doc-writer

作者: microsoft

規劃並撰寫 VS Code 新功能或更新功能的文件。當使用者要求「撰寫文件」、「為……新增文件」、「為……撰寫文件」或……時,一律使用此技能。

npx skills add https://github.com/microsoft/vscode-docs --skill doc-writer

Document a Feature

Help document a new or updated VS Code feature in the project documentation (the docs/ folder) and keep its required navigation metadata current. This skill works in two phases: it first researches the feature and proposes a documentation plan, then implements the changes only after you approve. It is acceptable to conclude that no documentation update is needed.

When to Use

Use this skill whenever the request maps to any of these — do not start editing docs directly without first running Phase 1:

  • The user says "document this", "document this functionality/feature", "add docs for…", "write docs for…", or "update the docs for…".
  • The user pastes a GitHub issue or PR link and asks you to document or address it in the docs.
  • Documenting a new or changed VS Code or GitHub Copilot feature.
  • Turning a GitHub issue or PR into concrete docs/ updates.
  • Auditing whether existing docs need updating after a feature change.

This applies even when the change looks small (a clarification, a few added steps, a single section). Run the plan-first workflow rather than jumping straight to edits.

Do not use this skill for release notes, API reference docs, redirects, image swaps, or pure copy-edits — those are handled by other skills (release-note-writer, content-redirect, frontmatter-description) or direct edits.

Guardrails

  • Docs and sitemap only. Limit content changes to the docs/ folder. When you add an article or change an article URL, update build/sitemap.xml in the same change. Do not update release notes or API docs (api/) unless the user explicitly asks.
  • Public functionality only. Do not add or retain documentation for internal builds, dogfood-only functionality, hidden settings or commands, source-only implementation, or manually settable internal feature flags.
  • Never edit enterprise/policies.md. This file is generated from the enterprise policy definitions in the VS Code source. Edit enterprise/policies-template.md instead, which is used to regenerate policies.md.
  • Screenshots are human work. When a screenshot needs to be added or updated, insert a TODO comment in the doc for a human to capture and insert it later — do not fabricate image references.
  • Style compliance. All writing must follow the docs-writing style guide.

Phase 1 — Research & Propose a Plan (no edits)

Do not modify any files in this phase.

  1. Understand the feature. Read the feature description, issue, or PR provided. If the description is ambiguous or lacks detail, ask clarifying questions before continuing.

  2. Establish the content framing. Identify the primary persona, reader intent, and article purpose by following the content-framing guidance. Infer them from the existing content, documentation journey, source material, and writer's request. If any part of the framing remains ambiguous and different interpretations would change the content, ask the writer to confirm before planning edits.

  3. Check the source if needed. To understand the implementation, inspect the source code in the microsoft/vscode repo. Use the gh CLI for all GitHub interactions (issues, PRs, code). See user memory gh-cli-powershell.md for PowerShell-specific gh patterns.

    Area being documentedPrimary source repo
    Core editor, workbench, debug, terminal, tasks, settings, commands, keybindingsmicrosoft/vscode
    Copilot Chat, inline chat, agent mode, chat tools, chat participants, MCP in chatmicrosoft/vscode-copilot-chat
    Enterprise policiesmicrosoft/vscode (policy definitions)
  4. Verify public availability. Check every feature, setting, and command against the public product channel intended for the article. Public Preview, Experimental, and Insiders functionality qualifies when labeled clearly. Source-code existence or manual configurability is not enough. Inspect registration metadata, product quality gates, feature flags, and the relevant public release. Treat metadata such as included: false as internal unless product evidence shows otherwise. If functionality is not public, do not document it. During an audit or review, propose removing existing coverage.

  5. Identify affected docs. Search the docs/ folder for the pages that need to be created or updated. Map each change to a specific file and section. Include docs/toc.json and build/sitemap.xml in the plan when you add an article or change an article URL.

  6. Present the plan. Summarize:

    • The primary persona, reader intent, and article purpose.
    • Which docs/ files and required navigation metadata you propose to create or change, and a short description of each edit.
    • Any TODO screenshot placeholders that will be needed.
    • Open questions or assumptions.

    If you conclude that no documentation update is needed, say so and ask the user to confirm before closing out.

  7. Stop and wait for approval. Do not proceed to Phase 2 until the user explicitly approves the plan (or adjusts it).

Phase 2 — Confirm the Branch & Implement (after approval)

Once the user approves the plan:

  1. Confirm the branch before editing. Use the user-question tool to ask whether to create a new branch from main before implementing the documentation updates. Offer these choices:

    • Yes, create a new branch from main (Recommended)
    • No, continue on the current branch

    Do not modify documentation files until the user answers. If the user chooses a new branch, ask for a branch name unless they already supplied one, then create the branch from main before editing. If uncommitted changes prevent safe branch creation, stop and explain the conflict instead of stashing or reverting changes. If the user chooses the current branch, continue without changing branches.

  2. Apply the documentation edits exactly as agreed, following the docs-writing style guide and the approved content framing.

  3. Add TODO comments where screenshots need to be captured by a human.

  4. Respect the guardrails above (docs content and required sitemap updates only; no release notes/API docs unless asked; never edit generated policies.md).

  5. Verify that the introduction, main content, examples, and next steps serve the approved primary persona and reader intent.

  6. Summarize the changes you made and call out any remaining TODOs for the user.

Orchestrate your documentation updates

  • Act as a reviewer and coordinator
  • Delegate work to GPT-5.6 Sol subagents, in parallel when feasible.
  • The subagents should not themselves run validation on their work, that's your job! You own the plan; delegate work in parallel when reasonable.
  • Monitor the progress of subagents and integrate their contributions into the final documentation.

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