来源：https://academy.claude.com/courses/ai-native-sdlc-playbook/introduction 读这篇之前：可直接读；课程页 https://academy.claude.com/courses/ai-native-sdlc-playbook 说明给谁看 不确定：ZH 是 Clint 译官方英文课体，不是 Anthropic 中文。locale pack 里没有本课正文。
本课词汇
| English | 中文 | 怎么记 |
|---|---|---|
| SDLC | 软件开发生命周期 | idea → 生产的那条链，课里拆成六段 |
| bottleneck | 瓶颈 | 最慢的一截。写代码不再是 |
| agentic | 能自己动手的 AI | 不只聊天，会改文件、跑命令 |
| play | 一条打法 | 可拆开上的改法，不是必须一次全上 |
| artifact | 工件 | 写进 Git 的那份：intent/spec/plan、diff、PR |
| intent.md | 意图文件 | 要什么、为什么、边界。链的第一环 |
| spec.md | 需求和设计说明 | 人审、agent 写，工程拿它做计划 |
| plan.md | 实现计划 | 改哪些文件、顺序、怎么证明 |
| CLAUDE.md | 给 Claude 的说明书 | 新同事第一天要知道的 |
| skill | 技能包 | 政策写成 agent 会加载的说明 |
| hook | 钩子 | Claude 动手前允许/询问/拦住 |
| control band | 控制带 | 指标允许的波动范围，越界就触发 |
| governance | 治理 | 谁拍板、怎么留下证据 |
| handoff | 交接 | 传统靠人传文档；这里靠提交工件触发下一阶段 |
对照正文
开头怎么说 / The opening
EN Organizations have started using AI to write code at a speed unthinkable one year ago, yet the processes around the code haven’t changed at the same pace.
ZH 组织已经开始用 AI 写代码，速度一年前还难以想象，但代码周围的流程没有同步变快。
EN Many engineering teams still have the same approval gates, reviews, handoffs, and policies, stalling productivity gains made by using agentic coding solutions like Claude Code.
ZH 很多工程团队还是同一套审批闸门、审查、交接和制度，把 Claude Code 这类能自己写代码的方案带来的效率，卡在原地。
传统 SDLC / The traditional SDLC
EN The software development lifecycle (SDLC) is the process that takes software from idea to production. Most organizations run some version of the same six stages, covering planning, design, building, testing, deploying, and maintaining software. Traditionally, each stage is a discrete phase owned by a different role. Product managers write requirements, technical architects turn them into designs, engineers build the designs, QA teams at regulated enterprises verify the software, release teams ship it, and operations monitors what is running. Work moves between the phases through documents, tickets, and sign-offs.
ZH 软件开发生命周期（SDLC）是软件从想法走到生产的过程。多数组织跑的是同一套六段的某个变体：计划、设计、构建、测试、发布、维护。传统上每段是独立阶段，由不同角色负责。产品经理写需求，架构师做成设计，工程师按设计实现，受监管企业里 QA 来验证，发布团队上线，运维盯着在跑的东西。阶段之间靠文档、工单和签字往前传。
EN The traditional software development lifecycle (SDLC) is process-heavy to ensure accountability and control at each step. However, the traditional SDLC was designed to maximize efficiency in an era where the most time-consuming and expensive stage was writing and implementing code, which is no longer the case. Product requirements documents (PRDs), estimation rituals, and product security reviews all existed to force alignment during what could be weeks, months, or quarters of development work.
ZH 传统 SDLC 流程重，是为了每一步都有人负责、有控制。但它是按「写代码最耗时、最贵」的年代来最大化效率的，现在已经不是这样。PRD、估点仪式、产品安全审查，都是为了在可能长达数周、数月甚至几个季度的开发里，把人强行对齐。
EN The traditional SDLC also features controls that assume every step is performed by humans. The organizations generating the most value have rebuilt their process around what agentic AI can now do, while ensuring that humans stay in the loop. In this guide, we walk through several of our Applied AI team’s best practices for integrating Claude internally across each stage of the SDLC to accelerate development and make processes run faster, inspired by working with our customers.
ZH 传统 SDLC 的控制还默认每一步都是人做的。拿到最多价值的组织，已经按 agentic AI 现在能做的事把流程重建了，同时保证人还在环里。这份指南走的是 Anthropic Applied AI 团队内部把 Claude 嵌进 SDLC 各阶段的若干实践，也受他们和客户一起干活的启发，目的是加快开发、让流程跑起来。
代码不再是瓶颈时 / When code is no longer the bottleneck
EN When code is no longer the bottleneck and the build phase runs faster than the traditional SDLC allows for, three things become true:
ZH 当代码不再是瓶颈、构建阶段跑得比传统 SDLC 允许的更快时，三件事会变成真的：
EN 1. The bottleneck moves to the stages on either side of the build phase: mainly plan, review/test, and deploy, which still run at human speed. 2. The controls stop matching reality and become intractable. Reviewing each line by hand made sense when a person had written it, but it can’t keep up once agents write most of the diff. 3. Governance costs increase because exceptions still route through meetings and committees that meet weekly or monthly.
