Claude Academy · Learn · Roles

Engineering — Learn with Claude

工程 —— 和 Claude 一起学习

Source / 原文:https://claude.ai/learning/roles/engineering

English paragraph first, Chinese translation below. 每段先英文,后中文译文。Copied 2026-11-22.

💡 点击任意英文单词:查看音标、中文释义,并可朗读单词 / 整句 / 整段。点击段落末尾的 🔊 朗读整段。词典:ECDICT(开源英汉词典)。

claude.ai/learning · Engineering

Try a prompt

试试这些提示词

Each prompt opens as a draft you can edit before sending.

每个提示词都会以草稿形式打开,发送前可以自行修改。

Debug an issue

调试问题

Diagnose a bug in your code and identify the root-cause.

诊断代码中的 bug,并找出根本原因。

I’m stuck on a bug. Below I’ll paste the relevant code, the error or wrong output, and what I expected. If the placeholder below is still there, ask me for it before doing anything else. Either way, ask me the language and version and how I’m running it if I haven’t said. Then: restate the problem in one sentence, walk through the most likely causes in order, point to the exact lines involved, and propose the smallest fix that addresses the root cause rather than the symptom. Give me a test or reproduction step that would fail before and pass after. If you can’t be sure without more information, say what you’d need. [paste code and error here]
我被一个 bug 卡住了。下面我会贴出相关代码、报错信息或错误的输出,以及我原本期望的结果。如果下方的占位文字还在,请先向我索要这些内容,再做其他事。无论如何,如果我没说明,请先问我用的是什么编程语言和版本、以及我是怎么运行它的。然后:用一句话复述问题,按可能性从高到低逐一分析最可能的原因,指出涉及的具体代码行,并提出能解决根本原因(而不是只掩盖症状)的最小修改方案。给我一个测试或复现步骤:修复前会失败,修复后会通过。如果没有更多信息就无法确定,请告诉我你还需要什么。 [在此粘贴代码和报错]

Review code changes

审查代码改动

Conduct a code review that identifies bugs, security issues, and nits.

进行一次代码审查,找出 bug、安全问题和细节瑕疵。

Please review this change as a careful senior engineer would. I’ll paste the diff or the changed file below. If the placeholder below is still there, ask me for it before doing anything else. Either way, ask me what the change is supposed to do and any conventions you should hold it to, rather than assuming them. Organize feedback as: bugs and correctness issues, security or data-handling concerns, performance, then style nits, most important first, each with the line it refers to and a concrete suggested change. Tell me what you’d want tested. Don’t pad the review; if it’s fine, say so and just list residual risks. [paste diff here]
请像一位细心的资深工程师那样审查这次改动。我会在下方贴出 diff(改动对比)或修改后的文件。如果下方的占位文字还在,请先向我索要,再做其他事。无论如何,请先问我这次改动要实现什么、以及应该遵守哪些约定,而不是自己假设。反馈请按以下顺序组织:bug 和正确性问题、安全或数据处理方面的隐患、性能,最后是代码风格的小问题;最重要的放在最前面,每条都注明对应的代码行,并给出具体的修改建议。告诉我你认为应该测试哪些内容。不要为了显得全面而凑数;如果改动没问题就直说,只列出剩余的风险即可。 [在此粘贴 diff]

Diagnose an infra failure

诊断基础设施故障

Analyze a deploy, CI job, permission, cluster, or network path that has broken.

分析出了问题的部署、CI 任务、权限、集群或网络链路。

Something in my infrastructure isn’t working: a deploy, a CI job, a permissions error, a cluster or networking problem. I’ll describe what I’m seeing and paste the error output below. If the placeholder below is still there, ask me for it before doing anything else. Either way, ask me what changed recently and which platform and tooling I’m using. Then help me narrow it down: the most likely causes ranked, the exact commands or checks to run to confirm or rule out each one (read-only first), and what to do once we know. Don’t tell me to run anything destructive without saying what it affects. Work through it with me until we have a fix or a clear next step. [paste error output here]
我的基础设施里有东西不正常:可能是部署、CI 任务、权限报错、集群或网络问题。我会描述看到的现象,并在下方贴出报错输出。如果下方的占位文字还在,请先向我索要,再做其他事。无论如何,请先问我最近改动了什么、用的是哪个平台和哪些工具。然后帮我逐步缩小范围:按可能性排序列出原因,给出用来确认或排除每个原因的具体命令或检查步骤(先用只读、不改动任何东西的方式),以及查明原因后该怎么处理。不要在没说明影响范围的情况下让我执行任何有破坏性的操作。和我一起排查,直到找到修复办法或明确的下一步。 [在此粘贴报错输出]

Write a design doc from the source code

根据源代码写设计文档

Create a design doc that’s generated from actual code that can be shared with others.

基于真实代码生成一份可以分享给他人的设计文档。

I need a design doc grounded in our actual code, not a generic template. First ask me to paste the relevant source files, what the change is, and whether we have an RFC template or past design docs to match. Don’t start without the code or a description detailed enough to stand in. Then read the relevant parts, note the constraints and prior art you find, and draft the doc as a markdown file: problem, goals and non-goals, two or three approaches with real trade-offs, your recommendation, risks, and a rollout plan. Cite the files that drove each constraint. Where you had to assume something about how the system works, say so and ask me to confirm it before I share the doc.
我需要一份立足于我们真实代码的设计文档,而不是套用通用模板。请先让我贴出相关的源代码文件、说明这次要做什么改动,并问我是否有 RFC 模板或以往的设计文档需要参照。在拿到代码、或足够详细可以替代代码的描述之前,不要开始写。然后阅读相关部分,记下你发现的限制条件和已有的先例,把文档起草成一个 markdown 文件,包含:问题、目标与非目标、两到三种方案及其真实的利弊权衡、你的推荐、风险,以及上线计划。为每条限制条件注明是哪些文件导致的。凡是你不得不对系统运作方式做出假设的地方,都要说明,并在我分享文档之前请我确认。

Reconstruct the incident

还原事故经过

Generate a blameless postmortem from an incident thread, logs, and the deploy history.

根据事故讨论串、日志和部署历史,生成一份不追究个人责任的事后复盘报告。

I need a postmortem written from what actually happened, not from memory. Start by asking me to point you at the sources: the incident channel or thread if I’ve connected our chat tool, a folder with exported logs, alerts, or notes, and the deploy or change history, or I’ll paste what I have. Don’t start until you have at least a timeline source. Then reconstruct the sequence of events with timestamps, write the customer impact summary, identify the root cause and contributing factors, and draft the postmortem in our template if I give you one, otherwise a standard one. Keep it blameless, name systems not people, and give every action item an owner placeholder and a due date. Quote the source for each timeline entry and mark anything you inferred.
我需要一份根据实际发生的情况写成的事后复盘报告,而不是凭记忆写。请先让我告诉你资料来源:如果我已经接入了聊天工具,就是事故频道或讨论串;一个存放导出日志、告警或笔记的文件夹;以及部署或变更历史——或者我直接把手头的内容贴给你。至少拿到一份能反映时间线的资料之前,不要开始。然后带上时间戳还原事件经过,写出对客户影响的概述,找出根本原因和促成因素,如果我提供了模板就按模板起草复盘报告,否则用标准格式。保持"不追责"的原则:点名系统而不是点名人;每个待办事项都要留一个负责人占位符和截止日期。时间线上的每一条都要引用出处,凡是你推断出来的内容都要标注出来。

Write tests for a module

为模块编写测试

Create a test file for a module in your repo that covers the real behavior, edge cases, and gaps in existing coverage.

为代码库中的某个模块创建测试文件,覆盖真实行为、边界情况,以及现有测试没覆盖到的地方。

I want tests written for part of my codebase, matched to how we actually test. First ask me to paste the module or file to cover and an existing test, and check how existing tests are laid out (framework, naming, fixtures) so you match them; if there are none yet, ask which framework to use. Then read the code and write a test file in the same style: the main paths, edge cases, and error handling, each test named for the behavior it checks, with no tests that only restate the implementation. Run the tests if you can and tell me what passed and failed; if you can’t run them, say so. Finish with a short list of behaviors you couldn’t test and why.
我想为代码库的一部分编写测试,并且要符合我们实际的测试习惯。请先让我贴出要覆盖的模块或文件,以及一个现有的测试,并查看现有测试是怎么组织的(测试框架、命名方式、测试数据准备),以便保持一致;如果还没有任何测试,就问我该用哪个框架。然后阅读代码,用同样的风格写一个测试文件:覆盖主要流程、边界情况和错误处理,每个测试都以它所检验的行为来命名,不要写那种只是把实现代码重复一遍的测试。如果你能运行测试,就运行并告诉我哪些通过、哪些失败;如果运行不了,就直说。最后简短列出你没法测试的行为以及原因。
Use case · 用例 · 10 min

Build an 'Ask the Company' agent

打造一个“问问公司”智能助手

Plain-English answers over your wiki, code, and warehouse.

基于你的 wiki、代码和数据仓库,用大白话给出答案。

Source / 原文:https://academy.claude.com/use-cases/ask-the-company

Claude Cowork is now just Claude. What you learn here still applies — simply ask Claude.

Claude Cowork 现在就是 Claude。您在这里学到的内容依然适用,直接询问 Claude 即可。

Film: the user asks Claude one question spanning a wiki, code repository, and data warehouse, then watches it cite sources, name an owner, and save the approach as a shared skill.

视频:用户向 Claude 提出一个横跨 wiki、代码仓库和数据仓库的问题,然后看着它引用来源、指出负责人,并把这套做法保存为一个可共享的技能(skill)。

Demo walkthrough

演示流程

Your team keeps asking who owns what, and the answers are scattered.

团队里总有人问“这个归谁管”,而答案散落在各处。

Ask Claude one real question, with each source cited and an owner named.

向 Claude 提一个真实的问题,并要求它为每个来源标注出处、指出负责人。

Claude checks each place in turn: your reference files, the wiki, the code and the warehouse.

Claude 依次查找每个地方:你的参考文件、wiki、代码和数据仓库。

You get the answer and the owner, with each source linked.

你会得到答案和负责人,并且每个来源都附有链接。

The answer names the owner and links every source it used.

答案会写明负责人,并链接到它用到的每一个来源。

Any source opens to the exact line behind the answer.

点开任何一个来源,都能直接定位到支撑这个答案的那一行。

When the answer is right, ask Claude to save it as a skill the whole team can run.

答案确认无误后,让 Claude 把它保存为全团队都能使用的技能。

Anyone on the team can ask who owns something and see the proof.

团队里任何人都能问某样东西归谁管,并看到依据。

You: Search our wiki, code and warehouse, cite the source for each part, and if you're not sure, tell me who to ask. Who owns the orders table?

你:搜索我们的 wiki、代码和数据仓库,每一部分都注明来源;如果不确定,告诉我该去问谁。orders 表归谁负责?

Claude: I traced the orders table through your reference files, the wiki, the code and the warehouse, and the five sources agree. Owner: Commerce Platform. The orders table is written by orders-service; CODEOWNERS and the service catalog both name that team. (Sources: CODEOWNERS · meridian/orders-service; service-catalog.yaml)

Claude:我在你的参考文件、wiki、代码和数据仓库中追踪了 orders 表,五个来源结论一致。负责人:Commerce Platform 团队。orders 表由 orders-service 写入;CODEOWNERS 和服务目录都写明是这个团队。(来源:CODEOWNERS · meridian/orders-service;service-catalog.yaml)

Claude: The Confluence schema page is two columns behind the warehouse. (Sources: Order data model (v3) · Confluence; COMMERCE.PUBLIC.ORDERS · Snowflake)

Claude:Confluence 上的表结构页面比数据仓库里的实际表少了两列,已经过时。(来源:Order data model (v3) · Confluence;COMMERCE.PUBLIC.ORDERS · Snowflake)

Claude: For write access or a new column, ask Priya Raman in #commerce-platform, listed as data owner. (Source: team-directory.md)

Claude:如需写入权限或新增一列,请在 #commerce-platform 频道联系 Priya Raman,她登记为数据负责人。(来源:team-directory.md)

Connectors: Confluence, GitHub, Snowflake

连接器:Confluence、GitHub、Snowflake

Set up

准备工作

Connect your tools

连接你的工具

Claude is more powerful when it works directly with your systems. You control permissions and access.

当 Claude 能直接对接你的系统时,它的能力会更强。权限和访问范围由你掌控。

Add a connector for each tool you want Claude to use.

为每一个你希望 Claude 使用的工具添加一个连接器(connector)。

Confluence

Confluence

Search the wiki and runbooks for the documented answer first. Confluence access comes through the Atlassian connector (Jira and Confluence).

先在 wiki 和运维手册(runbook)里查找有文档记录的答案。Confluence 的访问通过 Atlassian 连接器(包含 Jira 和 Confluence)实现。

GitHub

GitHub

Read code, READMEs, and CODEOWNERS to answer "how does X work" and "who owns X."

读取代码、README 和 CODEOWNERS,用来回答“X 是怎么运作的”和“X 归谁负责”。

Snowflake (optional)

Snowflake(可选)

Query the warehouse for "how many" and "why did the metric move" questions.

查询数据仓库,回答“有多少”以及“这个指标为什么变了”这类问题。

Want to try this task before setting anything up? Drag in your files, or give Claude access to a folder in the Claude desktop app. Then start with the prompt.

想在配置任何东西之前先试试这个任务?把文件拖进来,或者在 Claude 桌面应用中授权 Claude 访问一个文件夹,然后从下面的提示词开始。

Set your working folder

设置工作文件夹

Drag the files that anchor the agent (your service catalog export, the team directory, the glossary, the "where to find things" onboarding doc) into one folder and give Claude access to it in the Claude desktop app. These become the routing table the agent consults before searching anywhere else. Create a project from that folder so the catalog and ground rules stay attached as the agent evolves.

把作为这个助手“根基”的文件(导出的服务目录、团队通讯录、术语表、新人入职用的“东西都在哪儿找”文档)拖进同一个文件夹,并在 Claude 桌面应用中授权 Claude 访问它。这些文件就成了助手的“路由表”——它在去别处搜索之前会先查阅这里。再用这个文件夹创建一个项目(Project),这样随着助手不断完善,服务目录和基本规则会一直跟着它。

The prompt

提示词

Copy this into Claude

复制到 Claude 中

Build an internal "ask anything" agent for engineers. When someone asks "how do I get a staging API key," "what owns the orders table," or "why did deploys slow down last month," find the answer across our systems, cite the source, and tell them who to ask if you're not sure. Get one good answer first, then turn it into a shared skill.
为工程师打造一个内部的“有问必答”助手。当有人问“怎么拿到 staging 环境的 API key”“orders 表归谁负责”或“上个月部署为什么变慢了”时,在我们的各个系统中找到答案、注明来源;如果你不确定,就告诉他们该去问谁。先拿出一个好答案,再把它变成一个共享技能。

Why this works

为什么这样有效

Get a better draft

让初稿更好

Make Claude work for you

让 Claude 按你的方式工作

Once Claude answers your three questions well, tell it what to call the skill. Layer in which spaces and repos to search first, the "I'm not sure" threshold, the tone, and the systems that are off-limits. A few minutes of conversation and the skill speaks your stack from then on.

