microsoft/vscode
24 skills14.4k installs4M starsGitHub
Install
npx skills add https://github.com/microsoft/vscodeSkills in this repo
1Fix Errorsfix-errors is an agent skill that teaches solo builders and small teams how to remediate unhandled errors reported through the VS Code error telemetry dashboard. Instead of adding a typeof guard or try/catch at the line in the stack trace, you trace data flow upward through each frame, identify where invalid values are produced, and fix corruption at the source so the same payload cannot break other consumers later. The skill fits extension authors and anyone shipping editor-integrated products who sees recurring hits and user impact in telemetry. Use it when an issue bundles an error message, stack trace, hit count, and affected user count and you need a disciplined alternative to fallback values and masked failures. It pairs naturally with ship-phase hardening and operate-phase monitoring because the same invalid data often predates the visible crash.1.4kinstalls2Azure PipelinesAzure-pipelines is a Microsoft VS Code repo skill for validating Azure DevOps pipeline changes before you commit. Solo maintainers and contributors editing YAML under build/azure-pipelines/ use it to install or verify Azure CLI and the azure-devops extension, authenticate to monacotools/Monaco, queue definition 111 (main build), check run status, and read logs—shortening the push-and-wait CI cycle. Prerequisites include az login and configured defaults for organization and project. The skill is environment-specific to VS Code’s Azure Pipelines but the pattern applies to any team that can target a known definition ID. It complements unit testing skills by exercising the actual release pipeline machinery your users depend on for green builds.1.4kinstalls3AccessibilityAccessibility is Microsoft’s primary agent skill for VS Code UI work. Solo builders and contributors extending the workbench, editor, or extensions should invoke it whenever they add or change interactive surfaces—panels, views, widgets, commands, or dialogs—not only when a ticket mentions a11y. The skill treats accessibility as a default gate for new feature work: if you introduce a rich focusable component, you must supply an Accessibility Help dialog (help keybinding on focus), an Accessible View plain-text presentation when visual content is non-trivial, and a verbosity boolean setting when appropriate. It also guides ARIA announcements, labels, roles, signals, and keyboard paths so screen reader and keyboard-only users stay oriented. For Prism readers shipping editor-adjacent products, this is the procedural checklist that keeps contributions aligned with VS Code’s platform expectations before review or release.1.4kinstalls4Memory Leak AuditMemory Leak Audit encodes the disposable patterns Microsoft uses to prevent the most common class of bugs in VS Code extensions and rich clients. Solo builders shipping Electron apps, editor extensions, or complex React/TypeScript UIs can run the skill when reviewing listeners, DOM handlers, model lifecycle callbacks, or fixing reports of growing listener counts. The workflow is a step-by-step checklist: replace raw DOM hooks with tracked disposables, use Event.once for fire-once semantics, and consolidate cleanup with DisposableStore when setup runs more than once. It is most often invoked at Ship during code review, but the same rules apply while you are still building feature code that registers events in hot paths. Intermediate complexity—you need familiarity with TypeScript disposables and component/widget lifecycles.1.4kinstalls5Sessionssessions is an agent skill for contributors working inside Microsoft VS Code’s Agents window (`src/vs/sessions/`). It forces a disciplined workflow: read organization and coding guideline instructions, then the area-specific spec (layers, layout, chat widget, menus, contributions, or entry points) before writing code. The architecture is agents-first, with enforced internal layers and a documented grid layout so chat and auxiliary UI stay consistent. Solo builders maintaining a fork or contributing agent UI features use it to avoid cross-layer import violations and drift between code and specs. Complexity is advanced because it assumes TypeScript DI patterns, disposable management, and localized strings per repo conventions.1.3kinstalls6HygieneHygiene is a Microsoft VS Code agent skill that teaches your coding agent how to run the same hygiene pass the repository uses as a git pre-commit hook. Solo builders and small teams shipping in the VS Code codebase use it when they are about to stage and commit TypeScript or web assets and need confidence the change will not be rejected at commit time. The workflow centers on `npm run precommit`, which executes `node --experimental-strip-types build/hygiene.ts` against staged files, with an escape hatch to lint explicit paths before staging. Covered rules span unicode restrictions, quoting conventions, copyright headers, tab-only indentation, formatting, ESLint, and stylelint, matching what maintainers expect in upstream contributions. It is intermediate complexity because you must understand staging, suppress