REPO PACKAGE
microsoft vscode
OFFICIAL49skills
23ready to use
LABSCO SUMMARY
The skills split by what part of "building VS Code" they cover: contributor git workflow (commit, sync, merge, create-pr, create-draft-pr, fix-ci, code-review), engine-level debugging (heap-snapshot-analysis and cpu-profile-analysis for V8 profiles, memory-leak-audit for disposable and listener leaks, chat-perf for chat-pipeline regressions), the project's own test suites (smoke-tests, integration-tests, auto-perf-optimize), and a cluster for authoring the very customization files VS Code's Copilot integration reads (create-agent, create-hook, create-instructions, create-prompt, create-skill, agent-customization). A smaller group is infrastructure-specific: azure-pipelines for the build's Azure DevOps pipeline, and accessibility, which is described as required reading for any new UI work.
This is for someone building Code - OSS from source and working inside its own contribution process — filing PRs against microsoft/vscode, chasing a flaky smoke test, or debugging a renderer memory leak — not for a VS Code user customizing their own editor. We have descriptions for 40 of the 54 live skills; the other 14 sit outside our sample and are not characterized here. Of the 54, 23 need nothing beyond the agent, 19 assume a local tool (most plausibly a source build of Code OSS itself), 5 need an account credential, 6 could not be verified, and 1 talks to an MCP service.
READ THE FULL ANALYSIS
Two skills look like the same skill twice. project-setup-info-context7 and project-setup-info-local describe themselves in nearly identical wording — full project scaffolding, not single-file edits — differing mainly in which backend (a Context7 MCP tool, or a local equivalent) does the work. We read both; we did not find a third variant that explains why they aren't one skill with a fallback.
ALSO IN THIS PACKAGE
AGENT PERSONALITIES
2 itemsAGENT PERSONALITY
Data
Answer telemetry questions with data queries using Kusto Query Language (KQL)
AGENT PERSONALITY
Demonstrate
Agent for demonstrating VS Code features
PROMPT TEMPLATES
8 itemsPROMPT TEMPLATE
updateCopilotCLIToolMapping
Update the mapping of Copilot CLI tools from the source code for the CLI runtime
PROMPT TEMPLATE
chronicle:cost-tips
Get personalized tips to reduce token usage and Copilot cost
PROMPT TEMPLATE
chronicle:improve
Improve agent instructions based on friction patterns in your session history
PROMPT TEMPLATE
chronicle:reindex
Rebuild the local session index and sync to cloud
PROMPT TEMPLATE
chronicle:search
Search recent chat sessions by keyword, file path, or PR/issue ref
PROMPT TEMPLATE
chronicle:standup
Generate a standup report from recent chat sessions
PROMPT TEMPLATE
chronicle:tips
Get personalized tips based on your chat session usage patterns
PROMPT TEMPLATE
plan
Research and plan with the Plan agent
MCP SERVERS
1 itemMCP SERVER
vscode-automation-mcp
stdio · npm --prefix test/mcp run start-stdioWHAT'S INSIDE
49 showing · 49 totalReady to use
Nothing else to set up — install it and go.
49 of 49
SETUP
Showing all 49 skills
accessibility
The rules VS Code's own team follows so a new part of the editor works for people who navigate by keyboard alone or listen to the screen instead of seeing it.
act-on-feedback
You leave notes on the changes the assistant made; this is what sends it back through them, doing what each note asked and ticking it off.
agent-customization
Teaches the AI assistant in VS Code your project's rules and your own custom commands, by picking which kind of setup file you need and putting it where the assistant will actually find it.
author-contributions
A list of every file one person worked on in a set of changes, including the ones they wrote and somebody else later renamed, which a plain search would miss.
auto-perf-optimize
Puts VS Code through the same steps over and over while measuring its memory, to find out why it gets slower or heavier the longer it runs, and to prove the fix worked afterwards.
azure-pipelines
Tests a change to VS Code's build setup by starting a real build from your own machine, instead of committing the change and waiting to find out it was wrong.
chat-customizations-editor
Changing the screen inside VS Code where people manage their AI helpers — agents, skills, prompts, plugins — without breaking the other places the same information has to be kept in step.
chat-perf
Measures how fast VS Code's chat feels and how much memory it holds on to, so a change to it can be judged on numbers rather than on impressions.
chronicle
Digs through the record of your own past Copilot sessions stored on your machine, and answers questions about your work that you would otherwise have to remember yourself.
code-oss-logs
Where a development build of VS Code hides its log files, and which of the dozens it writes is the one worth opening.
code-review
A review of the code you just changed, delivered as notes attached to the exact lines they concern rather than as one lump of feedback at the end.
commit
Writes the commit message for you — the note that goes with a saved change — in whatever style the project already uses, by reading its own recent history first.
component-fixtures
Shows one piece of VS Code's interface on its own, outside the running app, so a picture of it can be taken and compared every time the code around it changes.