等 Claude 把你的三个问题都答好了,告诉它这个技能叫什么名字。再逐步补充:优先搜索哪些空间和代码仓库、到什么程度算“我不确定”、用什么语气,以及哪些系统不能碰。聊上几分钟,这个技能从此就能用你们团队自己的技术体系来回答问题。

Make what we've done in this task so far into a skill called /ask-the-company.
把我们在这个任务里到目前为止做的事情,做成一个名为 /ask-the-company 的技能。

Review the skill, and save it if Claude hasn't already.

检查一下这个技能;如果 Claude 还没保存,就把它保存下来。

Tip: tell Claude to edit the skill for you.

小提示:可以直接让 Claude 帮你修改这个技能。

Make it repeatable

让它可以重复运行

Share it as an artifact

以 artifact 形式分享

Once the agent answers your test questions well, publish it as an artifact so anyone in the org has one link to ask.

等助手把你的测试问题都答好后,把它发布为一个 artifact(可分享的页面),这样组织里的任何人都能通过同一个链接来提问。

Log every question it couldn't answer

记录它答不上来的每个问题

The agent gets better when you know where it falls short. Ask Claude to set up a scheduled task or open Scheduled in the sidebar, and a weekly run writes the unanswered questions to your folder so the platform team can fill the gaps.

知道助手哪里不足,它才能越来越好。让 Claude 设置一个定时任务,或者在侧边栏打开 Scheduled(定时),每周运行一次,把没答上来的问题写进你的文件夹,方便平台团队补齐缺口。

Set up a scheduled task for every Friday at 4pm: list the questions /ask-the-company couldn't confidently answer this week, group them by topic, and write them to Platform/ask-company/gaps-<week>.md with a suggested owner for each.
设置一个每周五下午 4 点运行的定时任务:列出本周 /ask-the-company 没能有把握回答的问题,按主题分组,写入 Platform/ask-company/gaps-<week>.md,并为每个问题建议一位负责人。

Scheduled task: Ask-company gap report

定时任务:Ask-company gap report(问答缺口报告)

Weekly, lists the questions /ask-the-company couldn't answer, groups them by topic, and writes a gap report with suggested owners.

每周列出 /ask-the-company 答不上来的问题,按主题分组,并写出一份附带建议负责人的缺口报告。

Active · Friday at 4:00pm

已启用 · 每周五下午 4:00

To use a folder on your computer during a scheduled task, keep your computer on and awake with the Claude desktop app open so Claude can access it.

如果定时任务要用到你电脑上的文件夹,请保持电脑开机、不休眠,并打开 Claude 桌面应用,这样 Claude 才能访问它。

Share with your teammates

分享给你的同事

Your /ask-the-company skill now carries your sources, your routing table, and your confidence rules. Share it workspace-wide so every engineer asks the company the same way, and the answer to "how do I…" stops depending on who's in the channel.

你的 /ask-the-company 技能现在已经包含了你的信息来源、路由表和把握度规则。把它分享给整个工作区,让每个工程师都用同样的方式“问公司”,这样“我该怎么……”的答案就不再取决于当时谁在频道里。

Share the skill

分享技能

Open /ask-the-company in your skills and select Publish to org, if your admin allows it. Everyone gets the same front door to wiki, code, and warehouse, with the platform team keeping it tuned.

在你的技能列表中打开 /ask-the-company,选择 Publish to org(发布到组织),前提是管理员允许。这样每个人都有了通往 wiki、代码和数据仓库的同一个入口,由平台团队负责持续调优。

What changes for the org

这对组织带来什么变化

Internal questions get a sourced answer from across your wiki, code, and warehouse, with an owner named when the agent isn't certain. Engineers check one response instead of searching each system.

内部问题能从 wiki、代码和数据仓库中得到有出处的答案;助手不确定时,会指出负责人。工程师只需看一条回复,而不用挨个系统去搜。

You did this for everyday engineering questions. The same approach covers on-call runbook lookup, data-catalog search, and new-hire onboarding — each one becomes a skill your team runs the same way.

你这次是针对日常工程问题做的。同样的方法也适用于值班手册查询、数据目录搜索和新员工入职——每一项都可以变成团队用同一种方式运行的技能。

Use case · 用例 · 10 min

Find the cloud cost anomaly

找出云成本异常

Where the bill jumped, why, and who owns the fix.

账单在哪里暴涨、为什么涨、该由谁来处理。

Source / 原文:https://academy.claude.com/use-cases/cloud-cost-anomalies

Claude Cowork is now just Claude. What you learn here still applies — simply ask Claude.

Claude Cowork 现在就是 Claude。您在这里学到的内容依然适用,直接询问 Claude 即可。

Film: the user asks Claude to compare a month's cloud billing export to trend by service and team, flagging anomalies with a cause and owner. Claude checks a warehouse and monitoring, confirms one spike was planned, then returns a chart and ranked list with owners and next steps.

视频:用户让 Claude 把一个月的云账单导出数据按服务和团队与历史趋势对比,标出异常项并给出原因和负责人。Claude 查询了数据仓库和监控系统,确认其中一次激增是计划内的,然后返回一张图表和一份按顺序排列的清单,附有负责人和下一步行动。

Demo walkthrough

演示流程

Your cloud bill jumped and you need to know where and why.

你的云账单突然暴涨,你需要知道涨在哪里、为什么涨。

Set a baseline and a threshold, and ask for a cause and an owner each time.

设定一个对比基准和一个阈值,并要求每一项都给出原因和负责人。

Claude compares the export with past months in the warehouse, then checks monitoring for what changed.

Claude 把这次的导出数据与数据仓库里过去几个月的数据对比,然后去监控系统里查看发生了什么变化。

When something is yours to decide, Claude stops and asks.

遇到需要你来拍板的事,Claude 会停下来问你。

Claude says what jumped and why, and gives you the chart and ranked list.

Claude 告诉你哪些费用涨了、为什么涨,并给你一张图表和一份排好序的清单。

Each item names its owner and a next step.

每一项都写明了负责人和下一步行动。

The chart shows what crossed your threshold.

图表展示了哪些项目超过了你设定的阈值。

Schedule it, and the report is in the folder the Monday after each month's export arrives.

把它设为定时任务,每个月的账单导出到位后的那个周一,报告就会出现在文件夹里。

You know what drove the bill up and who is looking into each part.

你清楚是什么推高了账单,也知道每一部分由谁在跟进。

You: Compare this month's AWS costs to the trailing three-month average by service and tag:team. Flag any line more than 20% over trend, trace each to the workload or change behind it, and give every item an owner and recommended action.

你:按服务和 tag:team(团队标签),把本月的 AWS 费用与过去三个月的平均值对比。标出所有超出趋势 20% 以上的条目,追查每一项背后对应的工作负载或变更,并为每一项指定负责人和建议的处理措施。

Claude asks: Was the jump in S3 backfill spend planned? Options: "Yes, planned, mark it expected" / "No, count it as an anomaly"

Claude 提问:S3 回填(backfill)费用的上涨是计划内的吗?选项:“是,计划内的,标记为预期内” / “不是,算作异常”

Claude: I compared April's AWS bill to the three-month trend by service and team: spend is 8.9% over, and four lines past 20% make up $11,900 of the $15,200 increase. Each has a cause, an owner and a next step.

Claude:我按服务和团队把 4 月的 AWS 账单与三个月趋势做了对比:总支出超出 8.9%,其中有四个条目超过 20%,它们占了 15,200 美元增幅中的 11,900 美元。每一项都有原因、负责人和下一步行动。

Output: april-cost-vs-trend (chart)

输出:april-cost-vs-trend(图表)

Connectors: Snowflake, Datadog, GitHub

连接器:Snowflake、Datadog、GitHub

Set up

准备工作

Connect your tools

连接你的工具

Claude is more powerful when it works directly with your systems. You control permissions and access.

当 Claude 能直接对接你的系统时,它的能力会更强。权限和访问范围由你掌控。

Add a connector for each tool you want Claude to use.

为每一个你希望 Claude 使用的工具添加一个连接器(connector)。

Snowflake

Snowflake

Query the cost-and-usage tables in the warehouse so the anomaly report runs against the same data Finance sees.

查询数据仓库中的成本与用量表,让异常报告所依据的数据和财务部门看到的完全一致。

Datadog

Datadog

Correlate a cost spike with the traffic, deploy, or job that ran in the same window.

把成本激增与同一时间段内的流量、部署或运行的任务关联起来。

GitHub (optional)

GitHub(可选)

Check what merged into the owning service around the date the spend changed.

查看在费用发生变化的那几天前后,负责的服务里合并了哪些代码改动。

Want to try this task before setting anything up? Drag in your files, or give Claude access to a folder in the Claude desktop app. Then start with the prompt.

想在配置任何东西之前先试试这个任务?把文件拖进来,或者在 Claude 桌面应用中授权 Claude 访问一个文件夹,然后从下面的提示词开始。

Set your working folder

设置工作文件夹

Drop the billing CSV, your tag-to-team mapping, and last month's report into one folder, then give Claude access to it in the Claude desktop app. With these files, Claude has the baseline and the ownership map. The anomaly report gets written back here. Create a project from the folder so your thresholds and known-okay exceptions persist month to month.

把账单 CSV 文件、标签与团队的对应表,以及上个月的报告放进同一个文件夹,然后在 Claude 桌面应用中授权 Claude 访问它。有了这些文件,Claude 就掌握了对比基准和“谁负责什么”的对应关系。异常报告也会写回到这个文件夹。再用这个文件夹创建一个项目(Project),这样你设定的阈值和已确认没问题的例外情况会每个月延续下去。

The prompt

提示词

Copy this into Claude

复制到 Claude 中

Compare this month's AWS costs to the trailing three-month average by service and tag:team. Flag any line more than 20% over trend, trace each to the workload or change behind it, and write the cost-anomaly report with an owner and recommended action for every item. Put a one-line summary for the infra channel at the top.
按服务和 tag:team(团队标签),把本月的 AWS 费用与过去三个月的平均值对比。标出所有超出趋势 20% 以上的条目,追查每一项背后对应的工作负载或变更,并写出成本异常报告,为每一项指定负责人和建议的处理措施。在报告顶部放一行给基础设施频道看的摘要。

Why this works

为什么这样有效

Get a better draft

让初稿更好

Finish it where the file lives

在文件所在的地方完成收尾

Claude for Excel

Claude for Excel

Check a flagged line against the billing export.

把被标记的条目与账单导出数据逐一核对。

Make Claude work for you

让 Claude 按你的方式工作

Tell Claude your own rules for this task, such as your tag scheme and the spikes you've already accepted, then save what you've done as a skill of your own. A few minutes of conversation and it runs with your standards from then on.

把你在这个任务上的规则告诉 Claude,比如你们的标签体系、已经认可的费用激增,然后把做过的事保存为你自己的技能(skill)。聊上几分钟,之后它就会按你的标准来执行。

Make what we've done in this task so far into a skill called /cost-anomalies.
把我们在这个任务里到目前为止做的事情,做成一个名为 /cost-anomalies 的技能。

Review the skill, and save it if Claude hasn't already.

检查一下这个技能;如果 Claude 还没保存,就把它保存下来。

Tip: tell Claude to edit the skill for you.

小提示:可以直接让 Claude 帮你修改这个技能。

Make it repeatable

让它可以重复运行

Run it when the bill closes

在账单结算后自动运行

The report should be waiting before anyone opens the console. Ask Claude to set up a scheduled task, or open Scheduled in the sidebar, and /cost-anomalies runs against the new export on the first Monday after it's in the folder.

报告应该在任何人打开控制台之前就已经准备好。让 Claude 设置一个定时任务,或者在侧边栏打开 Scheduled(定时),新的账单导出文件放进文件夹后的第一个周一,/cost-anomalies 就会自动对它运行。

Set up a scheduled task for every Monday at 7am: if a new month's billing export is in Infra/cloud-cost, run /cost-anomalies against it, write the anomaly report to that month's folder, and put the one-line summary for the infra channel at the top.
设置一个每周一早上 7 点运行的定时任务:如果 Infra/cloud-cost 里出现了新一个月的账单导出文件,就对它运行 /cost-anomalies,把异常报告写入该月的文件夹,并在顶部放一行给基础设施频道看的摘要。

Scheduled task: Monthly cloud-cost anomaly report

定时任务:Monthly cloud-cost anomaly report(月度云成本异常报告)

Runs /cost-anomalies on the closed month, writes the ranked anomaly report to the dated folder, and puts the one-line summary for the infra channel at the top.

对已结算的月份运行 /cost-anomalies,把排好序的异常报告写入对应日期的文件夹,并在顶部放一行给基础设施频道看的摘要。

Active · Monday at 7:00am

已启用 · 每周一上午 7:00

To use a folder on your computer during a scheduled task, keep your computer on and awake with the Claude desktop app open so Claude can access it.

如果定时任务要用到你电脑上的文件夹,请保持电脑开机、不休眠,并打开 Claude 桌面应用,这样 Claude 才能访问它。

Share with your teammates

分享给你的同事

Your /cost-anomalies skill now knows your tag scheme, your threshold, and the spikes you've already accepted. Share it so any infra engineer can run the same analysis mid-month when something looks off.

你的 /cost-anomalies 技能现在已经了解你们的标签体系、阈值,以及已经认可的费用激增。把它分享出去,这样任何一位基础设施工程师在月中发现异样时,都能跑一遍同样的分析。

Share the skill

分享技能

You can share a skill you created. In your skills, open the ... menu next to /cost-anomalies, select Share, and pick your teammates. For your whole organization, open the skill and select Publish to org, if your admin allows. They get the skill with your thresholds and exceptions baked in, so they don't have to build it again.

