the source code on GitHub using the link provided at the top of the page.

Related Guides

Best AI Skills for Claude

Explore the best AI skills for Claude and Claude Code across coding, research, workflow automation, documentation, and agent operations.

AI Agents for Coding

Browse AI agent skills for coding, debugging, testing, refactoring, code review, and developer workflows across Claude, Cursor, and Codex.

ChatGPT vs Claude for Agent Skills

Compare ChatGPT and Claude for AI agent skills across coding, writing, research, and reusable workflow execution.

SKILL.md Source

# Blueprint — 施工计划生成器

将单行目标转化为分步施工计划，任何编码代理都能冷启动执行。

## 何时使用

* 将大型功能拆分为多个具有明确依赖顺序的 PR
* 规划跨多个会话的重构或迁移
* 协调子代理间的并行工作流
* 任何因会话间上下文丢失而导致返工的任务

**请勿用于** 可在单个 PR 内完成、少于 3 次工具调用，或用户明确表示“直接做”的任务。

## 工作原理

Blueprint 运行一个 5 阶段流水线：

1. **研究** — 预检（git、gh auth、远程仓库、默认分支），然后读取项目结构、现有计划和记忆文件以收集上下文。
2. **设计** — 将目标分解为适合单次 PR 的步骤（通常 3–12 步）。为每个步骤分配依赖边、并行/串行顺序、模型层级（最强 vs 默认）和回滚策略。
3. **草拟** — 将自包含的 Markdown 计划文件写入 `plans/`。每个步骤都包含上下文摘要、任务列表、验证命令和退出标准 — 这样新的代理无需阅读先前步骤即可执行任何步骤。
4. **审查** — 委托最强模型子代理（例如 Opus）根据清单和反模式目录进行对抗性审查。在最终确定前修复所有关键发现。
5. **注册** — 保存计划、更新内存索引，并向用户展示步骤计数和并行性摘要。

Blueprint 自动检测 git/gh 可用性。如果具备 git + GitHub CLI，它会生成完整的分支/PR/CI 工作流计划。如果没有，则切换到直接模式（原地编辑，无分支）。

## 示例

### 基本用法

```
/blueprint myapp "将数据库迁移到PostgreSQL"
```

生成 `plans/myapp-migrate-database-to-postgresql.md`，包含类似以下的步骤：

* 步骤 1：添加 PostgreSQL 驱动程序和连接配置
* 步骤 2：为每个表创建迁移脚本
* 步骤 3：更新仓库层以使用新驱动程序
* 步骤 4：添加针对 PostgreSQL 的集成测试
* 步骤 5：移除旧数据库代码和配置

### 多代理项目

```
/blueprint chatbot "将LLM提供商提取到插件系统中"
```

生成一个尽可能包含并行步骤的计划（例如，在插件接口步骤完成后，“实现 Anthropic 插件”和“实现 OpenAI 插件”可以并行运行），分配模型层级（接口设计步骤使用最强模型，实现步骤使用默认模型），并在每个步骤后验证不变量（例如“所有现有测试通过”、“核心模块无提供商导入”）。

## 主要特性

* **冷启动执行** — 每个步骤都包含自包含的上下文摘要。无需先前上下文。
* **对抗性审查门控** — 每个计划都由最强模型子代理根据清单进行审查，涵盖完整性、依赖关系正确性和反模式检测。
* **分支/PR/CI 工作流** — 内置于每个步骤中。当 git/gh 缺失时，优雅降级为直接模式。
* **并行步骤检测** — 依赖图识别出没有共享文件或输出依赖的步骤。
* **计划变更协议** — 步骤可以按照正式协议和审计追踪进行拆分、插入、跳过、重新排序或放弃。
* **零运行时风险** — 纯 Markdown 技能。整个仓库仅包含 `.md` 文件 — 无钩子、无 shell 脚本、无可执行代码、无 `package.json`、无构建步骤。安装或调用时，除了 Claude Code 的原生 Markdown 技能加载器外，不运行任何内容。

## 安装

此技能随 Everything Claude Code 附带。安装 ECC 时无需单独安装。

### 完整 ECC 安装

如果您从 ECC 仓库检出中工作，请验证技能是否存在：

```bash
test -f skills/blueprint/SKILL.md
```

后续更新时，请在更新前查看 ECC 的差异：

```bash
cd /path/to/everything-claude-code
git fetch origin main
git log --oneline HEAD..origin/main       # review new commits before updating
git checkout <reviewed-full-sha>          # pin to a specific reviewed commit
```

### 独立安装（内嵌副本）

如果您在完整 ECC 安装之外仅内嵌此技能，请将 ECC 仓库中已审查的文件复制到 `~/.claude/skills/blueprint/SKILL.md`。内嵌副本没有 git 远程仓库，因此应通过从已审查的 ECC 提交中重新复制文件来更新，而不是运行 `git pull`。

## 要求

* Claude Code（用于 `/blueprint` 斜杠命令）
* Git + GitHub CLI（可选 — 启用完整的分支/PR/CI 工作流；Blueprint 检测到缺失时会自动切换到直接模式）

## 来源

灵感来源于 antbotlab/blueprint — 上游项目和参考设计。

DevelopmentClaude