ZH 1. 瓶颈挪到构建两侧：主要是计划、审查/测试、发布，这些还在按人的速度跑。 2. 控制对不上现实，变得搞不定。一行行人手审，在人写的时候说得通；agent 写了大部分 diff 之后就跟不上。 3. 治理成本上升，因为例外还在走每周或每月开一次的会和委员会。
图示 Two rows of software lifecycle stages. Before agents: Plan, Design, a long Build block, Test, Deploy, Maintain, every stage at human speed. After agents: the same stages with Build collapsed to a sliver and the freed width marked ‘cycle time reclaimed’; Plan and Design are labelled requirements, Test is labelled review, Deploy is labelled release.
原课配图文件未包含在迁入包中，见图注与上方来源链接。
EN The bottleneck moved: build collapses to agent speed while the stages around it still run at human speed.
ZH 瓶颈挪了：构建塌成 agent 速度，两侧阶段还在按人的速度跑。
EN Let’s use a security bottleneck as an example. Security teams are sized for human output, so when agents multiply code output, either the review queue builds or code ships under-reviewed. A regulated organization can’t accept either outcome, so its security and policy checks have to keep pace with the agents.
ZH 拿安全瓶颈当例子。安全团队是按人的产出配的，agent 把代码量乘上去之后，要么审查队列堆起来，要么代码没审够就上船。受监管组织两头都不能接受，所以安全和制度检查必须跟得上 agent。
EN To better realize the productivity gains of and secure agentic AI, the traditional SDLC lifecycle requires the same level of transformation as the implementation phase has undergone.
ZH 要把 agentic AI 的效率吃到、又要把它管住，传统 SDLC 需要跟实现阶段已经经历过的同等级改造。
什么是 AI-native SDLC？ / What is an AI-native SDLC?
EN The AI-native SDLC is a reimagined process that combines the old control objectives with new enforcement. Instead of a linear flow, the process becomes a loop, and AI is embedded at each point. The AI-native SDLC promotes automated handover and triggering of subsequent plays, helping to address the manual and clunky nature of handoff between the phases of the traditional SDLC.
ZH AI-native SDLC 是一套重想过的流程：控制目标还是原来那些，执行方式换成新的。不再是一条直线，而是一个环，每个点都嵌着 AI。它推动自动交接、自动触发下一招，用来对付传统 SDLC 阶段之间那种靠人手、很笨的交接。
图示 Left: the traditional lifecycle as a straight line, Plan to Design to Build to Test to Deploy to Maintain. Right: the same six stages arranged as a continuous loop around Claude, with arrows running clockwise from Plan back to Plan.
原课配图文件未包含在迁入包中，见图注与上方来源链接。
EN The traditional linear SDLC (left) and the AI-native continuous loop (right), with humans positioned above the loop instigating, directing, and governing.
ZH 左边是传统直线 SDLC，右边是绕着 Claude 转的连续环；人在环上方发起、指路、治理。
两边差在哪 / The shifts
EN The table below highlights the ends of the spectrum between traditional SDLC and AI-native SDLC, supported by Claude. Most organizations sit somewhere between the two columns.
ZH 下表标出传统 SDLC 和 Claude 支撑的 AI-native SDLC 两端。多数组织坐在两列中间某处。
EN
| Stage | Traditional SDLC | AI-native SDLC |
|---|---|---|
| Plan | Requirements gathered by committee, distilled through workshops and sign-offs, written up by hand | Claude synthesizes pain points straight from the sources and captures them within intent.md, which is human readable and machine actionable |
| Design | Spec written by analysts, parsed by designers | Requirements and design compressed into one working session with an agent, guided by standards encoded as skills, versioned in Git |
| Build | Tests and code are handwritten, and documentation is written after the main development happens | Tests and code are generated by AI, and institutional knowledge is maintained as versioned machine-readable CLAUDE.md files and skills |
| Test | QA gates at stage boundaries | Continuous evals woven through implementation |
| Deploy | Humans review every line of code, and governance occurs in review cycles, often inconsistently | Layers of agentic review with human review reserved for regulated and critical code. Governance is enforced as the AI acts, with hooks as approval gates |
| Maintain | Humans watch production for bugs | Agents monitor live deployments. Any breached control band is diagnosed and written back into the loop as a new intent.md |
ZH
| 阶段 | 传统 SDLC | AI-native SDLC |
|---|---|---|
| Plan | 委员会收需求，工作坊和签字蒸馏，人手写成文 | Claude 直接从源头归纳痛点，写进 intent.md：人能读，机器能执行 |
| Design | 分析师写 spec，设计师再解析一遍 | 需求和设计压进跟 agent 的一次工作会话，标准写成 skills，版本在 Git |
| Build | 测试和代码人手写，文档主开发之后再补 | 测试和代码由 AI 生成；组织知识维护成带版本、机器可读的 CLAUDE.md 和 skills |
| Test | 阶段边界上的 QA 闸门 | 持续 eval 织进实现过程 |
| Deploy | 人审每一行；治理靠审查周期，常常不一致 | 一层层 agent 审查，人审留给受监管和关键代码。治理在 AI 动手时用 hooks 当审批闸门 |
| Maintain | 人盯生产找 bug | Agent 盯线上。控制带一越界，诊断后写回环里，变成新的 intent.md |
EN What ties the AI-native column together is the committed artifact. Each stage ends by writing one to version control (including intent.md, spec.md, plan.md, the diff and its tests, the PR with its review findings, and the incident record), and the next stage begins by reading it. For the early stages, .md files are the predominant artifact because a product owner and an agent can both read and act on the same file. From Build onward, the artifact is code and its records. The chain of commits is also the audit trail: who asked for what, what the agent produced, and who approved it.