你可以分享自己创建的技能。在技能列表中,点开 /cost-anomalies 旁边的 ... 菜单,选择 Share(分享),然后选择你的同事。如果要分享给整个组织,打开这个技能并选择 Publish to org(发布到组织),前提是管理员允许。他们拿到的技能已经内置了你的阈值和例外规则,不用再从头搭一遍。

What changes for the bill review

这对账单审查带来什么变化

Cloud spend is checked against trend with each anomaly traced to a cause, costed, and assigned an owner — ready to act on instead of investigate.

云支出会与趋势做对比,每个异常都追查到原因、算清金额并指定负责人——拿到手就能直接处理,而不必再去调查。

You did this for one cloud account. The same approach covers other providers, SaaS tool spend, and per-team budget checks — each one becomes a skill your team runs the same way.

你这次是针对一个云账户做的。同样的方法也适用于其他云服务商、SaaS 工具支出,以及按团队的预算检查——每一项都可以变成团队用同一种方式运行的技能。

Use case · 用例 · 10 min

Write the design doc or RFC

撰写设计文档或 RFC

A structured proposal with prior art and trade-offs filled in.

一份结构清晰的提案,先例和利弊权衡都已填好。

Source / 原文:https://academy.claude.com/use-cases/design-doc

Claude Cowork is now just Claude. What you learn here still applies — simply ask Claude.

Claude Cowork 现在就是 Claude。您在这里学到的内容依然适用,直接询问 Claude 即可。

Film: the user asks Claude to research prior art and constraints, then draft a design doc in their template. Claude searches connected wiki and code, pauses to ask about sections only the user can decide, then returns a structured draft with trade-offs and open questions.

视频:用户让 Claude 先调研先例和限制条件,再按用户的模板起草一份设计文档。Claude 搜索已连接的 wiki 和代码,遇到只有用户能决定的章节时停下来询问,最后返回一份包含利弊权衡和待解决问题的结构化草稿。

Demo walkthrough

演示流程

You need to turn rough notes into a design doc your team can review.

你需要把一份粗略的笔记变成团队可以评审的设计文档。

Using the Engineering plugin's system design skill, ask for prior art and trade-offs before the draft.

借助 Engineering 插件里的系统设计技能,在起草之前先要求 Claude 查找先例、分析利弊。

Claude reads your notes and template, then searches the wiki and code for earlier decisions.

Claude 先读你的笔记和模板,再到 wiki 和代码里搜索以往做过的决策。

When something is yours to decide, Claude stops and asks.

遇到需要你来拍板的事,Claude 会停下来问你。

Claude tells you what it found, then gives you the draft RFC.

Claude 告诉你它找到了什么,然后交给你 RFC 草稿。

Each approach is weighed against the others and links to earlier proposals.

每种方案都与其他方案做了比较,并链接到以往的相关提案。

The parts you kept are titled and empty for reviewers to see.

你留给自己写的部分只保留了标题、内容为空,评审者一眼就能看到。

You can schedule the same draft for every new notes file in the inbox folder.

你可以设置定时任务,让收件箱文件夹里每出现一份新笔记,就自动起草一份同样的草稿。

Your team gets a complete design doc and reviews the decision, not the writing.

团队拿到的是一份完整的设计文档,评审的重点是决策本身,而不是文字。

You: RFC for moving our event pipeline to streaming. Look up prior art and constraints, then draft in our template with two or three approaches and trade-offs, a recommendation and open questions. I'll fill in the parts only I know.

你:写一份把我们的事件管道迁移到流式处理的 RFC。先查找先例和限制条件,再按我们的模板起草,包含两到三种方案及其利弊、一个推荐方案和待解决问题。只有我知道的部分由我来填。

Claude asks: Draft the rollout sections for you, or leave them empty? Options: "Leave them for me" / "Draft something I'll correct"

Claude 提问:上线部署相关的章节要我替你起草,还是留空?选项:“留给我来写” / “先起草,我再改”

Claude: I read your notes and template, searched Confluence and the two repos for prior art, and drafted RFC-0087 with three approaches and trade-offs, citing RFC-0072, RFC-0061 and ADR-019. Recommendation is A, phased; Rollout plan and Operational impact are titled and left for you, and three open questions are listed.

Claude:我读了你的笔记和模板,在 Confluence 和两个代码仓库里查找了先例,起草了 RFC-0087,包含三种方案及其利弊,并引用了 RFC-0072、RFC-0061 和 ADR-019。推荐方案 A(分阶段推进);“上线计划”和“运维影响”两节只写了标题、留给你来填,另外列出了三个待解决问题。

Output: 0087-streaming-events.md

输出:0087-streaming-events.md

Skill: /system-design (Engineering) · Connectors: Confluence, GitHub

技能:/system-design(Engineering)· 连接器:Confluence、GitHub

Set up

准备工作

Try a plugin

试用插件

The Engineering plugin ships with /system-design as a starting point, already structured to work through requirements and trade-offs, then write a design document with diagrams. If your admin manages plugins and it's not available yet, skip this step. The prompt below works without it.

Engineering 插件自带 /system-design 技能作为起点,它已经设计好了流程:先梳理需求和利弊,再写出附带图示的设计文档。如果你的插件由管理员统一管理、而这个插件还没开放,可以跳过这一步,下面的提示词不用插件也能用。

Engineering

Engineering

Streamline engineering workflows — standups, code review, architecture decisions, incident response, and technical documentation. Works with your existing tools or on its own.

简化工程工作流程——站会、代码审查、架构决策、事故响应和技术文档。可与您现有的工具配合使用,也可独立使用。

/system-design

/system-design

Design systems, services, and architecture.

设计系统、服务和架构。

/architecture

/architecture

Create or evaluate architecture decision records (ADRs).

创建或评估架构决策记录(ADR)。

The plugin includes 10 skills in total.

该插件共包含 10 项技能。

Connect your tools

连接你的工具

Claude is more powerful when it works directly with your systems. You control permissions and access.

当 Claude 能直接对接你的系统时,它的能力会更强。权限和访问范围由你掌控。

Add a connector for each tool you want Claude to use.

为每一个你希望 Claude 使用的工具添加一个连接器(connector)。

Confluence

Confluence

Look up prior RFCs and architecture pages and write the new doc straight into your engineering space. Confluence access comes through the Atlassian connector (Jira and Confluence).

查阅以往的 RFC 和架构页面,并把新文档直接写进你们的工程空间。Confluence 的访问通过 Atlassian 连接器(包含 Jira 和 Confluence)实现。

GitHub

GitHub

Read the relevant code and past ADRs so the proposal references the system as it actually is.

读取相关代码和以往的 ADR(架构决策记录),让提案引用的是系统的真实现状。

Notion (optional)

Notion(可选)

Pull the RFC template and publish the draft into your design-review database.

调取 RFC 模板,并把草稿发布到你们的设计评审数据库中。

Want to try this task before setting anything up? Drag in your files, or give Claude access to a folder in the Claude desktop app. Then start with the prompt.

想在配置任何东西之前先试试这个任务?把文件拖进来,或者在 Claude 桌面应用中授权 Claude 访问一个文件夹,然后从下面的提示词开始。

Set your working folder

设置工作文件夹

Drag the files you'll use (your RFC template, the one-pager or notes you've already written, related past RFCs, the service's architecture diagram) into one folder and give Claude access to it in the Claude desktop app. Claude reads from there and writes the draft back to it. Create a project from your RFCs folder so your template and past RFCs stay attached.

把要用到的文件(你的 RFC 模板、已经写好的一页纸简介或笔记、相关的历史 RFC、该服务的架构图)拖进同一个文件夹,并在 Claude 桌面应用中授权 Claude 访问它。Claude 会从这里读取资料,并把草稿写回这里。再用你的 RFCs 文件夹创建一个项目(Project),这样模板和历史 RFC 会一直关联在一起。

The prompt

提示词

Copy this into Claude

复制到 Claude 中

Write a design doc for moving our event pipeline from batch to streaming. Look up prior art and architecture constraints, then draft the proposal in our RFC template: problem, goals and non-goals, two or three approaches with trade-offs, the recommendation, and open questions. I'll fill in the parts only I know.
写一份设计文档,内容是把我们的事件管道从批处理迁移到流式处理。先查找先例和架构上的限制条件,再按我们的 RFC 模板起草提案:问题、目标与非目标、两到三种方案及其利弊、推荐方案,以及待解决问题。只有我知道的部分由我来填。

Why this works

为什么这样有效

Get a better draft

让初稿更好

Make Claude work for you

让 Claude 按你的方式工作

Tell Claude your own rules for this task, such as your prior-art sources and your reviewers' standing questions, then save what you've done as a skill of your own. A few minutes of conversation and it runs with your standards from then on.

把你在这个任务上的规则告诉 Claude,比如去哪里查先例、评审者每次都会问的问题,然后把做过的事保存为你自己的技能(skill)。聊上几分钟,之后它就会按你的标准来执行。

Make what we've done in this task so far into a skill called /rfc-draft.
把我们在这个任务里到目前为止做的事情,做成一个名为 /rfc-draft 的技能。

Review the skill, and save it if Claude hasn't already.

检查一下这个技能;如果 Claude 还没保存,就把它保存下来。

Tip: tell Claude to edit the skill for you.

小提示:可以直接让 Claude 帮你修改这个技能。

Make it repeatable

让它可以重复运行

Run it from any one-pager

从任意一页纸简介自动起草

Design docs start as a paragraph in a Slack thread or a notes file. Make an inbox folder inside your RFCs folder and save each new one-pager there. Ask Claude to set up a scheduled task or open Scheduled in the sidebar, and /rfc-draft turns any new one-pager in your RFCs/inbox folder into a structured first draft.

设计文档往往起源于 Slack 讨论串里的一段话或一份笔记。在你的 RFCs 文件夹里建一个 inbox(收件箱)文件夹,把每份新的一页纸简介存进去。让 Claude 设置一个定时任务,或者在侧边栏打开 Scheduled(定时),/rfc-draft 就会把 RFCs/inbox 里任何新的简介变成一份结构化的初稿。

Set up a scheduled task for weekdays at 9am: check RFCs/inbox for new one-pagers, run /rfc-draft on each one, and write the structured draft to a numbered folder under RFCs/.
设置一个每个工作日早上 9 点运行的定时任务:检查 RFCs/inbox 里有没有新的一页纸简介,对每一份运行 /rfc-draft,并把结构化草稿写入 RFCs/ 下一个带编号的文件夹。

Scheduled task: RFC inbox drafts

定时任务:RFC inbox drafts(RFC 收件箱草稿)

Weekdays at 9am, picks up new one-pagers from RFCs/inbox, runs /rfc-draft with prior-art lookup, and writes a structured draft to a numbered RFC folder.

每个工作日早上 9 点,从 RFCs/inbox 取出新的一页纸简介,运行带先例查找的 /rfc-draft,并把结构化草稿写入一个带编号的 RFC 文件夹。

Active · Weekdays at 9:00am

已启用 · 每个工作日上午 9:00

To use a folder on your computer during a scheduled task, keep your computer on and awake with the Claude desktop app open so Claude can access it.

如果定时任务要用到你电脑上的文件夹,请保持电脑开机、不休眠,并打开 Claude 桌面应用,这样 Claude 才能访问它。

Share with your teammates

分享给你的同事

Your /rfc-draft skill now carries your template, your prior-art sources, and your reviewers' standing questions. Share it so every tech lead's RFC arrives in the same shape, and architecture review spends time on the decision instead of the formatting.

你的 /rfc-draft 技能现在已经包含了你的模板、查找先例的来源,以及评审者每次都会问的问题。把它分享出去,让每位技术负责人提交的 RFC 都是同一种格式,架构评审就能把时间花在决策上,而不是格式上。

Share the skill

分享技能

You can share a skill you created. In your skills, open the ... menu next to /rfc-draft, select Share, and pick your teammates. For your whole organization, open the skill and select Publish to org, if your admin allows. They get the skill with your template and your reviewers' standing questions baked in, so they don't have to build it again.

你可以分享自己创建的技能。在技能列表中,点开 /rfc-draft 旁边的 ... 菜单,选择 Share(分享),然后选择你的同事。如果要分享给整个组织,打开这个技能并选择 Publish to org(发布到组织),前提是管理员允许。他们拿到的技能已经内置了你的模板和评审者的常见问题,不用再从头搭一遍。

What changes for design review

这对设计评审带来什么变化

You have a complete RFC draft in your template, with prior art and trade-offs filled in from your own sources — ready to edit and review instead of write from a blank page.

你手里有一份按你的模板写好的完整 RFC 草稿,先例和利弊权衡都已根据你自己的资料填好——可以直接修改和评审,而不用从一张白纸开始写。

You did this for one proposal. The same approach covers API changes, schema migrations, and architecture decisions — each one becomes a skill your team runs the same way.

你这次是针对一份提案做的。同样的方法也适用于 API 变更、数据库结构迁移和架构决策——每一项都可以变成团队用同一种方式运行的技能。

Use case · 用例 · 10 min

Draft your reports' performance reviews

为你的下属起草绩效评估

A grounded review draft with linked evidence, per report.

为每位下属生成一份有据可依、附带证据链接的评估草稿。

Source / 原文:https://academy.claude.com/use-cases/eng-performance-reviews

Engineering · Claude Cowork

工程 · Claude Cowork

Claude Cowork is now simply Claude. What you learn here still applies — just ask Claude directly.

Claude Cowork 现在就是 Claude。您在这里学到的内容依然适用,直接询问 Claude 即可。

Film: the user asks Claude to draft each report's performance review from a skill, pulling their pull requests, projects, and Slack threads against the team's career framework, then watches Claude produce linked-evidence drafts with ratings left blank for the manager's judgment.

视频:用户让 Claude 借助一个技能为每位下属起草绩效评估,依据团队的职业发展框架,调取他们的 pull request(代码合并请求)、项目和 Slack 讨论串;随后 Claude 生成附带证据链接的草稿,评级一栏留空,交由经理自行判断。

Demo walkthrough

演示流程

Draft your reports’ performance reviews

为你的下属起草绩效评估

You have performance reviews to write for each of your reports.

你需要为每位下属撰写绩效评估。

Ask the performance review skill for a draft per report, every claim linked, and no rating.

让绩效评估技能为每位下属各写一份草稿,每条结论都附上链接,并且不给出评级。

Claude reads your framework and last cycle’s reviews, then each report’s pull requests, projects and threads.

Claude 先阅读你的职业框架和上一周期的评估,再查看每位下属的 pull request、项目和讨论串。

Claude tells you what it read, then gives you a draft per report.

Claude 告诉你它读了哪些内容,然后为每位下属交付一份草稿。

Every claim links to the work behind it.

每条结论都链接到背后的具体工作成果。

The rating is left blank for you to write.