comments for rare unicode cases, and interpret linter output. Treat it as mandatory before declaring work complete on this monorepo—not a substitute for broader test suites or security review.1.3kinstalls7Author ContributionsAuthor-contributions is a Microsoft VS Code repo agent skill that traces every file a specific person changed on a branch compared to its upstream, following renames so moved code still counts as their work. Solo builders and small teams install it when they need a fast authorship map before merge—who owns which diffs, what came through renames, and how many lines moved—without manually chaining git log and diff commands. The procedure insists on exact author strings from git history, walks commits in the upstream..branch range, and formats results as a markdown table with Status, File Path, and Lines (+/-) plus a summary line. Because it performs heavy git work, SKILL.md directs running it as a subagent with a tight prompt. It fits ship-time review and accountability checks, and it also helps operate-time forensics when you need to understand contribution boundaries on a long-lived branch.1.3kinstalls8Tool Rename Deprecationtool-rename-deprecation is a Microsoft VS Code agent skill that enforces backward compatibility when built-in language model tools or tool sets change names. Any rename of toolReferenceName or a tool set referenceName—or a move that changes the toolSet/toolName path—must append the previous identifier to the correct legacy array so prompt files, tool configurations, and saved references keep resolving. The skill documents entity-specific registration surfaces (TypeScript IToolData versus extension package.json languageModelTools), stresses that stable ids must not change, and applies equally when reviewing pull requests that touch registration code. Extension and agent-tool authors use it as a procedural guardrail so Copilot-style tool references do not silently break across VS Code updates. It is checker-oriented: identify what changed, verify legacy arrays, and treat missing deprecations as regressions.1.3kinstalls9Update ScreenshotsUpdate Screenshots is a Microsoft VS Code agent skill for maintaining confidence in component-level visual tests when screenshot baselines are hosted externally rather than checked into git. Solo extension maintainers and small teams hit this when a PR triggers the screenshot-test GitHub Action and the Checking Component Screenshots workflow surfaces diffs. The skill clarifies that you no longer download baselines into test/componentFixtures or commit LFS images—instead you read CI comments, pull the screenshots artifact with the GitHub CLI, and compare captures locally if intent versus regression is unclear. After merge to main, the service treats that commit’s images as the new baseline. Use it in ship and review loops so agents do not repeat obsolete “commit baseline folder” steps that the repository has deliberately retired.1.3kinstalls10Agent Sessions Layoutagent-sessions-layout is a procedural skill for contributors working inside the VS Code repo on the Agents workbench shell. It tells your coding agent to treat LAYOUT.md as the single source of truth for grid structure, which parts exist, default visibility, titlebar configuration, and editor modal behavior before touching implementation files. The workflow is opinionated about parity: any change to positions, sizing, menus, or CSS must be mirrored in the spec and logged in the revision table so future fixes do not drift from documented behavior. Solo builders are rarely the audience unless they maintain a VS Code fork or internal editor distribution; for them it removes guesswork when debugging broken agent panels or sidebar footer layout. Use it when implementing new agent-session UI affordances or fixing regressions in the sessions workbench rather than for general extension theming.914installs11LaunchLaunch is a Microsoft VS Code repo skill implemented as a bash script that solo contributors use to spin up Code OSS without trashing their daily driver profile. It clones or slims an authenticated user-data directory so Copilot and GitHub sign-in survive on macOS keychain-backed setups, while forcing a separate shared-data dir so two OSS windows do not crash each other. Callers get structured JSON on stdout—ports for renderer CDP, extension host, main process, and optional agent host—so automated tests or agent loops can attach cleanly. Flags trade startup time against extension fidelity: default empty extensions dir is fastest; --clone-extensions or --full recover third-party stacks when needed. Advanced complexity: you must have a real vscode checkout, understand dev auth env vars, and accept stderr-only human logs. Narrow audience—VS Code from-source hackers—not general “open my project in VS Code” workflows.199installs12Component FixturesComponent-fixtures is an agent skill for Microsoft VS Code contributors who need isolated workbench UI renders for screenshot testing. Fixtures live under src/vs/workbench/test/browser/componentFixtures/ and pair with the Component Explorer Server task and MCP tools when available. Solo builders extending VS Code or large web