← 返回 Skills 市场
sydpz

AI-DLC

作者 BingWang · GitHub ↗ · v1.1.0 · MIT-0
cross-platform ✓ 安全检测通过
221
总下载
0
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install ai-dlc
功能描述
AI-Driven Development Life Cycle (AI-DLC) adaptive workflow for software development. Use when: starting a new project, new feature, bug fix, refactoring, mi...
使用说明 (SKILL.md)

AI-DLC: AI-Driven Development Life Cycle

AI-DLC 是一种结构化但自适应的软件开发方法论,由 AI 主导流程、人审批把关。

核心原则

流程适应工作,而不是反过来。 AI 根据请求复杂度、现有代码库状态和风险评估,智能决定需要执行哪些阶段。

  • 简单变更 → 只执行必要阶段
  • 复杂项目 → 完整三阶段 + 所有保障机制
  • 每个阶段都有审批门控,必须人确认后才能继续

三阶段概览

用户请求 → INCEPTION → CONSTRUCTION → OPERATIONS → 完成
            (计划)      (构建)        (运营)

🔵 INCEPTION 阶段 — 做什么 + 为什么

  • Workspace Detection (始终执行) — 分析是 greenfield 还是 brownfield
  • Reverse Engineering (brownfield 有现有代码时执行) — 分析现有代码库
  • Requirements Analysis (始终执行,深度自适应) — 收集需求并提问澄清
  • User Stories (条件执行) — 创建用户故事和角色
  • Workflow Planning (始终执行) — 确定执行计划
  • Application Design (条件执行) — 高层组件设计
  • Units Generation (条件执行) — 拆解为工作单元

🔵 INCEPTION 阶段 — 做什么 + 为什么

  • Workspace Detection (始终执行) — 分析是 greenfield 还是 brownfield
  • Reverse Engineering (brownfield 有现有代码时执行) — 分析现有代码库
  • Requirements Analysis (始终执行,深度自适应) — 收集需求并提问澄清
  • User Stories (条件执行) — 创建用户故事和角色
  • Workflow Planning (始终执行) — 确定执行计划
  • Application Design (条件执行) — 高层组件设计
  • Units Generation (条件执行) — 拆解为工作单元

🟢 CONSTRUCTION 阶段 — 怎么做

  • Per-Unit Loop (每个工作单元依次执行):
    • Functional Design (条件执行)
    • NFR Requirements (条件执行)
    • NFR Design (条件执行)
    • Infrastructure Design (条件执行)
    • Code Generation (始终执行) — Part 1 计划 → Part 2 生成
  • Build and Test (始终执行) — 构建、单元测试、集成测试
  • Changelog (始终执行) — 每次合并到 master 后更新 aidlc-docs/aidlc-state.md 的 changelog(格式:时间/commit/变更内容),SubAgent 靠此了解进展

🟡 OPERATIONS 阶段 — 部署 (占位)


MANDATORY: 工作流执行规则

规则文件加载顺序

开始工作流时,必须按此顺序加载规则文件:

  1. 加载 references/common/welcome-message.md — 显示欢迎消息(新项目只执行一次)
  2. 加载 references/common/process-overview.md — 工作流概览图
  3. 加载 references/common/session-continuity.md — 会话恢复指引
  4. 加载 references/common/question-format-guide.md — 问题格式规范
  5. 扫描 references/extensions/ 目录 — 加载所有扩展规则

MANDATORY: 内容验证

在创建任何文件之前,必须按 references/common/content-validation.md 验证内容。

MANDATORY: 问题文件格式

禁止在聊天中直接提问。所有问题必须写入 .md 文件,使用 [Answer]: 标签格式,详见 references/common/question-format-guide.md

MANDATORY: 审计日志

所有用户输入必须原样记录aidlc-docs/audit.md(时间戳用 ISO 8601)。禁止总结或改写用户原话。

MANDATORY: 审批门控

每个阶段完成后,必须等待用户明确批准才能进入下一阶段。使用标准两选项格式:

  • 🔧 Request Changes — 请求修改
  • Continue — 继续下一阶段

MANDATORY: 复选框更新

完成任何计划步骤后,必须立即在该交互中将步骤标记为 [x]


完整工作流

Step 0: 初始化(新项目)

  1. 检查是否存在 aidlc-docs/aidlc-state.md
    • 存在 → 读取状态,从上次阶段恢复
    • 不存在 → 新项目,创建状态文件,继续检测工作区
  2. 检查是否存在现有代码
  3. 显示欢迎消息(新项目只执行一次)
  4. 自动进入下一阶段

Step 1: Workspace Detection (始终执行)

读取 references/inception/workspace-detection.md,按步骤执行。

  • 扫描工作区判断 greenfield / brownfield
  • 更新 aidlc-docs/aidlc-state.md
  • 自动进入下一阶段(无需用户审批)

Step 2: Reverse Engineering (brownfield 条件执行)