评级留空,由你亲自填写。

You can schedule it so the drafts are ready when each new cycle’s folder appears.

你可以把它设为定时任务,这样每个新周期的文件夹一出现,草稿就已准备好。

You start each review from the evidence and write the rating yourself.

每份评估你都从证据出发,评级则由你亲自来写。

For each engineer in team-roster.md, draft their Q1 review in our framework with evidence linked for every claim, from this quarter's work and last review's growth areas. Flag where I need my own judgment and do not write the rating.
针对 team-roster.md 中的每位工程师,按照我们的框架起草他们的第一季度评估,依据本季度的工作和上次评估中的成长方向,每条结论都附上证据链接。标出需要我自己判断的地方,不要写评级。

I drafted five reviews from 87 pull requests, 9 Linear projects and 31 Slack threads, each under the framework’s four competencies with every claim linked to its source. Ratings are blank for you, and seven places are marked “Needs your judgment”; Anika’s draft has two.

我根据 87 个 pull request、9 个 Linear 项目和 31 个 Slack 讨论串起草了五份评估,每份都按框架的四项能力维度组织,每条结论都链接到出处。评级留空待你填写,共有七处标注了“需要你的判断”,其中 Anika 的草稿有两处。

Output files: anika-sorensen-2026-Q1.md, devraj-pillai-2026-Q1.md

输出文件:anika-sorensen-2026-Q1.md、devraj-pillai-2026-Q1.md

Skill: /performance-review (Human Resources)

技能:/performance-review(Human Resources)

Connectors: GitHub, Linear, Slack

连接器:GitHub、Linear、Slack

Set up

准备工作

Try a plugin

试用插件

The Human Resources plugin ships with /performance-review as a starting point, already structured to gather a report's work for the period and fill in its manager review template. If your admin manages plugins and it's not available yet, skip this step. The prompt below works without it.

Human Resources 插件自带 /performance-review 技能作为起点,它已经设计好流程:收集下属在评估周期内的工作,并填入经理评估模板。如果插件由你的管理员统一管理、而你暂时还用不了,可以跳过这一步。下面的提示词不依赖这个插件也能用。

Human Resources

Human Resources

Streamline people operations, covering recruiting, onboarding, performance reviews, compensation analysis, and policy guidance. Stay compliant and keep your team running smoothly.

简化人事运营工作,涵盖招聘、入职、绩效评估、薪酬分析和政策指导。保持合规,让您的团队顺畅运转。

/performance-review

/performance-review

Plan the structure of performance reviews, including self-assessments, manager review templates, and calibration prep.

规划绩效评估的结构,包括自我评估、经理评估模板和校准准备。

/people-report

/people-report

Generate headcount, attrition, diversity, or org-health reports.

生成员工人数、人员流失、多元化或组织健康度报告。

The plugin includes 9 skills in total.

该插件共包含 9 项技能。

Connect your tools

连接你的工具

Claude is more powerful when it works directly with your systems. You control permissions and access.

当 Claude 能直接接入你的系统时,它的能力会更强。权限和访问范围由你掌控。

Add a connector for each tool you want Claude to use.

为每个你希望 Claude 使用的工具添加一个连接器。

GitHub

GitHub

Pull each report's merged PRs, reviews given, and incidents responded to over the period.

调取每位下属在评估周期内已合并的 PR、给出的代码审查,以及参与响应的故障事件。

Slack

Slack

Surface the threads where they led, unblocked a teammate, or drove a decision.

找出他们发挥带头作用、帮队友扫除障碍或推动决策的讨论串。

Linear (optional)

Linear(可选)

List the projects and epics they shipped and the cycle outcomes they owned.

列出他们交付的项目和史诗任务(epic,即大型需求合集),以及他们负责的迭代周期成果。

Want to try this task before setting anything up? Drag in your files, or give Claude access to a folder in the Claude desktop app. Then start with the prompt.

想在做任何设置之前先试试这个任务?把你的文件拖进来,或者在 Claude 桌面应用中授权 Claude 访问一个文件夹,然后从提示词开始。

Set your working folder

设置工作文件夹

Drag the files you'll use (your career framework, the review template, last cycle's reviews, your team roster) into one folder and give Claude access to it in the Claude desktop app. Claude reads the framework from there and writes one draft per report back to it. Create a project from that folder so your framework and your writing voice stay attached.

把要用到的文件(职业发展框架、评估模板、上一周期的评估、团队名单)拖进同一个文件夹,并在 Claude 桌面应用中授权 Claude 访问它。Claude 会从那里读取框架,并为每位下属写一份草稿存回该文件夹。再用这个文件夹创建一个项目(Project),让你的框架和写作风格始终与之绑定。

The prompt

提示词

Copy this into Claude

把这段复制到 Claude 中

I'm writing quarterly reviews for my reports. For each engineer in my team list, draft the review in our framework with evidence linked for every claim, drawing on their work this quarter and their last review's growth areas. Flag where I need to add my own judgment; do not write the rating.
我正在为下属撰写季度评估。针对我团队名单中的每位工程师,按照我们的框架起草评估,依据他们本季度的工作和上次评估中的成长方向,每条结论都附上证据链接。标出需要我补充自己判断的地方;不要写评级。

Why this works

为什么有效

Get a better draft

让草稿更好

Make Claude work for you

让 Claude 按你的方式工作

Tell Claude your own rules for this task, such as which evidence sources to use and the tone you want, then save what you've done as a skill of your own. A few minutes of conversation and it runs with your standards from then on.

告诉 Claude 你对这项任务的规则,比如使用哪些证据来源、想要什么语气,然后把你做过的事保存成你自己的技能。只需几分钟的对话,之后它就会一直按你的标准来执行。

Make what we've done in this task so far into a skill called /team-review-drafts.
把我们在这项任务中到目前为止做的事,做成一个名为 /team-review-drafts 的技能。

Review the skill, and save it if Claude hasn't already.

检查一下这个技能;如果 Claude 还没保存,就把它保存下来。

Tip: tell Claude to edit the skill for you.

小贴士:可以直接让 Claude 帮你修改这个技能。

Make it repeatable

让它可以反复运行

Run it at the start of every cycle

在每个周期开始时运行

The drafts should be ready before review season opens. Ask Claude to set up a scheduled task, or open Scheduled in the sidebar, and /team-review-drafts builds the drafts once you add a folder for the new quarter.

草稿应该在评估季开始之前就准备好。让 Claude 设置一个定时任务,或者打开侧边栏中的 Scheduled(定时任务),这样只要你为新季度添加了文件夹,/team-review-drafts 就会自动生成草稿。

Set up a scheduled task for every Monday at 9am: if a new quarter folder exists under Reviews and has no drafts yet, run /team-review-drafts for everyone in team-roster.md and write each draft to Reviews/<quarter>/<name>.md.
设置一个每周一上午 9 点运行的定时任务:如果 Reviews 下出现了新的季度文件夹且里面还没有草稿,就为 team-roster.md 中的每个人运行 /team-review-drafts,并把每份草稿写入 Reviews/<quarter>/<name>.md。

Scheduled task · Active

定时任务 · 活跃

Quarterly review drafts

季度评估草稿

When a new quarter folder appears, runs /team-review-drafts for each report against GitHub, Linear, and Slack and writes evidence-linked drafts to the cycle folder.

每当出现新的季度文件夹,就基于 GitHub、Linear 和 Slack 为每位下属运行 /team-review-drafts,并把附带证据链接的草稿写入该周期的文件夹。

Monday at 9:00am

每周一上午 9:00

To use a folder on your computer during a scheduled task, keep your computer on and awake with the Claude desktop app open so Claude can access it.

如果定时任务需要用到你电脑上的文件夹,请让电脑保持开机且不休眠,并保持 Claude 桌面应用打开,这样 Claude 才能访问它。

Share with your teammates

与团队成员共享

Your /team-review-drafts skill now carries your career framework, your evidence sources, and your tone. Share it so every manager in the org drafts against the same framework, and calibration compares apples to apples.

你的 /team-review-drafts 技能现在已经包含了你的职业框架、证据来源和语气。把它共享出去,让组织里的每位经理都按同一套框架起草,校准时比较的才是同一标准下的结果。

Share the skill

共享技能

You can share a skill you created. In your skills, open the ... menu next to /team-review-drafts, select Share, and pick your peer managers. For your whole organization, open the skill and select Publish to org, if your admin allows. They get the skill with your framework baked in, so they don't have to build it again.

你可以共享自己创建的技能。在你的技能列表中,打开 /team-review-drafts 旁边的 ... 菜单,选择 Share(共享),然后选择你的同级经理。如果要共享给整个组织,在管理员允许的情况下,打开该技能并选择 Publish to org(发布到组织)。他们拿到的技能已经内置了你的框架,无需重新搭建。

What changes for review season

评估季会有什么变化

Each report has a review draft with evidence linked for every claim and the places that need your judgment clearly marked. Your time goes to the assessment and the conversation, not to gathering what they shipped.

每位下属都有一份评估草稿,每条结论都附有证据链接,需要你判断的地方也标得清清楚楚。你的时间花在评估本身和与下属的面谈上,而不是去搜罗他们交付了什么。

You did this for one review cycle. The same approach covers promotion packets, calibration prep, and peer-feedback summaries — each one becomes a skill your team runs the same way.

你这次只是为一个评估周期做了这件事。同样的方法也适用于晋升材料、校准准备和同事反馈汇总——每一项都可以变成团队以同样方式运行的技能。

Use case · 用例 · 10 min

Draft the incident postmortem

起草故障事后复盘

Timeline, root cause, and action items from the incident channel.

从故障频道中整理出时间线、根本原因和行动项。

Source / 原文:https://academy.claude.com/use-cases/incident-postmortem

Engineering · Claude Cowork

工程 · Claude Cowork

Claude Cowork is now simply Claude. What you learn here still applies — just ask Claude directly.

Claude Cowork 现在就是 Claude。您在这里学到的内容依然适用,直接询问 Claude 即可。

Film: a self-playing walkthrough showing the user asking Claude to draft an incident postmortem from a Slack channel and alert timeline, with Claude pausing to ask a judgment call before returning a structured, blameless draft with owned action items.

视频:一段自动播放的演示,展示用户让 Claude 根据 Slack 频道和告警时间线起草故障事后复盘;Claude 中途停下来请用户做一个需要判断的决定,然后交付一份结构清晰、不追究个人责任(blameless)的草稿,每个行动项都有负责人。

Demo walkthrough

演示流程

Draft the incident postmortem

起草故障事后复盘

The incident is over and you need to write the postmortem.

故障已经结束,你需要撰写事后复盘。

Using the Engineering plugin’s incident response skill, ask for a timestamped sequence and owned action items.

使用 Engineering 插件的故障响应技能,要求给出带时间戳的事件经过,以及有明确负责人的行动项。

Claude reads the channel, the alert timeline and your template.

Claude 阅读故障频道、告警时间线和你的模板。

When something is yours to decide, Claude stops and asks.

遇到需要你来决定的事情时,Claude 会停下来问你。

Claude says what it rebuilt and gives you the postmortem and a draft note.

Claude 说明它还原了哪些内容,然后交给你复盘文档和一条频道通知草稿。

Every timeline entry comes from a message or an alert.

时间线上的每一条记录都来自某条消息或某次告警。

Every action item already has an owner and a date.

每个行动项都已经有负责人和截止日期。

You approve the channel note first, and can schedule a first draft after every incident.

频道通知要先经你批准才会发出;你还可以设置在每次故障之后自动生成初稿。

The postmortem is written from what actually happened, ready to review.

复盘文档依据真实发生的事情写成,可以直接进入审阅。

From #inc-2026-04-22-checkout-5xx and the PagerDuty timeline, reconstruct what happened with timestamps, then draft the postmortem in our template: impact, root cause, contributing factors, and action items with an owner on each. Be blameless. Draft a channel note linking it.
根据 #inc-2026-04-22-checkout-5xx 频道和 PagerDuty 时间线,带时间戳地还原事件经过,然后按我们的模板起草复盘:影响、根本原因、促成因素,以及每项都有负责人的行动项。不要追究个人责任。再起草一条附上复盘链接的频道通知。

Claude asks: When did the incident start: first errors or the alert? Options: 14:04, first errors on the graph / 14:07, when the alert fired

Claude 提问:故障从什么时候算起——第一次出现错误,还是告警触发时?选项:14:04,图表上首次出现错误 / 14:07,告警触发时

I rebuilt the incident from the channel and the PagerDuty timeline as 13 timestamped entries, counted impact from 14:04, and gave all five action items an owner and a date. Two numbers I am unsure of are under Open questions, and the channel note waits for your approval.

我根据频道和 PagerDuty 时间线把这次故障还原成 13 条带时间戳的记录,影响从 14:04 开始计算,并为全部五个行动项指定了负责人和日期。有两个我不确定的数字放在了“待确认问题”部分,频道通知正在等你批准。

Output files: checkout-5xx-postmortem-2026-04-22.md, inc-48213-action-items.csv

输出文件:checkout-5xx-postmortem-2026-04-22.md、inc-48213-action-items.csv

Skill: /incident-response (Engineering)

技能:/incident-response(Engineering)

Connectors: Slack, PagerDuty

连接器:Slack、PagerDuty

Set up

准备工作

Try a plugin

试用插件

The Engineering plugin ships with /incident-response and other incident-and-ops skills as a starting point, already structured to write a blameless postmortem with a timestamped timeline. If your admin manages plugins and it's not available yet, skip this step. The prompt below works without it.

Engineering 插件自带 /incident-response 以及其他故障和运维相关技能作为起点,它们已经设计好流程,可以写出带时间戳时间线、不追究个人责任的事后复盘。如果插件由你的管理员统一管理、而你暂时还用不了,可以跳过这一步。下面的提示词不依赖这个插件也能用。

Engineering

Engineering

Streamline engineering workflows — standups, code review, architecture decisions, incident response, and technical documentation. Works with your existing tools or on its own.

简化工程工作流程——站会、代码审查、架构决策、事故响应和技术文档。可与您现有的工具配合使用,也可独立使用。

/incident-response

/incident-response

Run the incident response workflow — triage, communicate, and write the postmortem.

运行事件响应工作流——分诊、沟通并撰写事后复盘。

/standup

/standup

Generate a standup progress update from recent activity.

根据近期活动生成站会进展更新。

The plugin includes 10 skills in total.

该插件共包含 10 项技能。

Connect your tools

连接你的工具

Claude is more powerful when it works directly with your systems. You control permissions and access.

当 Claude 能直接接入你的系统时,它的能力会更强。权限和访问范围由你掌控。

Add a connector for each tool you want Claude to use.