cpu-profile-analysis
Turns the unreadable recording a performance profiler spits out into a plain answer about where a program's time actually went.
create-agent
Hands one kind of job to a specialist — a version of the AI kept to a single role, built out of the way you have already been steering this conversation.
create-draft-pr
Puts the work from this conversation up on GitHub as a draft: visible to the team, but flagged as unfinished, so nobody spends a review on it yet.
create-hook
Some rules should not be advice the AI can ignore — this turns one into an automatic check that runs at a set moment and can stop the action outright.
create-instructions
A rule you keep having to repeat to the AI, written down once so that it holds in every conversation about this project from now on.
create-pr
Opens a pull request for the work done in this conversation, so you do not have to leave the editor, write it up and push it yourself.
create-prompt
For a request you type again and again, a saved copy you can call up by name instead of writing the whole thing out each time.
create-skill
Packages a job that takes several steps into something the AI can run from start to finish on its own, the same way every time.
fix-ci
Reads the automatic checks that failed on your change, works out what actually broke, and fixes the code until they pass.
fix-ci-failures
Digs into a failed build on GitHub: which check went red, what its logs actually say once you get hold of them, and what to change so it passes next time.
generate-run-commands
Works out how your project is normally started — install this, then run that — and puts those commands behind a Run button so nobody has to remember them.
get-search-view-results
Hands the assistant whatever matches are already showing in the editor's search panel, instead of it searching the project all over again.
github-copilot-upgrader
Runs the Copilot library upgrade as a checklist you can leave and come back to: snapshot the old version, update, compare what changed, then compile and test until nothing fails.
heap-snapshot-analysis
Finds out why a program's memory keeps growing: which things were never thrown away, and what is still holding on to them.
init
Reads a project it has never seen before and writes down what an AI assistant needs to know about it — how to build it, what the conventions are, where the traps are.
install-vscode-extension
Installs an add-on into Visual Studio Code by name, so you do not have to go and find it in the marketplace yourself.
integrated-browser
The mental model behind VS Code's built-in web browser: why the page floats on top of the editor rather than sitting inside it, and why that one fact decides where new code has to live.
integration-tests
VS Code's slower tests — the ones that boot a real editor and real extensions to check the pieces work together — and which of the two scripts runs which half of them.
launch
When you are working on VS Code itself, this starts a throwaway copy built from your source and already signed in. Each copy takes its own ports, so a test script can drive one while a debugger is attached and others run alongside.
memory-leak-audit
The checklist VS Code's team runs over new code to catch the handful of mistakes behind most of its memory leaks, the commonest being a listener that is never unhooked.
merge
Folds the work from this conversation back into the main line of the project, stopping to sort out anything that clashes on the way.
otel
VS Code records what its AI agents do in two separate systems that are easy to mix up; this says which one you are actually in before you change anything.
project-setup-info-context7
Looks up how to start a new project in whichever framework you named, fetching current instructions from an outside documentation service rather than working from memory.
project-setup-info-local
Starts a brand-new project from nothing, using the official starter command for the kind of thing you are building rather than a hand-made folder of files.
sessions
VS Code's Agents Window has a dozen design documents and a handful of hard boundaries; this works out which of them your change actually has to obey.
smoke-tests
Drives a real copy of VS Code through the things a user would actually do, and has a trick for catching the test that only fails once in twenty runs.
sync
Brings your branch level with the shared copy and pushes your own commits up — or, if the branch has never left your machine, puts it on the server for the first time.
sync-upstream
When a branch has fallen a long way behind, this replays your work on top of the latest code and bends your changes to fit it, never the other way round.
troubleshoot
Answers the question of why the assistant did what it did, by reading the detailed record the conversation left behind instead of guessing from the screen.
unit-tests
The quick tests that run in seconds, cut down to the single file or the single behaviour you are working on right now.
update-pr
A pull request you opened earlier, brought back in line with the work that has happened since — on the shared branch and on your own machine.
update-screenshots
When a check catches that a screen looks different from before, this decides whether the difference was intentional and, only if it was, records the new look as the one to compare against from now on.
update-skills
Write down what you learned the hard way this session so the next one does not have to rediscover it, and work out where it belongs — a page of its own, or a line added to notes that already exist.
ux-css-layout
VS Code has a house style for the code behind its own interface, and this is it — the conventions a reviewer will expect, down to how text should behave when a panel gets too narrow to hold it.
ux-theming
Nothing in VS Code's interface picks its own colors — they come from whichever theme the user has chosen, and this is how you wire a new piece of interface into that system.
vscode-dev-workbench
Serves the browser version of VS Code, Agents window included, straight from the source code sitting on your own machine, so a change can be clicked through while it is still being written.
No skill in this package matches that.
HOW TO GET IT
WHOLE PACKAGEInstall the whole package (49 skills)
npx skills add microsoft/vscodeSINGLE SKILLOr install a single skill
npx skills add microsoft/vscode --skill <name> --full-depthPick the skill name from the Skills tab — each entry there installs independently.