读取 references/inception/reverse-engineering.md,按步骤执行。

  • 分析现有代码库,生成 architecture.md, component-inventory.md, api-documentation.md 等
  • 必须等待用户明确批准后才能继续

Step 3: Requirements Analysis (始终执行,自适应深度)

读取 references/inception/requirements-analysis.md,按步骤执行。

深度级别:

  • Minimal — 简单请求,只记录意图

  • Standard — 正常复杂度,收集功能和 NFR

  • Comprehensive — 复杂高风险,详细需求 + 追溯性

  • 始终创建 aidlc-docs/inception/requirements/requirement-verification-questions.md(除非需求极其清晰)

  • 等待用户回答所有 [Answer]: 标签

  • 分析答案中的矛盾/歧义,必要时创建追问文件

  • 生成 aidlc-docs/inception/requirements/requirements.md

  • 必须等待用户明确批准

Step 4: User Stories (条件执行)

读取 references/inception/user-stories.md,按步骤执行。

执行条件(满足任一即执行):

  • 新用户功能、用户体验变化、多用户类型
  • 需要验收标准的复杂业务逻辑
  • 跨团队协作

跳过条件

  • 纯内部重构、明确范围的 bug 修复、技术债务清理

分为两部分:

  • Part 1 Planning — 创建故事计划 + 提问 → 等待批准
  • Part 2 Generation — 生成 stories.md + personas.md → 等待批准

Step 5: Workflow Planning (始终执行)

读取 references/inception/workflow-planning.md,按步骤执行。

  • 加载所有前序上下文
  • 创建 aidlc-docs/inception/plans/execution-plan.md
  • 展示推荐方案(执行哪些阶段,跳过哪些,说明原因)
  • 必须等待用户明确批准

Step 6: Application Design (条件执行)

读取 references/inception/application-design.md,按步骤执行。

  • 必须等待用户明确批准

Step 7: Units Generation (条件执行)

读取 references/inception/units-generation.md,按步骤执行。

  • 必须等待用户明确批准

Step 8: Construction — Per-Unit Loop

对于每个工作单元,依次执行(跳过不适用的):

  1. Functional Design (条件执行) — 读取 references/construction/functional-design.md
  2. NFR Requirements (条件执行) — 读取 references/construction/nfr-requirements.md
  3. NFR Design (条件执行) — 读取 references/construction/nfr-design.md
  4. Infrastructure Design (条件执行) — 读取 references/construction/infrastructure-design.md
  5. Code Generation (始终执行) — 读取 references/construction/code-generation.md
    • Part 1: 创建详细代码生成计划 → 等待批准
    • Part 2: 执行计划生成代码 → 等待批准

每个阶段完成后用标准两选项消息请求批准。

Step 9: Build and Test (始终执行)

读取 references/construction/build-and-test.md,按步骤执行。

  • 生成 build-and-test/ 目录下的构建说明、测试说明文件
  • 必须等待用户明确批准

Step 10: Operations (占位)

读取 references/operations/operations.md。当前为占位阶段,构建测试活动已在 CONSTRUCTION 完成。


文件结构约定

\x3CWORKSPACE-ROOT>/           # ⚠️ 应用代码放这里
├── [项目特定结构]           # 按项目类型(见 code-generation.md)
│
├── aidlc-docs/            # 📄 文档放这里
│   ├── inception/
│   │   ├── plans/
│   │   ├── reverse-engineering/   # brownfield
│   │   ├── requirements/
│   │   ├── user-stories/
│   │   └── application-design/
│   ├── construction/
│   │   ├── plans/
│   │   ├── {unit-name}/
│   │   │   ├── functional-design/
│   │   │   ├── nfr-requirements/
│   │   │   ├── nfr-design/
│   │   │   ├── infrastructure-design/
│   │   │   └── code/             # markdown 摘要
│   │   └── build-and-test/
│   ├── operations/
│   ├── aidlc-state.md
│   └── audit.md

关键规则

  • 应用代码:工作区根目录(禁止放 aidlc-docs/ 内)
  • 文档:aidlc-docs/
  • 审计日志:追加到 audit.md,禁止覆盖

会话恢复(Resume)

读取 references/common/session-continuity.md

  1. 检查 aidlc-state.md,获取当前状态
  2. 加载所有已完成阶段的工件
  3. 显示 "Welcome back" 提示,包含当前阶段 + 下一步
  4. 用户选择继续或回顾

Bug 修复场景

读取 references/inception/requirements-analysis.md(Minimal 深度):

  1. Workspace Detection
  2. Requirements Analysis (Minimal) — 记录 bug 描述和复现路径
  3. Workflow Planning — 通常跳过大部分阶段
  4. Code Generation — 修复 + 测试
  5. Build and Test

对于 bug 修复,User Stories、NFR、Application Design 通常都跳过。


详细规则参考

本 SKILL.md 提供了完整工作流概览。每个阶段的详细步骤、工件格式、问题模板和审批格式,请查阅对应参考文件:

阶段 参考文件
欢迎消息 references/common/welcome-message.md
工作流概览 references/common/process-overview.md
会话恢复 references/common/session-continuity.md
问题格式 references/common/question-format-guide.md
内容验证 references/common/content-validation.md
深度自适应 references/common/depth-levels.md
ASCII 图规范 references/common/ascii-diagram-standards.md
过度自信预防 references/common/overconfidence-prevention.md
错误处理 references/common/error-handling.md
变更管理 references/common/workflow-changes.md
术语表 references/common/terminology.md
工作区检测 references/inception/workspace-detection.md
需求分析 references/inception/requirements-analysis.md
用户故事 references/inception/user-stories.md
工作流规划 references/inception/workflow-planning.md
应用设计 references/inception/application-design.md
单元生成 references/inception/units-generation.md
逆向工程 references/inception/reverse-engineering.md
代码生成 references/construction/code-generation.md
构建测试 references/construction/build-and-test.md
NFR 需求 references/construction/nfr-requirements.md
NFR 设计 references/construction/nfr-design.md
功能设计 references/construction/functional-design.md
基础设施设计 references/construction/infrastructure-design.md
运营 references/operations/operations.md
安全基线(扩展) references/extensions/security/baseline/security-baseline.md
安全使用建议
This skill is internally consistent with a development workflow assistant, but it will read your repository and persist user input and artifacts to aidlc-docs/ (including audit.md) verbatim. Before installing or running: (1) ensure the workspace you grant it access to does not contain secrets, credentials, or sensitive data you don't want recorded; (2) be aware that the skill enforces 'never ask in chat' and records all user answers into files — review those files if you need to redact sensitive content; (3) if you plan to run this on production or large repos, try it first on a safe, trimmed copy to verify behavior; (4) confirm how your agent runtime enforces file access boundaries and whether the skill can read outside intended project paths; and (5) if you rely on external SubAgents or integrations mentioned in docs (e.g., 'SubAgent'), verify what other components will consume the generated state files.
功能分析
Type: OpenClaw Skill Name: ai-dlc Version: 1.1.0 The bundle implements a highly structured AI-Driven Development Life Cycle (AI-DLC) framework. It includes comprehensive instructions for workspace analysis, requirements gathering, architectural design, and code generation. Notable features include a mandatory 'Security Baseline' extension (security-baseline.md) that enforces OWASP-aligned security checks and a strict audit logging mechanism (SKILL.md) to track development decisions. While the agent is granted broad access to scan the workspace and modify files, these capabilities are directly aligned with its stated purpose as a software development orchestrator, and the bundle lacks any indicators of malicious intent, data exfiltration, or unauthorized persistence.
能力评估
Purpose & Capability
The name/description (AI-Driven Development Life Cycle) matches the instructions: workspace detection, requirements, design, code generation, build/test and lots of file creation under aidlc-docs/. The skill asks the agent to read code and workspace artifacts and to create state, audit and question files — behavior that is expected for a development lifecycle assistant.
Instruction Scope
SKILL.md is prescriptive and broad: it mandates reading many reference files bundled with the skill and also mandates scanning/reading the user's workspace and all relevant artifacts (including 'all code files' during code stages). It enforces file-based question/answer flow (never ask in chat) and requires recording all user input verbatim to aidlc-docs/audit.md. These are coherent with an audit-focused dev workflow but give the agent broad discretion over what files to read and what content to persist.
Install Mechanism
No install spec and no code files to run; this is an instruction-only skill. That minimizes installation risk because nothing will be downloaded or executed by the installer.
Credentials
The skill requests no environment variables or external credentials (appropriate), but it explicitly instructs reading the entire workspace and writing audit/state files. That behavior is proportionate for a codebase analysis tool, yet it presents a privacy/secret-exposure risk: any secrets or credentials present in the workspace or pasted by users will be recorded verbatim into audit.md and other artifacts. The skill does not request or justify access to unrelated external credentials, which is good.
Persistence & Privilege
The skill does not request 'always: true' and has no install-time persistence. It does instruct creating and updating workspace files (aidlc-docs/, aidlc-state.md, audit.md) which is normal for a development assistant and limited to the project workspace rather than system-wide configuration.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install ai-dlc
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /ai-dlc 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.1.0
Add mandatory changelog step to CONSTRUCTION phase: update aidlc-state.md after each merge
v1.0.0
Initial release
元数据
Slug ai-dlc
版本 1.1.0
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 2
常见问题

AI-DLC 是什么?

AI-Driven Development Life Cycle (AI-DLC) adaptive workflow for software development. Use when: starting a new project, new feature, bug fix, refactoring, mi... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 221 次。

如何安装 AI-DLC?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install ai-dlc」即可一键安装,无需额外配置。

AI-DLC 是免费的吗?

是的,AI-DLC 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

AI-DLC 支持哪些平台?

AI-DLC 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 AI-DLC?

由 BingWang(@sydpz)开发并维护,当前版本 v1.1.0。

💬 留言讨论