为每个你希望 Claude 使用的工具添加一个连接器。

Slack

Slack

Read the incident channel end to end and pull every timestamp, decision, and status update.

从头到尾阅读故障频道,提取每个时间戳、每项决定和每次状态更新。

PagerDuty

PagerDuty

Pull the alert timeline, who was paged, and when the incident was acknowledged and resolved.

调取告警时间线、被呼叫的人员,以及故障被确认和解决的时间。

Datadog (optional)

Datadog(可选)

Attach the error-rate and latency graphs for the impact window.

附上受影响时段内的错误率和延迟图表。

Want to try this task before setting anything up? Drag in your files, or give Claude access to a folder in the Claude desktop app. Then start with the prompt.

想在做任何设置之前先试试这个任务?把你的文件拖进来,或者在 Claude 桌面应用中授权 Claude 访问一个文件夹,然后从提示词开始。

Set your working folder

设置工作文件夹

Drag the files you'll use (your postmortem template, the runbook, any log exports or graph screenshots) into one folder and give Claude access to it in the Claude desktop app. Claude reads the template from there and writes the draft back to it. Create a project from your incidents folder so your template, severity definitions, and blameless-language guide stay attached.

把要用到的文件(复盘模板、运维手册 runbook、导出的日志或图表截图)拖进同一个文件夹,并在 Claude 桌面应用中授权 Claude 访问它。Claude 会从那里读取模板,并把草稿写回该文件夹。再用你的故障文件夹创建一个项目(Project),让模板、严重级别定义和“不追责”用语指南始终与之绑定。

The prompt

提示词

Copy this into Claude

把这段复制到 Claude 中

From #inc-2026-04-22-checkout-5xx, reconstruct the sequence of events with timestamps, write the customer impact summary, identify the root cause and contributing factors, and draft the postmortem in our template with owners on every action item. Be blameless: describe what the system did, not who made a mistake.
根据 #inc-2026-04-22-checkout-5xx 频道,带时间戳地还原事件经过,撰写客户影响摘要,找出根本原因和促成因素,并按我们的模板起草复盘,每个行动项都要有负责人。不要追究个人责任:描述系统做了什么,而不是谁犯了错。

Why this works

为什么有效

Get a better draft

让草稿更好

Make Claude work for you

让 Claude 按你的方式工作

Tell Claude your team's standards for postmortems, such as your severity definitions and your blameless-language rules, then save what you've done as a skill of your own. A few minutes of conversation and it runs with your standards from then on.

告诉 Claude 你们团队的复盘标准,比如严重级别定义和“不追责”用语规则,然后把你做过的事保存成你自己的技能。只需几分钟的对话,之后它就会一直按你的标准来执行。

Make what we've done in this task so far into a skill called /postmortem-draft.
把我们在这项任务中到目前为止做的事,做成一个名为 /postmortem-draft 的技能。

Review the skill, and save it if Claude hasn't already.

检查一下这个技能;如果 Claude 还没保存,就把它保存下来。

Tip: tell Claude to edit the skill for you.

小贴士:可以直接让 Claude 帮你修改这个技能。

Make it repeatable

让它可以反复运行

Run it when the incident closes

在故障关闭时运行

The draft should exist before the review meeting is scheduled. Ask Claude to set up a scheduled task, or open Scheduled in the sidebar, and /postmortem-draft runs on each resolved incident and writes the first draft into the matching folder.

复盘会议排期之前,草稿就应该已经写好。让 Claude 设置一个定时任务,或者打开侧边栏中的 Scheduled(定时任务),这样 /postmortem-draft 会针对每个已解决的故障运行,并把初稿写入对应的文件夹。

Set up a scheduled task for every weekday at 6pm: check PagerDuty for incidents resolved since the last run, run /postmortem-draft on each one, and write the draft to Incidents/<incident-folder>/postmortem-draft.md.
设置一个每个工作日下午 6 点运行的定时任务:在 PagerDuty 中查找自上次运行以来已解决的故障,对每一个运行 /postmortem-draft,并把草稿写入 Incidents/<incident-folder>/postmortem-draft.md。

Scheduled task · Active

定时任务 · 活跃

Postmortem first draft

复盘初稿

Every weekday at 6pm, finds incidents resolved since the last run, runs /postmortem-draft against the channel and timeline, and writes the draft to the incident folder.

每个工作日下午 6 点,查找自上次运行以来已解决的故障,基于频道和时间线运行 /postmortem-draft,并把草稿写入该故障的文件夹。

Weekdays at 6:00pm

每个工作日下午 6:00

To use a folder on your computer during a scheduled task, keep your computer on and awake with the Claude desktop app open so Claude can access it.

如果定时任务需要用到你电脑上的文件夹,请让电脑保持开机且不休眠,并保持 Claude 桌面应用打开,这样 Claude 才能访问它。

Share with your teammates

与团队成员共享

Your /postmortem-draft skill now carries your template, your severity definitions, and your blameless-language rules. Share it so every on-call engineer writes the doc the same way, and the review meeting starts at "is this right" instead of "who's writing this up."

你的 /postmortem-draft 技能现在已经包含了你的模板、严重级别定义和“不追责”用语规则。把它共享出去,让每位值班工程师都用同样的方式写复盘,复盘会议一开始讨论的就是“这样写对不对”,而不是“谁来写这份文档”。

Share the skill

共享技能

You can share a skill you created. In your skills, open the ... menu next to /postmortem-draft, select Share, and pick your teammates. For your whole organization, open the skill and select Publish to org, if your admin allows. They get the skill with your template and tone baked in, so they don't have to build it again.

你可以共享自己创建的技能。在你的技能列表中,打开 /postmortem-draft 旁边的 ... 菜单,选择 Share(共享),然后选择你的团队成员。如果要共享给整个组织,在管理员允许的情况下,打开该技能并选择 Publish to org(发布到组织)。他们拿到的技能已经内置了你的模板和语气,无需重新搭建。

What changes for incident review

故障复盘会有什么变化

You have a postmortem drafted from the incident record, with an owner on every action item — ready to review instead of write.

你手上有一份根据故障记录起草的复盘,每个行动项都有负责人——你要做的是审阅,而不是从头写。

You did this for one incident. The same approach covers on-call handoffs, runbook updates, and sprint retros — each one a skill your team runs the same way.

你这次只是为一次故障做了这件事。同样的方法也适用于值班交接、运维手册更新和迭代回顾——每一项都可以成为团队以同样方式运行的技能。

Use case · 用例 · 10 min

On-call handoff brief

值班交接简报

Everything that happened on your shift, written before you log off.

在你下班之前,把这一班发生的所有事情写好。

Source / 原文:https://academy.claude.com/use-cases/oncall-handoff-brief

Engineering · Claude Cowork

工程 · Claude Cowork

Claude Cowork is now simply Claude. What you learn here still applies — just ask Claude directly.

Claude Cowork 现在就是 Claude。您在这里学到的内容依然适用,直接询问 Claude 即可。

Film: a self-playing walkthrough where the user asks Claude to write an on-call handoff from pages, escalations and deploys, watches Claude pause for a judgment call, and gets a status-grouped brief ready to post and schedule for every rotation change.

视频:一段自动播放的演示,用户让 Claude 根据告警呼叫(page)、升级工单和部署记录撰写值班交接;Claude 中途停下来请用户做一个需要判断的决定,最后交付一份按状态分组的简报,可以直接发布,并设为每次轮班交接时定时生成。

Demo walkthrough

演示流程

Write the on-call handoff

撰写值班交接

Your on-call shift is ending and the next person needs to know what happened.

你的值班即将结束,下一位值班人需要知道这期间发生了什么。

With the Engineering plugin’s standup skill, ask what’s resolved, what’s open and what to watch.

借助 Engineering 插件的站会技能,询问哪些已解决、哪些仍未解决,以及需要关注什么。

Claude reads last week’s brief first, then the week’s pages, escalations and deploys.

Claude 先阅读上周的简报,再查看本周的告警呼叫、升级工单和部署记录。

PagerDuty alert: p99 latency > 2 s for 5 min — #5102

PagerDuty 告警:p99 延迟超过 2 秒并持续 5 分钟 — #5102

When something is yours to decide, Claude stops and asks.

遇到需要你来决定的事情时,Claude 会停下来问你。

You get the brief and a draft post.

你会拿到简报和一条待发布的消息草稿。

Each item has a status, so open ones are read first.

每一项都标有状态,所以未解决的事项会被优先看到。

Anything still open says what to watch and links the fix.

凡是仍未解决的事项,都会写明需要关注什么,并附上修复方案的链接。

Approve the post, then schedule the brief for every rotation change.

批准这条消息后,再把简报设为每次轮班交接时定时生成。

The next person on call starts knowing what is still open.

下一位值班人一上岗就清楚还有哪些事没解决。

Write the on-call handoff for my shift: pages, #support-escalations and prod deploys from the last 7 days. For each item note status (resolved, monitoring, still open), a one-line cause, and what the next on-call should watch. Post it to #oncall. What does the next on-call need to know?
为我这一班撰写值班交接:涵盖过去 7 天的告警呼叫、#support-escalations 频道和生产环境部署。每一项都注明状态(已解决、观察中、仍未解决)、一句话说明原因,以及下一位值班人应该关注什么。把它发到 #oncall。下一位值班人需要知道些什么?

Claude asks: Are the webhook delivery delays fixed, or still being watched? Options: Fixed, mark it resolved / Still watching it

Claude 提问:webhook 投递延迟的问题是已经修好了,还是仍在观察?选项:已修好,标记为已解决 / 仍在观察

I read 11 pages, four incident channels, 16 deploys and last week’s brief, and wrote the week 35 handoff: 8 items, 1 still open, 3 monitoring, 4 resolved. The open item is auth-service v2.31.0 until #2287 merges, and the #oncall post is ready for you to approve.

我查看了 11 次告警呼叫、四个故障频道、16 次部署和上周的简报,写好了第 35 周的交接:共 8 项,1 项仍未解决,3 项观察中,4 项已解决。未解决的那项是 auth-service v2.31.0,要等 #2287 合并后才算解决;#oncall 的消息已准备好,等你批准。

Output file: handoff-2026-W35.md

输出文件:handoff-2026-W35.md

Skill: /standup (Engineering)

技能:/standup(Engineering)

Connectors: PagerDuty, Slack, GitHub

连接器:PagerDuty、Slack、GitHub

Set up

准备工作

Try a plugin

试用插件

The Engineering plugin ships with /standup and other incident-and-ops skills as a starting point. Use /standup to summarize recent activity and blockers, and /incident-response to write up an incident. If your admin manages plugins and it's not available yet, skip this step. The prompt below works without it.

Engineering 插件自带 /standup 以及其他故障和运维相关技能作为起点。用 /standup 汇总近期动态和阻碍事项,用 /incident-response 撰写故障报告。如果插件由你的管理员统一管理、而你暂时还用不了,可以跳过这一步。下面的提示词不依赖这个插件也能用。

Engineering

Engineering

Streamline engineering workflows — standups, code review, architecture decisions, incident response, and technical documentation. Works with your existing tools or on its own.

简化工程工作流程——站会、代码审查、架构决策、事故响应和技术文档。可与您现有的工具配合使用,也可独立使用。

/standup

/standup

Generate a standup progress update from recent activity.

根据近期活动生成站会进展更新。

/incident-response

/incident-response

Run the incident response workflow — triage, communicate, and write the postmortem.

运行事件响应工作流——分诊、沟通并撰写事后复盘。

The plugin includes 10 skills in total.

该插件共包含 10 项技能。

Connect your tools

连接你的工具

Claude is more powerful when it works directly with your systems. You control permissions and access.

当 Claude 能直接接入你的系统时,它的能力会更强。权限和访问范围由你掌控。

Add a connector for each tool you want Claude to use.

为每个你希望 Claude 使用的工具添加一个连接器。

PagerDuty

PagerDuty

Pull every page from your shift with ack time, resolution, and the alert that fired.

调取你值班期间的每一次告警呼叫,包括确认时间、处理结果和触发的告警。

Slack

Slack

Read the incident channels and #support-escalations for what CS raised during the week.

阅读故障频道和 #support-escalations,了解客服(CS)本周提出了哪些问题。

GitHub (optional)

GitHub(可选)

List every production deploy that shipped during your shift, with the change summary.

列出你值班期间上线的每一次生产环境部署,并附上变更摘要。

Want to try this task before setting anything up? Drag in your files, or give Claude access to a folder in the Claude desktop app. Then start with the prompt.

想在做任何设置之前先试试这个任务?把你的文件拖进来,或者在 Claude 桌面应用中授权 Claude 访问一个文件夹,然后从提示词开始。

Set your working folder

设置工作文件夹

Drag the files you'll use (your handoff template, last week's brief, the runbook index) into one folder and give Claude access to it in the Claude desktop app. Claude reads the template from there and writes the brief back to it. Create a project from your on-call folder so the template and runbook index stay attached every rotation.

把要用到的文件(交接模板、上周的简报、运维手册索引)拖进同一个文件夹,并在 Claude 桌面应用中授权 Claude 访问它。Claude 会从那里读取模板,并把简报写回该文件夹。再用你的值班文件夹创建一个项目(Project),让模板和运维手册索引在每次轮班时都始终与之绑定。

The prompt

提示词

Copy this into Claude

把这段复制到 Claude 中

Write the on-call handoff for my shift. Cover pages and incident channels from the last 7 days, CS escalations from #support-escalations, and production deploys. For each, note status (resolved, monitoring, still open), the one-line cause if known, and what the next on-call should watch. Write it to the on-call folder and post it to #oncall.
为我这一班撰写值班交接。涵盖过去 7 天的告警呼叫和故障频道、#support-escalations 中的客服升级工单,以及生产环境部署。每一项都注明状态(已解决、观察中、仍未解决)、已知的话用一句话说明原因,以及下一位值班人应该关注什么。把它写入值班文件夹,并发到 #oncall。

Why this works

为什么有效

Get a better draft

让草稿更好

Make Claude work for you

让 Claude 按你的方式工作

Tell Claude your own rules for this task, such as your channel list and your status labels, then save what you've done as a skill of your own. A few minutes of conversation and it runs with your standards from then on.

告诉 Claude 你对这项任务的规则,比如要看哪些频道、用哪些状态标签,然后把你做过的事保存成你自己的技能。只需几分钟的对话,之后它就会一直按你的标准来执行。

Make what we've done in this task so far into a skill called /oncall-handoff.
把我们在这项任务中到目前为止做的事,做成一个名为 /oncall-handoff 的技能。

Review the skill, and save it if Claude hasn't already.