workbench surfaces use it when designing components to be fixture-friendly—correct file endings, themed groups, editor services, and scoping rules. It bridges implementation and visual QA: you author .fixture.ts files during frontend work, then validate appearance in Ship without booting the full product shell every time.186installs13Add Policyadd-policy is an advanced workflow skill for contributors and maintainers working in the VS Code repository who need to add, modify, or review configuration policies. Enterprise admins lock settings through Windows Group Policy, macOS managed preferences, Linux policy files, or Copilot account-level policy data; this skill documents how those layers connect in code—from NativePolicyService and FilePolicyService through AccountPolicyService and MultiplexPolicyService. Solo builders shipping editor forks, OSS enterprise builds, or internal VS Code distributions use it whenever a configuration property gains a `policy:` field so registration, export, and platform artifacts stay consistent. It is not a generic app-settings tutorial; it is the canonical procedure for policy registration and review in the VS Code policy architecture.164installs14Unit Testsunit-tests teaches agents and contributors how to run unit tests inside the microsoft/vscode repository without guessing entrypoints. Prefer the runTests tool when available for structured pass/fail output, file lists, test name filters, and coverage mode; otherwise use platform scripts from the repo root that bootstrap Electron and launch Mocha. The skill covers common patterns such as passing absolute test file paths and filtering by test name—critical for a monorepo where wrong flags waste long CI-like local runs. Intermediate complexity reflects Electron-backed runner setup and VS Code-specific layout. Not a generic Jest/Vitest guide—it is scoped to this codebase’s official harness.141installs15Chat Customizations Editorchat-customizations-editor orients contributors to the VS Code pane where users manage agents, skills, instructions, hooks, prompts, MCP servers, and plugins across workspace, user, extension, and plugin scopes. Solo builders shipping on VS Code’s AI stack—or maintaining a fork—need the harness model (IHarnessDescriptor, ISectionOverride) so UI changes stay consistent between workbench and sessions windows. The skill maps key folders under chat/common, aiCustomization, and sessions/contrib, and insists the AI_CUSTOMIZATIONS spec stays the single source of truth. Intermediate complexity reflects TypeScript DI, duplicate registration sites, and optional Playwright runs. It is for codebase contributors, not end users who only install skills from a marketplace.132installs16Fix Ci FailuresFix CI Failures is a procedural agent skill from the VS Code ecosystem that walks solo builders through red GitHub Actions on a pull request without tab-hopping guesswork. You start from the branch you already have locally, resolve the associated PR, and inspect statusCheckRollup style failures through the GitHub CLI. The skill emphasizes extracting run IDs from job URLs, downloading logs for failed jobs, and translating noisy CI output into a single actionable defect. That maps directly to the Ship phase where lint, test, typecheck, and packaging pipelines protect main. It is not a substitute for designing CI from scratch; it is the tactical loop when a change you believed was ready suddenly fails remote-only assumptions. Keep gh installed and authenticated, iterate fix-push-watch, and use it alongside code review skills when failures are flaky or environment-specific.125installs17Integration Testsintegration-tests is a contributor-focused agent skill for the Microsoft VS Code repository. It tells solo builders and maintainers exactly how to run integration tests through the official shell and batch wrappers, and when those wrappers run only Node integration tests versus full extension-host launches. The skill clarifies that unfiltered runs execute all Node integration tests followed by every extension host suite, while node-specific filters intentionally skip extension host work. It also documents how --grep propagates to every runner when no narrower filter is set. Use it when you are validating changes locally or debugging CI failures tied to Mocha integration files or built-in extension API tests. It reduces wrong commands and wasted full VS Code startups during tight iteration loops.103installs18Heap Snapshot AnalysisHeap Snapshot Analysis is a Microsoft VS Code repository skill that compares paired V8 heap snapshots and reports where memory grew between runs. Solo builders shipping editor extensions, Electron shells, or other Chromium/V8 hosts can use it to see summary node counts, total size deltas, top constructors by size and count increase, entirely new object groups, and VS Code–oriented class changes. It fits late Build or Ship when a feature branch suddenly balloons RAM during manual or automated runs. You typically capture snapshots from Chrome DevTools or VS Code’s profiling workflow, then let the skill diff structured snapshot data. It does not replace full production APM; it accelerates local root-cause