ZH 把 AI-native 那一列串起来的，是提交进版本库的工件。每阶段结束都往版本库写一份（包括 intent.md、spec.md、plan.md、diff 和测试、带审查结论的 PR、事故记录），下一阶段从读它开始。前面几段以 .md 为主，因为产品负责人和 agent 都能读、都能对同一份文件动手。从 Build 起，工件变成代码和它的记录。提交链也是审计链：谁要了什么、agent 产出了什么、谁批的。
EN Humans remain accountable for every decision that requires judgment. In the agentic SDLC world, the human attention shifts along with the artifacts that must be reviewed.
ZH 凡是需要判断的决定，责任仍在人。在 agentic SDLC 里，人的注意力跟着必须审的工件一起挪。
这些打法怎么用 / How the plays work
EN The plays are the core of the playbook and are grouped into six non-linear stages (Plan, Design, Build, Test, Deploy, Maintain), which together cover the complete lifecycle.
ZH 打法是这本手册的核心，按六个非线阶段分组（Plan、Design、Build、Test、Deploy、Maintain），合起来覆盖完整生命周期。
EN Each play covers:
- What changes
- Getting started
- Concrete steps for implementation
- Governance considerations
- How you measure whether it worked
ZH 每条打法都覆盖：
- 什么变了
- 怎么起步
- 可执行的步骤
- 治理上要注意什么
- 怎么衡量有没有用
EN The plays are modular, and organizations may choose to prioritize transforming different stages at different times based on their unique needs. Each play names its dependencies under “Prerequisites,” which the dependency graph further illustrates.
ZH 打法是模块化的，组织可以按自己的需要，不同时间优先改不同阶段。每条打法在 Prerequisites 里点名依赖，依赖图会再画一遍。
EN A stage ends by committing an artifact with the commit initiating the next stage. An accepted intent.md triggers the requirements and design pass, an approved spec.md triggers plan mode, a merged PR triggers the pipeline, and a breached control band in production writes the next intent.md, and so the loop continues.
ZH 一阶段结束于提交一份工件，这次提交启动下一阶段。被接受的 intent.md 触发需求和设计那一趟；批过的 spec.md 触发 plan mode；合进的 PR 触发流水线；生产里控制带越界就写下一份 intent.md。环就这样转。
EN At first, you prompt each step by hand, with the end state being a loop in which each accepted artifact fires the next gate. Human attention concentrates at the gates, reviewing what the agent flagged rather than starting each stage from scratch.
ZH 一开始每步还是你亲手 prompt。终点是一个环：每份被接受的工件打下一扇闸。人的注意力集中在闸门上，审 agent 标出来的东西，而不是每阶段从零开始。
图示 Dependency graph of the twelve plays in five rows. Row 1, no prerequisites: Capture intent, CLAUDE.md, Skills, Feedback loop, Hooks, Plan mode. Row 2: Subagents (needs CLAUDE.md; the feedback loop helps), Evals (needs CLAUDE.md and the feedback loop). Row 3: Requirements and design (needs Capture intent and Skills), PR review (needs Evals and Subagents; Skills help). Row 4: CI/CD (needs PR review and Hooks). Row 5: Closing the loop (needs CI/CD and Capture intent). CLAUDE.md helps Skills but is not required.
原课配图文件未包含在迁入包中，见图注与上方来源链接。
EN The play dependency graph. Plays in the top row have no prerequisites; a solid arrow points to the plays that build on it, a dotted arrow means it helps but is not required.
ZH 打法依赖图。最上排没有先决条件；实线箭头指向建立在它上面的打法，虚线表示有帮助但不是必须。
读完能记住的三句话
EN Once agents write most of the code, the bottleneck is plan, review/test, and deploy — still at human speed. ZH agent 写了大部分代码之后，瓶颈是计划、审查/测试、发布——还在按人的速度。
EN Each stage ends by committing an artifact the next stage can read: intent.md → spec.md → plan.md → diff/tests → PR → production → a new intent.md.
ZH 每阶段结束都提交下一阶段能读的工件：intent.md → spec.md → plan.md → diff/测试 → PR → 生产 → 新的 intent.md。
EN Humans stay accountable at the gates. Plays are modular; you do not have to transform every stage at once. ZH 闸门上的责任仍在人。打法可拆，不必一次改完全部阶段。