检查一下这个技能;如果 Claude 还没保存,就把它保存下来。

Tip: tell Claude to edit the skill for you.

小贴士:可以直接让 Claude 帮你修改这个技能。

Make it repeatable

让它可以反复运行

Run it at rotation change

在轮班交接时运行

Handoff happens the same hour every week. Ask Claude to set up a scheduled task, or open Scheduled in the sidebar, and /oncall-handoff writes the brief at the end of every rotation, whether you remembered or not.

交接每周都在同一时间进行。让 Claude 设置一个定时任务,或者打开侧边栏中的 Scheduled(定时任务),这样无论你记没记得,/oncall-handoff 都会在每轮值班结束时写好简报。

Set up a scheduled task for every Monday at 9am: run /oncall-handoff over the prior 7 days of pages, #support-escalations, and prod deploys, write the handoff to Platform/Oncall/handoff-<week>.md, and prepare the #oncall post for me to approve.
设置一个每周一上午 9 点运行的定时任务:针对前 7 天的告警呼叫、#support-escalations 和生产环境部署运行 /oncall-handoff,把交接写入 Platform/Oncall/handoff-<week>.md,并准备好 #oncall 的消息等我批准。

Scheduled task · Active

定时任务 · 活跃

Weekly on-call handoff

每周值班交接

Runs /oncall-handoff over the week's pages, CS escalations, and deploys, writes the brief to the on-call folder, and prepares the #oncall post for you to approve.

针对本周的告警呼叫、客服升级和部署运行 /oncall-handoff,把简报写入值班文件夹,并准备好 #oncall 的消息等你批准。

Mondays at 9:00am

每周一上午 9:00

To use a folder on your computer during a scheduled task, keep your computer on and awake with the Claude desktop app open so Claude can access it.

如果定时任务需要用到你电脑上的文件夹,请让电脑保持开机且不休眠,并保持 Claude 桌面应用打开,这样 Claude 才能访问它。

Share with your teammates

与团队成员共享

Your /oncall-handoff skill now carries your channel list, your status labels, and your team's handoff template. Share it so every engineer in the rotation produces the same brief, and nobody starts a shift blind because the last person was too tired to write it up.

你的 /oncall-handoff 技能现在已经包含了你的频道列表、状态标签和团队的交接模板。把它共享出去,让轮班中的每位工程师都产出同样格式的简报,也不会再有人因为上一班的同事太累没写交接,而两眼一抹黑地开始值班。

Share the skill

共享技能

You can share a skill you created. In your skills, open the ... menu next to /oncall-handoff, select Share, and pick your teammates. For your whole organization, open the skill and select Publish to org, if your admin allows. They get the skill with your instructions baked in, so they don't have to build it again.

你可以共享自己创建的技能。在你的技能列表中,打开 /oncall-handoff 旁边的 ... 菜单,选择 Share(共享),然后选择你的团队成员。如果要共享给整个组织,在管理员允许的情况下,打开该技能并选择 Publish to org(发布到组织)。他们拿到的技能已经内置了你的指令,无需重新搭建。

What changes for the rotation

轮班会有什么变化

The handoff is written and posted with each item's status, cause, and what to watch — the next engineer starts informed instead of searching channels for context.

交接简报已经写好并发布,每一项都有状态、原因和关注点——下一位工程师上岗时就掌握了情况,而不用到各个频道里翻找来龙去脉。

You did this for one rotation. The same approach covers incident postmortems, release summaries, and the weekly ops review — each one a skill your team runs the same way every time.

你这次只是为一轮值班做了这件事。同样的方法也适用于故障事后复盘、版本发布总结和每周运维回顾——每一项都可以成为团队每次都以同样方式运行的技能。

Use case · 用例 · 10 min

Write the sprint retro and on-call handoff

撰写冲刺回顾和值班交接文档

What shipped, what slipped, and what the next on-call needs to know.

哪些交付了,哪些延期了,以及下一位值班人员需要知道什么。

Source / 原文:https://academy.claude.com/use-cases/sprint-retro-handoff

Engineering · Claude Cowork. Claude Cowork is now just Claude. What you learn here still applies — just ask Claude.

工程 · Claude Cowork。Claude Cowork 现在就是 Claude。您在这里学到的内容依然适用,直接询问 Claude 即可。

Demo walkthrough

演示流程

Film: the user asks Claude to read a closed sprint cycle and team channels, then watches it draft a sourced retro and a separate on-call handoff, asking which themes to discuss before writing, with a post pending approval.

演示视频:用户让 Claude 读取一个已结束的冲刺周期和团队频道,然后看着它起草一份注明出处的回顾文档和一份单独的值班交接文档;Claude 在动笔前先询问要讨论哪些主题,最后留下一条待审批的频道帖子。

Demo setup: Linear, Slack, and PagerDuty connected; working folder Platform / retros / 2026-sprint-08 containing sprint-07-retro.md (last retro, format to match) and retro-template.md (team template).

演示环境:已连接 Linear、Slack 和 PagerDuty;工作文件夹为 Platform / retros / 2026-sprint-08,其中有 sprint-07-retro.md(上一次回顾,按其格式撰写)和 retro-template.md(团队模板)。

Write the sprint retro and handoff

撰写冲刺回顾和交接文档

The sprint has closed and you need the retro and the on-call handoff written.

冲刺已经结束,你需要写好回顾文档和值班交接文档。

Give Claude the cycle and channels, and ask why things slipped.

把周期和频道交给 Claude,并问它事情为什么延期。

Claude reads the cycle for what moved and the channels for why.

Claude 从周期里看哪些事项有进展,从频道里看背后的原因。

Wherever a decision is yours to make, Claude asks instead of guessing.

凡是需要由你来做决定的地方,Claude 都会先问你,而不是自己猜。

Claude tells you what it found, then gives you the retro and handoff.

Claude 先告诉你它发现了什么,然后交给你回顾文档和交接文档。

Every slipped item comes with its reason and where it was said.

每个延期事项都附有原因,以及这个原因是在哪里被提到的。

The next on-call gets what to watch, once you approve it.

经你批准后,下一位值班人员就能拿到需要重点关注的事项。

You approve the post first, and can schedule Claude to check each week for a closed cycle and draft both.

帖子要先经你批准;你还可以设置定时任务,让 Claude 每周检查是否有刚结束的周期,并起草这两份文档。

The retro and the handoff are written from what actually happened this sprint.

回顾文档和交接文档都基于本次冲刺中实际发生的事情来写。

You: Read the just-closed Linear cycle and two weeks of #team-platform and #oncall-platform. Write the sprint retro: shipped, slipped with reasons, themes worth discussing. Write a separate on-call handoff: open alerts, toil, what to watch. Keep each short enough to cover in a 30-minute meeting. Draft the handoff post to #oncall-platform for me to approve.

你:读取刚结束的 Linear 周期,以及 #team-platform 和 #oncall-platform 两周的消息。撰写冲刺回顾:已交付的、延期的(附原因)、值得讨论的主题。另写一份单独的值班交接文档:未处理的告警、重复性杂活、需要关注的事项。每份都要精简到 30 分钟的会议能讲完。把交接帖子起草到 #oncall-platform,等我批准。

Claude: Which of these should the retro meeting discuss? Staging environment instability (3 slips) / Review wait time (3-day waits) / Mid-cycle scope (3 late adds) / Alert noise (already tuned)

Claude:回顾会议应该讨论下面哪些主题?预发布环境不稳定(3 项延期)/ 代码审查等待时间(等了 3 天)/ 周期中途范围变动(3 项后加需求)/ 告警噪音(已调整过)

Claude: I read Sprint 08’s 21 issues and two weeks of #team-platform and #oncall-platform: 16 closed, and each slipped issue has its reason from the channel. Your two themes are written up as questions for Monday; Lena’s handoff is separate, and the #oncall-platform post is drafted for you to approve.

Claude:我读了 Sprint 08 的 21 个事项,以及 #team-platform 和 #oncall-platform 两周的消息:16 个已关闭,每个延期事项都附上了频道里提到的原因。你选的两个主题已经写成周一要讨论的问题;Lena 的交接文档单独成篇,#oncall-platform 的帖子也已起草好,等你批准。

Output files: sprint-08-retro.md, oncall-handoff-2026-04-24.md

输出文件:sprint-08-retro.md、oncall-handoff-2026-04-24.md

Connectors: Linear, Slack, PagerDuty

连接器:Linear、Slack、PagerDuty

Set up

准备工作

Connect your tools

连接你的工具

Claude is more powerful when it works directly with your systems. You control permissions and access.

当 Claude 能直接对接你的系统时,它的能力会更强。权限和访问范围由你来控制。

Add a connector for each tool you want Claude to use.

为你希望 Claude 使用的每个工具添加一个连接器。

Linear

Linear

Read the closed cycle: what was planned, what shipped, what carried over and why.

读取已结束的周期:计划了什么、交付了什么、哪些顺延到下一周期以及原因。

Slack

Slack

Pull the team and on-call channels for the context the tickets don't capture: the why-it-slipped and the 2am page.

拉取团队频道和值班频道,获取工单里没有记录的背景信息:为什么延期,以及凌晨 2 点的告警呼叫。

PagerDuty (optional)

PagerDuty(可选)

List every page in the rotation so toil is counted, not estimated.

列出本轮值班中的每一次告警呼叫,让重复性杂活的工作量有实数可查,而不是靠估算。

Want to try this task before setting anything up? Drag in your files, or give Claude access to a folder in the Claude desktop app. Then start with the prompt.

想在设置任何东西之前先试试这个任务?把文件拖进来,或者在 Claude 桌面应用中给 Claude 授权访问一个文件夹,然后从提示词开始。

Set your working folder

设置你的工作文件夹

Keep your retro template, the running action-items list, and last sprint's retro in one folder, then give Claude access to it in the Claude desktop app. Claude reads the format from there and writes both docs back so the retro and the handoff live next to their predecessors. Create a project from the folder so your team's definition of "shipped" and the discussion-prompt style carry over every cycle.

把回顾模板、持续更新的待办行动清单和上一次冲刺的回顾放在同一个文件夹里,然后在 Claude 桌面应用中授权 Claude 访问它。Claude 会从中读取格式,并把两份文档写回这个文件夹,让新的回顾和交接文档与以往的放在一起。用这个文件夹创建一个项目,这样团队对“已交付”的定义和讨论问题的写法就能在每个周期延续下去。

Folder Platform / retros / 2026-sprint-08: retro-template.md, sprint-07-retro.md, retro-action-items.md

文件夹 Platform / retros / 2026-sprint-08:retro-template.md、sprint-07-retro.md、retro-action-items.md

The prompt

提示词

Copy this into Claude — in Cowork’s chat bar, with the folder Platform / retros / 2026-sprint-08:

把下面的内容复制到 Claude 中——在 Cowork 的聊天栏里,选中文件夹 Platform / retros / 2026-sprint-08:

Read the just-closed Linear cycle and two weeks of #team-platform and #oncall-platform. Write the sprint retro (shipped, slipped with reasons, themes worth discussing) and a separate on-call handoff (open alerts, toil, what the next rotation should watch). Pull quotes where they help. Keep each section to what's discussable in a 30-minute meeting.
读取刚结束的 Linear 周期,以及 #team-platform 和 #oncall-platform 两周的消息。撰写冲刺回顾(已交付的、延期的及原因、值得讨论的主题),再另写一份值班交接文档(未处理的告警、重复性杂活、下一轮值班应关注的事项)。在有帮助的地方引用原话。每个部分都控制在 30 分钟会议能讨论完的范围内。

Why this works

为什么这样有效

Get a better draft

得到更好的初稿

Make Claude work for you

让 Claude 为你所用

Tell Claude your own rules for this task, such as your retro template and what counts as a theme versus a status update, then save what you've done as a skill of your own. A few minutes of conversation and it runs with your standards from then on.

告诉 Claude 你做这项任务的规则,比如你的回顾模板,以及什么算“讨论主题”、什么只算“进度汇报”,然后把这次的做法保存成你自己的技能(Skill)。只需几分钟的对话,之后它就会一直按你的标准来做。

Make what we've done in this task so far into a skill called /sprint-retro.
把我们在这个任务中到目前为止做的事情做成一个名为 /sprint-retro 的技能。

Folder: Platform / retros

文件夹:Platform / retros

Review the skill, and save it if Claude hasn't already.

检查一下这个技能,如果 Claude 还没保存,就把它保存下来。

Tip: tell Claude to edit the skill for you.

提示:可以直接让 Claude 帮你修改这个技能。

Make it repeatable

让它可以重复运行

Run it when the cycle closes

在周期结束时运行

Retro prep shouldn't be your Friday afternoon. Ask Claude to set up a scheduled task, or open Scheduled in the sidebar, and /sprint-retro drafts both docs the week the cycle closes.

准备回顾不该占掉你的周五下午。让 Claude 设置一个定时任务,或者在侧边栏打开 Scheduled(定时任务),/sprint-retro 就会在周期结束的那一周起草好两份文档。

Set up a scheduled task for every Friday at 3pm: if a cycle closed this week, run /sprint-retro on it and the last two weeks of channel history, write the retro and the handoff to Platform/retros/<sprint>/, and draft the handoff post to #oncall-platform for me to approve.
设置一个每周五下午 3 点运行的定时任务:如果本周有周期结束,就针对该周期和最近两周的频道记录运行 /sprint-retro,把回顾文档和交接文档写到 Platform/retros/<sprint>/,并把交接帖子起草到 #oncall-platform,等我批准。

Folder: Platform / retros

文件夹:Platform / retros

Sprint retro + on-call handoff

冲刺回顾 + 值班交接

Scheduled task · Active · Friday at 3:00pm

定时任务 · 活跃 · 每周五下午 3:00

If a cycle closed this week, runs /sprint-retro, writes the retro and the handoff to the dated retro folder, and drafts the handoff post to the on-call channel for you to approve.

如果本周有周期结束,就运行 /sprint-retro,把回顾文档和交接文档写入以日期命名的回顾文件夹,并把交接帖子起草到值班频道,等你批准。

To use a folder on your computer during a scheduled task, keep your computer on and awake with the Claude desktop app open so Claude can access it.

如果定时任务需要用到你电脑上的文件夹,请让电脑保持开机且不休眠,并打开 Claude 桌面应用,这样 Claude 才能访问它。

Share with your teammates

与队友共享

Your /sprint-retro skill now knows your tracker, your channels, your template, and what counts as a theme versus a status update. Share it so every squad runs the same retro prep and on-call handoffs read the same way org-wide.