analysis before you merge a leak.80installs19Auto Perf OptimizeAuto Perf Optimize is a Microsoft VS Code ecosystem skill for extension and agent-ui builders who need evidence-backed performance fixes—not gut feel. It documents how to spin one-off Playwright scenario runners in a dated scratchpad, capture findings with ideas tried and rejected, and reuse generic smoke scripts for chat memory and session switching. Solo builders shipping Copilot-style chat in an editor hit regressions here: scroll leaks, heap growth across turns, slow tab switches. The skill is multi-phase because you can run the same pattern while building new chat features, before release perf gates, or when operating on user-reported slowness—but the canonical shelf is Ship → perf. It pairs procedural knowledge with concrete CLI flags (--message, --iterations, --reuse) so agents do not invent incompatible harnesses.79installs20Cpu Profile Analysiscpu-profile-analysis is an agent skill bundled with Microsoft’s VS Code skill set for reading Chrome and V8 performance artifacts. Solo builders optimizing editors, Electron apps, or web UIs use it when a .cpuprofile from the VS Code profiler or a Trace-*.json export from DevTools needs interpretation: which functions dominate samples, how call trees differ between implementations, and whether trace events show expensive layout, paint, or long tasks. It is tagged multi-phase with a primary Ship perf shelf because the usual trigger is shipping or hardening performance, but the same analysis supports Operate when production feels slow and Build when refactoring hot paths during development. The skill is read-only on profile files and emphasizes structured comparison and bottleneck narration rather than running new benchmarks. Review Security Audits on Prism before pointing agents at traces that may contain URLs or user paths from local sessions.72installs21Chat PerfChat Perf is a VS Code repository skill for running chat performance regressions and memory leak checks against local dev builds or pinned releases. It targets contributors changing chatListRenderer, chatInputPart, markdown rendering, streaming SSE handling, or disposable lifecycle code. Quick starts use npm run perf:chat with scenarios such as text-only, optional --no-baseline measurement, apples-to-apples comparisons between two builds, and production-build baselines against release versions like 1.115.0. A separate leak harness sends multiple chat messages in one session to surface retained disposables. Solo maintainers of Copilot-style chat forks or VS Code extensions can treat it as a Ship-phase perf gate before merge.65installs22Vscode Dev Workbenchvscode-dev-workbench teaches agents how to run Microsoft’s vscode.dev server from the vscode-dev checkout so you can hit the VS Code web workbench—or the Agents window—from a local browser while reading sources from a sibling microsoft/vscode tree. It targets contributors and solo builders extending VS Code for the web who need a tight feedback loop on UI and agent-surface behavior. The skill stresses layout assumptions, the correct npm run dev working directory, the vscode-quality=dev query parameter, and patterns for browser-driven testing. Optionally you wire a mock agent host when exercising the Agents window. This is narrow, high-signal tooling: wrong repo root or cwd silently leaves you in the wrong package and breaks the dev script. For Prism’s audience it matters when your product is the editor itself or a web-only extension path, not when you only use VS Code as a generic coding agent host.59installs23OtelThe otel skill is a procedural guide for Microsoft Copilot Chat extension developers who need consistent OpenTelemetry across agent runs. It directs you to read the user-facing and architecture monitoring specs before touching instrumentation, then apply shared conventions for spans, metrics, and events across the documented execution paths. Solo builders shipping agent features in VS Code use it when adding a new agent surface, changing the Copilot CLI bridge, adjusting Claude-related span emission, or updating exports described in agent_monitoring*.md. The skill emphasizes parity between implementation and documentation so operators can configure backends from the user doc while contributors follow the arch spec and HTML flow diagram. It is narrow in scope—Copilot extension OTel—not generic app observability, but it prevents drift between telemetry code and the published monitoring contract.50installs24Code Oss LogsCode-oss-logs teaches coding agents how to find and read logs from Code OSS and the Agents app during local development. Indie extension and editor hackers hit opaque failures in renderer, extension host, agent host, or main process threads; this skill standardizes where logs live, how to pick the latest run folder, and when a custom user-data-dir overrides defaults. It explicitly covers launch scenarios that stash data under .build/ so agents do not read stale directories. Use it whenever you are debugging a dev build rather than production telemetry—you get a repeatable shell-first procedure instead of guessing paths under your home directory.40installs