write-my-release-notes

作者: microsoft

為你在里程碑中開發的功能生成 VS Code 發行說明。當使用者要求「為我的…撰寫發行說明」時,一律使用此技能。

npx skills add https://github.com/microsoft/vscode-docs --skill write-my-release-notes

VS Code Release Features Writer

This skill turns the features you worked on in a VS Code milestone into polished, user-focused release note sections. It gathers the features you own from GitHub using the gh CLI, then documents every one of them in the milestone's release notes file.

Use this skill when the user wants to draft or update the feature sections of a milestone's release notes based on the issues and test plan items they worked on.

Prerequisites

  • Prefer the GitHub MCP server for all GitHub queries when it is available. Use its issue search and issue read capabilities to run the queries and read the bodies, labels, and comments described below.
  • Fall back to the gh CLI only when the GitHub MCP server is not available. In that case, ensure it is authenticated (gh auth status); the gh commands below implement the same queries.
  • Queries use @me, so results are scoped to the authenticated user.

Input parameters

  • Milestone name: If the user does not provide a milestone name, ask the user for it. Do not continue until you have the milestone name.

Locate the release notes file

  1. Confirm the milestone name with the user.
  2. Find the release notes file under the /release-notes folder. The file name does not match the milestone name, so search the folder contents for the text TOCTitle: <milestone name> (for example, TOCTitle: April 2016).
  3. If no file exists for the milestone, create one using the appropriate template in /templates and a filename of the form v<version>.md (for example, v1_109.md).

Identify features to document

Gather the features the user owns by running the following GitHub search queries. Use the GitHub MCP server if available; otherwise use the gh CLI commands shown below. Substitute <milestone> with the milestone name (keep the surrounding quotes). Collect each issue's number, title, body, url, labels, and comments.

Step 0 — Consult prior context (optional).

If you have persistent memory, notes, or prior context about this milestone (coverage decisions, features you own that the label queries won't surface, work intentionally excluded), review it before the GitHub queries and reconcile against the results. Skip if you have none.

Step 1 — Feature-request issues assigned to you

gh search issues 'org:microsoft milestone:"<milestone>" label:feature-request assignee:@me' \
  --limit 100 --json number,title,body,url,labels,repository

Process the results:

  • Skip any issue labeled *duplicate or *out-of-scope.
  • If an issue is labeled on-testplan, set it aside — do not document it directly. It will be attached as a related issue to its test plan item in Step 2.
  • All remaining issues are release features to document.

Step 2 — Test plan items authored by you

gh search issues 'repo:microsoft/vscode milestone:"<milestone>" label:testplan-item author:@me' \
  --limit 100 --json number,title,body,url,labels

Each test plan item is also a release feature to document. List the test plan item features first, before the feature-request features from Step 1. For each test plan item, collect related issues that provide the real feature detail:

  • Issues filed against the test plan item — closed issues that reference it during testing:

    gh search issues 'repo:microsoft/vscode is:closed Testing <testplan-item-number>' \
      --limit 100 --json number,title,body,url,labels
    
  • Set-aside on-testplan issues from Step 1 whose issue number appears in the test plan item's body — attach these as related issues too.

Step 2.5 — Reconcile against all your milestone work (catch mislabeled features)

The label queries in Steps 1 and 2 only find issues labeled feature-request or testplan-item. Real features are often tracked under other labels during the endgame — for example a feature that shipped a fix may carry bug, verification-needed, or polish instead. Do not trust the label queries to be exhaustive. Run these broader queries and reconcile the results against what Steps 1 and 2 already surfaced:

gh search issues --owner microsoft --assignee @me --milestone "<milestone>" \
  --limit 100 --json number,title,url,labels,state
gh search issues --owner microsoft --author @me --milestone "<milestone>" \
  --limit 100 --json number,title,url,labels,state

For every issue these return that is not already covered by Steps 1 and 2, judge it on its content, not its label:

  • If it describes a user-facing capability (a new command, setting, UI, deep link, workflow), document it as a feature — regardless of whether it is labeled bug, verification-needed, polish, or something else. A "bug" that adds or restores a capability users will notice is a release feature.
  • If it is a pure defect fix, internal cleanup, or minor polish with no user-facing capability, leave it out of the feature sections (it belongs in "Notable fixes" at most).
  • When in doubt, read the body and comments (Step 3) before deciding, and ask the user whether to include it.

If you have persistent memory or prior context about this milestone (Step 0), cross-check each remembered feature you own against the combined results here — a feature noted in memory but missing from the query output is exactly the kind of item that slips through mislabeling.

Step 3 — Fetch comments for each issue

The search queries above may not return comments (the gh search issues --json output does not include them). For every feature and related issue you plan to use, read the full body and comments for context — use the GitHub MCP server's issue read capability if available, otherwise the gh CLI:

gh issue view <number> --repo <owner/repo> --json number,title,body,url,labels,comments

Document every feature gathered from Steps 1, 2, and 2.5 — none should be skipped. Most features are identified by one of these labels:

  • feature-request — a standard feature request issue. Its description and comments contain the feature details.
  • testplan-item — a structured test plan for the feature. Its description contains in-depth details plus setup and test steps.

Some features surface only through Step 2.5 under other labels (for example bug, verification-needed, or polish). Treat those as features too when they describe a user-facing capability — the label does not disqualify them.

Use the related issues (from Step 2) to pull in additional context (their summary, description, and comments).

Build each feature section

Gather and analyze all relevant details before writing:

  • The summary, description, and comments of the feature itself.
  • The summary, description, and comments of every related issue.