你的 /sprint-retro 技能现在已经了解你的任务跟踪工具、你的频道、你的模板,以及什么算讨论主题、什么只算进度汇报。把它共享出去,让每个小组都用同样的方式准备回顾,全公司的值班交接文档也保持统一风格。

Share the skill

共享技能

You can share a skill you created. In your skills, open the ... menu next to /sprint-retro, select Share, and pick your teammates. For your whole organization, open the skill and select Publish to org, if your admin allows. Your teammates get the skill with your template and channel defaults baked in, so they don't have to build it again.

你可以共享自己创建的技能。在你的技能列表中,打开 /sprint-retro 旁边的 ... 菜单,选择 Share(共享),然后选择队友。如果要共享给整个组织,在管理员允许的前提下,打开该技能并选择 Publish to org(发布到组织)。队友拿到的技能会自带你的模板和默认频道设置,不用再从头搭建。

What changes for retro day

回顾日会有什么变化

The sprint retro and on-call handoff are drafted from the closed cycle and channel history — ready to review and edit instead of write from scratch.

冲刺回顾和值班交接文档会根据已结束的周期和频道记录自动起草好——你只需审阅和修改,不用从零开始写。

You did this for one sprint. The same approach covers incident postmortems, release notes, and the weekly status summary — each one becomes a skill your team runs the same way.

你这次只是为一个冲刺做了这件事。同样的方法也适用于故障复盘、版本发布说明和每周进度总结——每一项都可以变成团队用同一种方式运行的技能。

Next: Draft the incident postmortem

下一篇:起草故障复盘报告

Tutorial · 教程 · 7 min

Customize Claude

定制 Claude

The set up that makes Claude specific to your work or team.

让 Claude 贴合你的工作或团队的设置方法。

Source / 原文:https://academy.claude.com/tutorials/customize-claude-cowork

Claude Cowork is now just Claude. What you learn here still applies — just ask Claude.

Claude Cowork 现在就是 Claude。您在这里学到的内容依然适用,直接询问 Claude 即可。

Claude runs full tasks across your files and tools. Customizing it once means every task after that starts with your systems connected and your team's process already in place. For the quickstart, see Get started in three steps.

Claude 可以跨你的文件和工具完成整项任务。只要定制一次,之后的每个任务一开始就已连好你的系统、备好团队的工作流程。快速入门请参阅“三步上手”。

Level 1: Your context and tools

第 1 级:你的背景信息和工具

Connectors

连接器(Connectors)

Connectors plug Claude into the systems where your work already is — Slack, Salesforce, Microsoft 365, Jira, your company's internal tools — giving it the context to understand a task and the ability to act on it. With a connector enabled, Claude can read your data and write back: update a ticket, draft a reply, post to a channel, save a file.

连接器把 Claude 接入你日常工作所在的系统——Slack、Salesforce、Microsoft 365、Jira,以及公司内部工具——让它既有理解任务所需的背景信息,又能实际动手处理。启用连接器后,Claude 可以读取你的数据并写回结果:更新工单、起草回复、在频道发帖、保存文件。

Add a connector for each tool you want Claude to use. Authorize once; Claude can then use that tool in any conversation.

为你希望 Claude 使用的每个工具添加一个连接器。只需授权一次,之后 Claude 就能在任何对话中使用该工具。

Permissions. For enterprise users, your admin controls which connectors are available, including Claude in Chrome, and whether each has read-only or write authorization. You can adjust which tools are enabled for a given conversation in the chat bar. More on connector permissions.

权限。对企业用户来说,管理员决定哪些连接器可用(包括 Claude in Chrome),以及每个连接器是只读权限还是可写权限。你可以在聊天栏中调整某个对话启用哪些工具。详见连接器权限说明。

Instructions

指令(Instructions)

Instructions are standing rules you write for how Claude should work — tone, formatting, which sources to check first, conventions to follow. You set them at two levels, and your organization may add a third:

指令是你为 Claude 写下的长期工作规则——语气、格式、优先查看哪些资料来源、要遵循的惯例等。你可以在两个层级设置指令,你所在的组织还可能再加一个层级:

Instructions apply to every task — they're like the background rules Claude follows regardless of what you're working on. A Skill is for one specific kind of task, loads only when relevant, and can be shared with teammates. General rules go in Instructions; a repeatable process goes in a Skill.

指令适用于每一个任务——就像 Claude 无论你在做什么都会遵守的背景规则。技能(Skill)则针对某一类特定任务,只在相关时才加载,而且可以分享给队友。通用规则写进指令;可重复的流程写成技能。

Level 2: Capture your process

第 2 级:把你的流程沉淀下来

Skills

技能(Skills)

Skills are instruction files for Claude to complete specific tasks in a repeatable way. Use skills to encode your personal best practices — as well as your team's expertise — into a repeatable set of instructions.

技能是一种指令文件,让 Claude 以可重复的方式完成特定任务。用技能把你个人的最佳做法——以及团队的专业经验——整理成一套可以反复使用的指令。

To use a skill, name it directly (/skill-name) or describe the task in plain language. Claude can recognize when a skill applies and load it.

使用技能时,可以直接点名(/skill-name),也可以用日常语言描述任务。Claude 能识别何时该用某个技能并自动加载它。

You can also create a skill with Claude in a prompt. Run through a workflow the way you normally would. At the end of the work, ask Claude:

你也可以通过一条提示词和 Claude 一起创建技能。按你平时的方式把某个工作流程走一遍,工作结束时对 Claude 说:

Package what we just did into a skill.
把我们刚才做的事情打包成一个技能。

The built-in skill-creator captures the steps, templates, and source locations so the next run is one prompt. Create your first skill.

内置的 skill-creator 会记录下步骤、模板和资料来源位置,下次运行只需一条提示词。动手创建你的第一个技能吧。

Level 3: Bundle and share

第 3 级:打包与分享

Plugins

插件(Plugins)

Plugins bundle skills and connectors so teammates have what they need to get started.

插件把技能和连接器打包在一起,让队友拿到就能上手。

Diagram: an equation graphic showing Connectors plus Skills equals Plugins, with each labeled as access to systems, best practices for work, and both in a single package.

示意图:一个等式图形,表示“连接器 + 技能 = 插件”,三者分别标注为“访问系统”“工作最佳做法”和“两者合为一个包”。

Connectors (access to your systems and tools) + Skills (best practices for the work) = Plugins (both, in a single package)

连接器(访问你的系统和工具)+ 技能(工作的最佳做法)= 插件(两者合为一个包)

Example plugins

插件示例

Anthropic publishes ready-made plugins for common roles, each pairing the connectors that role relies on with skills for its core workflows:

Anthropic 为常见岗位发布了现成的插件,每个插件都把该岗位依赖的连接器与其核心工作流程所需的技能搭配在一起:

Browse more plugins, or ask Claude to bundle your own connectors and skills into a custom plugin to share with your team. Ask Claude to find plugins for your role.

浏览更多插件,或者让 Claude 把你自己的连接器和技能打包成自定义插件,分享给团队。也可以让 Claude 帮你找适合你岗位的插件。

Where to go from here

接下来做什么

Most people start with connectors and Instructions, write a skill after running the same task a few times, and share it once a teammate asks how to set theirs up the same way. Plugins are how your admin distributes the org's standard set across roles.

大多数人先从连接器和指令开始,同一个任务做过几次后再写成技能,等有队友问怎么照样设置时再把它分享出去。插件则是管理员向各个岗位分发组织标准配置的方式。

Once you're set up, additional features can take you further:

设置好之后,还有更多功能可以帮你更进一步:

For more on working with Claude:

了解更多使用 Claude 的方法:

Tutorial · 教程 · 7 min

Intro to Projects

Projects 入门

Learn how Projects can streamline your workflow with Claude for more organized and efficient conversations.

了解 Projects 如何简化你与 Claude 的工作流程,让对话更有条理、更高效。

Source / 原文:https://academy.claude.com/tutorials/intro-to-projects

This is a video tutorial (about 7 minutes). Watch it on Claude Academy or on YouTube.

这是一个视频教程(约 7 分钟),可在 Claude Academy 或 YouTube 上观看。

Tutorial · 教程 · 10 min

Teach Claude your way of working using skills

用 Skills 教会 Claude 你的工作方式

Learn how Claude Skills let you package proven work approaches so Claude automatically applies them across conversations.

了解 Claude Skills 如何把你验证有效的工作方法打包起来,让 Claude 在各个对话中自动套用。

Source / 原文:https://academy.claude.com/tutorials/teach-claude-your-way-of-working-using-skills

What are skills?

什么是 Skills?

Think about the last time you created something with Claude that turned out really well. Maybe it was a presentation, an analysis, or a report. What made it work? Claude brought intelligence and capability, but you also played a crucial part—you explained your standards, you described the structure that works for your needs, you provided examples showing what you wanted, and you caught and fixed things that weren't quite right.

回想一下上一次你和 Claude 一起做出一份效果特别好的成果——也许是一份演示文稿、一份分析或一份报告。它为什么成功?Claude 贡献了智能和能力,但你也起了关键作用:你讲清了自己的标准,描述了适合你需求的结构,提供了示例说明你想要什么,还发现并修正了不太对的地方。

Instead of having to repeat this process the next time you complete a similar task, Skills let you package what you've learned. Then, whenever the context matches, Claude applies the expertise automatically. You're not explaining the same standards repeatedly. You're not hunting for that document with your guidelines. You're not fixing generic output to match what you know works. Claude recognizes the situation, loads your approach, and applies it.

下次做类似任务时,你不必再把这个过程重复一遍——Skills 让你把学到的经验打包起来。之后只要情境吻合,Claude 就会自动运用这些专业知识。你不用一遍遍解释相同的标准,不用到处翻找写着你规范的那份文档,也不用把千篇一律的输出改成你认可的样子。Claude 会识别出当前情境,加载你的方法,并加以应用。

Creating a skill is an act of description. By the time you've built one, you can name the format a task needs, the tone that fits your voice, and an example of what a good result looks like, and hand all three to Claude in one move.

创建一个 skill,本质上是一次描述。等你建好它时,你已经能说清一项任务需要的格式、符合你风格的语气,以及好结果的范例,并一次性把这三样交给 Claude。

Why skills matter for your work

为什么 Skills 对你的工作很重要

You have tasks where you've figured out what works—a way of structuring reports, a framework for analysis, a style for client communication. But every time you start a new conversation with Claude, you're explaining it all over again or fixing the output to match your standards.

有些任务你已经摸索出了行之有效的做法——一种组织报告的方式、一套分析框架、一种与客户沟通的风格。但每次和 Claude 开启新对话,你都得从头再解释一遍,或者动手修改输出,让它符合你的标准。

Skills solve this. Package your approach once, and Claude applies it automatically in any chat.

Skills 解决了这个问题。把你的方法打包一次,Claude 就会在任何对话中自动应用它。

What you get out of Claude is mostly decided by what you put in. Your format, your standards, and your examples are the most valuable context you can give it, so they're worth writing down once, properly, instead of reconstructing from memory in every chat.

你从 Claude 那里得到什么,很大程度上取决于你给了它什么。你的格式、你的标准和你的示例,是你能提供给它的最有价值的背景信息,所以值得认真地写下来一次,而不是每次对话都凭记忆重新拼凑。

Before skills: You ask Claude to create a quarterly business review. Claude can already build presentations, but making them match your standards means explaining your preferences each time. You could document guidelines in a Project, but they apply only inside that Project.

没有 Skills 时:你让 Claude 制作一份季度业务回顾(QBR)。Claude 本来就会做演示文稿,但要让它符合你的标准,就得每次都解释你的偏好。你可以把规范写进某个 Project(项目),但它们只在那个 Project 里生效。

With skills: After creating a QBR skill, you ask for the same presentation. Just by mentioning a task related to an existing skill’s name or purpose, Claude can recognize it. In this case, Claude will load your QBR skill. If you open the steps Claude took, you’ll see one that loads the skill. Claude automatically applies your brand guidelines and preferred structure, generating a deck that matches your standards from the start.

有了 Skills 后:创建一个 QBR skill 之后,你再要求做同样的演示文稿。只要你提到的任务与某个已有 skill 的名称或用途相关,Claude 就能识别出来。在这个例子里,Claude 会加载你的 QBR skill。展开 Claude 执行的步骤,你会看到其中有一步是加载这个 skill。Claude 会自动应用你的品牌规范和偏好的结构,从一开始就生成符合你标准的演示文稿。

Where Skills fit into how you work with Claude

Skills 在你使用 Claude 的方式中处于什么位置

Projects

Projects(项目)

You might already use Projects for certain work. Projects make sense when context accumulates over time. Think: a product launch with evolving plans, research that builds on previous findings, a campaign that unfolds over weeks. You can upload documents, build conversation history, and everything stays organized in one workspace. But Projects are bounded, and whatever context you upload to a project only exists there.

你可能已经在某些工作中用上了 Projects。当背景信息会随时间不断积累时,Projects 就很合适。比如:计划不断演变的产品发布、在先前发现基础上推进的研究、持续数周的营销活动。你可以上传文档、积累对话记录,所有内容都整齐地放在一个工作空间里。但 Projects 是有边界的,你上传到某个项目的背景信息只存在于那个项目中。

Skills work everywhere. Create a Skill once, and it's available in any conversation—regular chats, inside Projects, across all your work with Claude. Skills activate automatically when you reference them, regardless of where you're working.

Skills 则处处可用。创建一次,它就能在任何对话中使用——普通聊天、Projects 内部,以及你与 Claude 的所有工作中。无论你在哪里工作,只要提到它们,Skills 就会自动启用。

Use Skills for repeatable procedures you want applied automatically—your brand guidelines, your analysis framework, your report format. Use Projects for work that needs accumulated context—a product launch, a research project, an ongoing campaign. Use both together when your work benefits from persistent context and standardized procedures.

对于希望自动套用的可重复流程——你的品牌规范、分析框架、报告格式——用 Skills。对于需要积累背景信息的工作——产品发布、研究项目、持续进行的营销活动——用 Projects。如果你的工作既需要长期保存的背景信息,又需要标准化流程,就两者一起用。

Custom Instructions

自定义指令(Custom Instructions)

You might also use custom instructions. These are preferences about how Claude works with you in general, like "ask clarifying questions before starting" or "keep explanations concise." They apply universally to everything Claude does for you.

你可能也在用自定义指令。它们是关于 Claude 总体上如何与你协作的偏好,比如“开始前先问清楚问题”或“解释要简洁”。它们普遍适用于 Claude 为你做的所有事情。

Unlike custom instructions, Skills can contain far more detail. They can include extensive instructions, complete reference libraries and detailed frameworks that would otherwise clutter custom instructions. Skills are also specific to certain types of work and activate only when relevant. For example, "When creating financial models, use this validation framework" lives in a skill, as it won’t apply to everything, only financial modeling work.

与自定义指令不同,Skills 可以包含多得多的细节。它们可以收录大量说明、完整的参考资料库和详细的框架——这些内容如果放进自定义指令,会让它变得杂乱。Skills 还针对特定类型的工作,只在相关时才启用。例如,“制作财务模型时,使用这套校验框架”就适合放在 skill 里,因为它并不适用于所有事情,只适用于财务建模工作。

Regular prompting

普通提示

When you need Claude to do something once, just prompt well. Explain what you want, provide context, refine the output. This works perfectly for one-off tasks or exploratory work.

当你只需要 Claude 做一次某件事时,写好提示词就行。说明你想要什么,提供背景信息,再逐步打磨输出。这对一次性任务或探索性工作非常合适。

Skills work differently. They are active everywhere you work with Claude. This means you’re not re-uploading materials and re-explaining standards. Skills also combine automatically when work requires multiple areas of expertise, coordinating without you specifying each one.

Skills 的工作方式不同。在你使用 Claude 的任何地方,它们都处于可用状态。这意味着你不必重新上传材料、重新解释标准。当工作需要多个领域的专业知识时,Skills 还会自动组合、相互配合,不需要你一一指定。

Skills also unlock higher quality outputs that would be difficult to achieve through prompting alone. They can package together complete reference libraries, validation frameworks, quality standards, and proven methodologies. This built-in expertise produces results that otherwise would require extensive explanation and rounds of prompting.

Skills 还能带来仅靠提示词难以达到的更高质量输出。它们可以把完整的参考资料库、校验框架、质量标准和经过验证的方法论打包在一起。这些内置的专业知识所产出的结果,否则需要大量解释和多轮提示才能得到。

Recognizing when a skill will help your work

判断什么时候 skill 能帮上你的工作

Skills range from simple formatting rules to complete workflows that pull data from multiple sources, analyze it, and generate reports. Create a skill when you've figured out how you want a task done and want Claude to follow that approach consistently. This applies in different situations:

Skills 既可以是简单的格式规则,也可以是完整的工作流程——从多个来源拉取数据、加以分析并生成报告。当你已经想清楚某项任务该怎么做,并希望 Claude 始终按这种方式来做时,就可以创建一个 skill。这适用于以下几种情况:

You've refined an approach through experience

你已经通过经验打磨出了一套方法

While working with Claude, you've figured out how something should be done, and find yourself explaining it to Claude multiple times. When you want Claude to follow that refined approach automatically rather than you explaining it fresh each time, that's a good skill opportunity.

在与 Claude 合作的过程中,你已经弄清楚某件事该怎么做,并发现自己一再向 Claude 解释它。当你希望 Claude 自动遵循这套成熟的方法,而不是每次都由你重新解释时,就很适合做成一个 skill。

Quality depends on having specific materials

质量取决于是否有特定的材料

Some work needs specific examples, templates, domain knowledge, or assets to meet your standards. Skills become powerful for packaging everything together so quality is consistent.

有些工作需要特定的示例、模板、领域知识或素材才能达到你的标准。这时 Skills 就能大显身手,把所有东西打包在一起,保证质量始终如一。

Setup normally requires assembling multiple pieces

准备工作通常需要拼凑多样东西

Even occasional work might require gathering multiple files or tools, explaining how they work together, coordinating different requirements. Skills eliminate the setup burden—package everything once and have it available whenever that situation arises.

即便是偶尔才做的工作,也可能需要收集多个文件或工具、解释它们如何配合、协调不同的要求。Skills 免去了这些准备负担——把一切打包一次,以后每当遇到这种情况就能直接使用。

Ready to create a skill?

准备好创建 skill 了吗?

Skills are available on Free, Pro, Max, Ultra, Team, and Enterprise plans. They also need Create and edit files (also called Code execution and file creation) turned on, since Skills run in Claude’s secure sandboxed computing environment. On Free, Pro, Max, and Ultra plans, check that this setting is enabled in Settings. On Team and Enterprise plans, an organization Owner controls both this setting and Skills. To find the two switches, an Owner can open Organization settings, search for “Skills” and click the first result.

Skills 适用于 Free、Pro、Max、Ultra、Team 和 Enterprise 套餐。它们还需要开启“创建和编辑文件”(Create and edit files,也叫“代码执行与文件创建”),因为 Skills 运行在 Claude 安全的沙盒计算环境中(沙盒:一个与外界隔离的运行空间)。在 Free、Pro、Max 和 Ultra 套餐中,请在“设置”里确认已启用这项设置。在 Team 和 Enterprise 套餐中,这项设置和 Skills 都由组织所有者(Owner)控制。要找到这两个开关,所有者可以打开“组织设置”,搜索“Skills”,然后点击第一个结果。

To get started, add a pre-built skill. From there, when you describe a task to Claude that references a skill’s name or description, Claude recognizes it and loads the skill automatically.

入门时,可以先添加一个预置的 skill。之后,当你向 Claude 描述的任务涉及某个 skill 的名称或描述时,Claude 就会识别出来并自动加载这个 skill。

Learn how to write your own custom skills, incorporating guidance from our skills best practices to understand the principles that make skills most effective.

学习如何编写你自己的自定义 skill,并结合我们的 skills 最佳实践指南,理解让 skill 发挥最大效果的原则。

You also don’t have to be technical to start creating skills. Try creating skills using Claude. Start by identifying a well-suited task and describing it for Claude to turn into a skill. Claude will build and structure it into a properly formatted skill file.

你也不需要懂技术就能开始创建 skill。试试用 Claude 来创建。先找一个合适的任务,把它描述给 Claude,让它转化成 skill。Claude 会搭建并整理出一个格式规范的 skill 文件。

Practice: turn one task into a skill

练习:把一项任务变成 skill

Pick a task you've explained to Claude more than once. The description you give Claude does most of the work, and three moves carry it:

挑一项你已经向 Claude 解释过不止一次的任务。你给 Claude 的描述承担了大部分工作,而这份描述靠三个动作撑起来:

Then describe the skill to Claude in one message, with your own details in place of the team-update ones:

然后用一条消息把这个 skill 描述给 Claude,把团队周报的细节换成你自己的:

Create a skill for my weekly team update. The format: three sections, wins, blockers, and priorities, each with at most three bullets. The tone: direct and factual, written for my manager to skim. Here's an example of an update that worked well: [paste one you were happy with].
为我的团队周报创建一个 skill。格式:分三个部分——成果、阻碍、优先事项,每部分最多三条要点。语气:直接、基于事实,写给我的经理快速浏览。下面是一份效果很好的周报示例:[粘贴一份你满意的周报]。

Check that your skill holds up

检验你的 skill 是否靠得住

A skill starts saving you time once you've tested it. Open a new conversation and ask for the task the way you normally would, such as "draft this week's team update from these notes". Two checks tell you whether it worked:

skill 经过测试之后,才会真正开始帮你节省时间。打开一个新对话,像平常那样提出任务,比如“根据这些笔记起草本周的团队周报”。通过两项检查就能知道它是否奏效:

The habit worth keeping goes beyond skills themselves: whenever you hand Claude work that matters, name the format, the tone, and an example of what good looks like. When you catch yourself naming the same three things for the same task twice, that's your cue to package them into a skill.

值得保持的习惯不止于 skills 本身:每当你把重要的工作交给 Claude 时,都说明格式、语气,并给出一个好结果的示例。当你发现自己为同一项任务第二次说明这三样东西时,就是该把它们打包成 skill 的信号了。

Additional Resources

更多资源

Getting started

入门

Going deeper

深入了解

Tutorial · 教程 · 15 min

Using LSEG for financial market data analysis

使用 LSEG 进行金融市场数据分析

Set up and use LSEG's financial market data integration with Claude for real-time pricing, analytics, and valuation tools across asset classes.

在 Claude 中设置并使用 LSEG 的金融市场数据集成,获取覆盖多个资产类别的实时报价、分析和估值工具。

Source / 原文:https://academy.claude.com/tutorials/using-lseg-for-financial-market-data-analysis

The LSEG integration provides Claude with access to LSEG’s comprehensive financial market data ecosystem, spanning across asset classes and domains. This article explains how to set up and use LSEG to retrieve real-time market data, analytics, and perform complex financial calculations. The LSEG integration relies upon Claude’s ability to use remote connectors.

LSEG 集成让 Claude 能够访问 LSEG(伦敦证券交易所集团)覆盖各资产类别和领域的完整金融市场数据生态。本文介绍如何设置和使用 LSEG 来获取实时市场数据和分析结果,并完成复杂的金融计算。LSEG 集成依赖于 Claude 使用远程连接器(remote connectors,即让 Claude 连接外部服务的接口)的能力。

What This Integration Provides

这项集成能提供什么

Capabilities

功能

The LSEG integration enables Claude to access institutional-grade market data, analytics, and valuation tools directly into conversational AI workflows.

LSEG 集成让 Claude 能把机构级的市场数据、分析和估值工具直接带入对话式 AI 工作流程中。

How Claude Uses LSEG Data

Claude 如何使用 LSEG 数据

Claude applies LSEG’s market data to support your financial analysis.

Claude 会运用 LSEG 的市场数据来支持你的金融分析。

Setting Up LSEG Integration

设置 LSEG 集成

Contact LSEG to get access to the MCP server and its technical documentation.

联系 LSEG,获取 MCP 服务器(让 Claude 接入外部工具的服务)的访问权限及其技术文档。

For Organization Owners

组织所有者

Navigate to Organization settings > Connectors.

前往“组织设置 > 连接器”(Organization settings > Connectors)。

Use the Add button to browse all available connectors, then find LSEG and add it for your team.

点击“添加”按钮浏览所有可用的连接器,然后找到 LSEG,为你的团队添加它。

For Individual Users

个人用户

Open LSEG in the connectors directory, connect it to Claude, and sign in when asked. If an owner added it earlier by URL, connect it from your connectors instead. Learn about finding and connecting tools.

在连接器目录中打开 LSEG,将它连接到 Claude,并在提示时登录。如果所有者之前已通过 URL 添加了它,那就改为从你的连接器列表中连接。了解如何查找和连接工具。

Common Use Cases

常见使用场景

Currency Exchange Rate Analysis

货币汇率分析

Example input prompt:

示例输入提示词:

What’s the current USD/EUR spot rate, and what are the 3-month and 6-month forward rates? Calculate the implied forward points for each tenor.
当前 USD/EUR 的即期汇率是多少?3 个月和 6 个月的远期汇率是多少?请计算每个期限的隐含远期点数。

When to use: Evaluating currency positions, pricing FX forwards, or analyzing currency basis.

适用场景:评估货币头寸、为外汇远期定价,或分析货币基差。

Tip: Compare spot vs. forward rates to understand market expectations for currency movements.

提示:对比即期汇率与远期汇率,了解市场对汇率走势的预期。

Interest Rate Curve Comparison

利率曲线比较

Example input prompt:

示例输入提示词:

Show me the USD SOFR swap curve. Plot the curve and identify any unusual shapes or inflection points.
给我看 USD SOFR 互换曲线。画出这条曲线,并找出任何异常形态或拐点。

When to use: Assessing interest rate risk, pricing swaps, or understanding monetary policy expectations.

适用场景:评估利率风险、为互换定价,或理解市场对货币政策的预期。

Typical analysis: Compare curves across different currencies or time periods to identify relative value opportunities.

典型分析:比较不同币种或不同时间段的曲线,发现相对价值机会。

Multi-Currency Bond Analysis

多币种债券分析

Example input prompt:

示例输入提示词:

Price the following bonds using their ISIN codes: FR0014012II5 (French government bond) and US91282CNT44 (US Treasury). Compare their yields and calculate the spread.
使用 ISIN 代码为以下债券定价:FR0014012II5(法国国债)和 US91282CNT44(美国国债)。比较它们的收益率并计算利差。

When to use: Portfolio construction, relative value analysis, or credit assessment across sovereign issuers.

适用场景:构建投资组合、进行相对价值分析,或对不同主权发行人进行信用评估。

Note: Ensure proper currency conversion when comparing international bonds.

注意:比较不同国家的债券时,要确保正确进行货币换算。

FX Forward Curve Construction

外汇远期曲线构建

Example input prompt:

示例输入提示词:

For AUD/USD, show me the complete forward curve using AONIA and SOFR conventions. Display the forward points and implied forward rates for standard tenors.
针对 AUD/USD,按照 AONIA 和 SOFR 惯例给我展示完整的远期曲线。列出各标准期限的远期点数和隐含远期汇率。

When to use: Structuring FX derivatives, pricing swaps, or analyzing currency carry strategies.

适用场景:设计外汇衍生品结构、为互换定价,或分析货币套息策略。

Works well with: Custom valuation dates to assess forward pricing at specific points in time.

搭配使用效果更好:自定义估值日期,以评估特定时间点的远期定价。

Cross-Asset Correlation Analysis

跨资产相关性分析

Example input prompt:

示例输入提示词:

Get the USD SOFR curve and EUR ESTR curve, then analyze how they’ve moved relative to each other. Are there any divergences that suggest trading opportunities?
获取 USD SOFR 曲线和 EUR ESTR 曲线,然后分析两者之间的相对走势。是否存在暗示交易机会的背离?

When to use: Identifying cross-currency basis trades or understanding global rate relationships.

适用场景:发掘跨币种基差交易机会,或理解全球利率之间的关系。

Key benefit: LSEG provides consistent data across asset classes for reliable correlation analysis.

主要优势:LSEG 在各资产类别间提供口径一致的数据,让相关性分析更可靠。

Real-Time Market Data for Trading

用于交易的实时市场数据

Example input prompt:

示例输入提示词:

I’m looking at pricing a currency swap. Get me the current EUR/USD spot rate and the EUR ESTR and USD SOFR curves for accurate valuation.
我正在考虑为一笔货币互换定价。请给我当前的 EUR/USD 即期汇率,以及 EUR ESTR 和 USD SOFR 曲线,以便准确估值。

When to use: Pre-trade analysis, pricing verification, or mark-to-market calculations.

适用场景:交易前分析、定价核实,或按市值计价(mark-to-market)计算。

Why it matters: Real-time access ensures your analysis reflects current market conditions.

为什么重要:实时获取数据,能确保你的分析反映当前的市场状况。

Tips for Using LSEG

使用 LSEG 的技巧