Then write a comprehensive, user-focused section for each feature:

  • Feature title
    • Use a concise, descriptive title that identifies the feature. Use sentence case.
    • Do not take the title from a testplan-item, because that title is written for testing.
    • If the feature is in Preview or Experimental, include that status in the heading, for example ### New JavaScript debugging experience (Preview).
  • Feature description
    • Do not add a "Feature Description" header — start directly with the prose under the heading.
    • Explain the feature, its purpose, and any constraints users should be aware of.
    • Never include testing content: no setup, test instructions, or validation steps.
  • Value proposition — every section must answer "Why should the user care?"
    • Lead with the user benefit or the problem being solved, then explain the mechanics. State the pain point first, then the resolution.
    • Avoid describing only what changed. A mechanism ("buffer-based rendering") is not a benefit ("chat responses feel more fluid").
    • Link to relevant VS Code documentation when referencing tools, APIs, or concepts that are not common developer knowledge. Expand inline when docs coverage is thin.
    • For admin and policy features, cover both the administrative use case (compliance, security, cost control) and the developer-productivity impact.
  • Feature continuity for multi-release features
    • For features that span multiple releases (for example, Agents, Customization UI, Sandboxing), add a 1-2 sentence reminder of what the feature is and its goal before describing what changed this release.
    • The first release of a feature should lead with benefit-oriented bullets. Later releases should still re-establish context.

Writing guidelines

Follow the release notes writing instructions first, then the VS Code documentation writing guidelines. Key rules:

  • Use sentence case for headings; do not apply italic, bold, or inline code styling to headings.
  • Use absolute documentation links that start with https://code.visualstudio.com/docs/ and omit the .md suffix. Use descriptive link text — never "click here", "this link", or "here".
  • Image alt text must describe the content and start with "Screenshot showing" or "Screenshot of". Video alt text must start with "Video showing" or "Video of" and use the <video> tag: <video src="[URL]" title="[description]." autoplay loop controls muted></video>.
  • Format notes and tips as block quotes with a bold Note or Tip prefix. Use asterisks for list items.
  • Reference settings with the setting(setting.name) format and keyboard shortcuts with the kb(command.commandId) format.
  • Avoid marketing language and the word "now". State what changed directly and prefer concrete before/after examples over vague claims like "improved support".
  • Group features that share a theme (for example, several agent improvements) and lead with the most impactful one.

Verify technical details

  • Only document settings and commands that actually exist. Verify setting names, values, and command IDs before referencing them.
  • Fact-check all technical details and confirm every link and reference resolves.
  • Keep terminology and structure consistent with existing release notes.

Writing to the file

  • Write only the feature documentation sections. Do not add version headers, welcome messages, or update summaries.
  • Start directly with ### feature headings.
  • Do not leave TODO placeholders.
  • Save the changes to the milestone release notes file.

來自 microsoft 的更多技能

oss-growth
microsoft
開源增長駭客角色
agent-framework-azure-ai-py
microsoft
使用Microsoft Agent Framework Python SDK(agent-framework-azure-ai)构建Azure AI Foundry代理。适用于使用AzureAIAgentsProvider创建持久化代理、使用托管工具(代码解释器、文件搜索、网络搜索)、集成MCP服务器、管理对话线程或实现流式响应。涵盖函数工具、结构化输出和多工具代理。
development
airunway-aks-setup
microsoft
在AKS上設定AI Runway——從裸叢集到執行模型。涵蓋叢集驗證、控制器安裝、GPU評估、供應商設定及首次部署。時機:「設定AI Runway」、「上線AKS叢集」、「安裝AI Runway」、「airunway設定」、「部署模型至AKS」、「在AKS上進行GPU推論」、「在AKS上設定KAITO」、「在AKS上執行LLM」、「在AKS上使用vLLM」、「在AKS上設定模型服務」、「AI Runway控制器」。
devops
appinsights-instrumentation
microsoft
使用Azure Application Insights檢測Web應用程式的指南。提供遙測模式、SDK設定與組態參考。適用時機:如何檢測應用程式、App Insights SDK、遙測模式、什麼是App Insights、Application Insights指南、檢測範例、APM最佳實踐。
devops
applicationinsights-web-ts
microsoft
使用Application Insights JavaScript SDK(@microsoft/applicationinsights-web)為瀏覽器/Web應用程式進行檢測。適用於真實使用者監控(RUM)——頁面檢視、點擊、AJAX/fetch依賴、例外、自訂事件,以及與後端OpenTelemetry追蹤關聯的瀏覽器端GenAI代理追蹤。涵蓋SDK載入器指令碼與npm設定、框架擴充(React、React Native、Angular)、點擊分析、遙測初始化器,以及從瀏覽器發出的代理/工具/模型span的OTel GenAI語意慣例。
devops
azure-ai-anomalydetector-java
microsoft
使用適用於 Java 的 Azure AI 異常偵測器 SDK 建置異常偵測應用程式。在實作單變量/多變量異常偵測、時間序列分析或 AI 驅動監控時使用。
development
azure-ai-language-conversations-py
microsoft
使用 azure-ai-language-conversations Python SDK 實作對話語言理解(CLU)。當使用 ConversationAnalysisClient 分析對話意圖與實體、建置 NLP 功能,或將語言理解整合至應用程式時使用。
development
azure-ai-ml-py
microsoft
Azure Machine Learning SDK v2 for Python。用於機器學習工作區、作業、模型、資料集、計算資源與管線。 觸發詞:「azure-ai-ml」、「MLClient」、「workspace」、「model registry」、「training jobs」、「datasets」。
development