microsoft/vscode-team-kit

Profil microsoft

Dernière activité source enregistrée: 19 octobre 2026 à 20:29

Catalogue SkillsMP mis à jour: 26 octobre 2026 à 11:37

skills collectés: 26

Étoiles GitHub: 63
Forks GitHub: 12

README du dépôt

Fichier source du README : README.md

Lire le README source

Afficher le README complet

# VS Code Team Kit These are the agent plugins the VS Code team uses to build VS Code — code reviews, PR workflows, notification triage, and code quality guardrails, all running inside Copilot's agent mode. We're sharing them because they're useful and because we think seeing how a team actually works with AI agents is more helpful than docs alone. ## Getting Started Install any plugin directly in VS Code: 1. Run **Chat: Install Plugin from Source** 2. Enter `microsoft/vscode-team-kit` 3. Pick the plugin you want **If you install one thing**, make it [review-areas](review-areas/). It runs an area-focused code review after any non-trivial change — correctness, security, tests, performance — and catches things a single-pass review misses. Just say "review my code" in agent mode. **If you install three things**, add [monitor-pr](monitor-pr/) (watches CI and Copilot's review in the background so you can move on) and [handy-commands](handy-commands/) (slash commands for creating PRs, addressing review comments, and fixing issues end-to-end). ## Plugins ### Review — catch problems before they ship | Plugin | What it does | |---|---| | [review-areas](review-areas/) | Fans out parallel subagents across correctness, tests, security, performance, and product — then synthesizes only the findings worth blocking a PR for | | [model-council](model-council/) | Sends the same review to GPT, Claude, and Gemini independently, then surfaces where they agree and where they disagree | | [review-plan](review-plan/) | Reviews implementation plans *before* you start coding — checks completeness, feasibility, sequencing, scope creep, and risk | ### Ship — from commit to merge | Plugin | What it does | |---|---| | [monitor-pr](monitor-pr/) | Launches background terminals that watch CI and Copilot's code review, then notifies the agent when either finishes — no polling, no tab-switching | | [handy-commands](handy-commands/) | Slash commands for the things you do every day: `/commit-and-pr`, `/pr-comments`, `/fix-issue` | ### Triage — stay on top of GitHub | Plugin | What it does | |---|---| | [github-inbox](github-inbox/) | Smart notification triage with sub-agents for review, investigation, and memory — groups by repo, applies rules, acts on your behalf | ### Guard — enforce standards automatically | Plugin | What it does | |---|---| | [ban-ast](ban-ast/) | Bans code patterns via Tree Sitter AST queries — intercepts edits and rejects banned patterns unless the change includes a justification | ### Explore — visibility into builds and components | Plugin | What it does | |---|---| | [build-health](build-health/) | Analyzes VS Code's rolling build pipeline on Azure DevOps — finds failures, shows break/fix transitions, links to commits | | [component-explorer](component-explorer/) | Sets up fixture-based visual testing with MCP-powered screenshot comparison for UI components | ### Debug — inspect runtime behavior with the right tool | Plugin | What it does | |---|---| | [dap-cli](https://github.com/roblourens/dap-cli) ↗ | DAP debugger controls — breakpoints, stepping, stack and variable inspection, and launch configuration | | [debug](debug/) | Hypothesis-driven temporary instrumentation and local structured runtime evidence, with mandatory human reproduction and cleanup gates | | [session-timeline](session-timeline/) | Live session timeline canvas with nested tool/subagent activity, prompt inspection, filters, and search | ### Test — show testers how to verify a change | Plugin | What it does | |---|---| | [test-videos](test-videos/) | `/test-videos:generate-test-video <TPI or issues>` records a short narrated "how to test" video in real VS Code Insiders, checks every step in the app, pauses on what's wrong, and opens with a verdict (verified, issues, partial…) | ### Explain — show what a change does | Plugin | What it does | |---|---| | [pr-studio](pr-studio/) | `/pr-studio:generate-video <PR>` makes a short narrated explainer video of a pull request, checked against the diff and reviewed before it's delivered; `/pr-studio:generate-presentation <PR>` makes a slide deck of it in the same look, with every code line checked against the PR | ## How we use these Every plugin here runs in our daily workflow. A few patterns that emerged: - **review-areas after every non-trivial change.** It's the default. We run it before pushing, not after — fixing issues locally is cheaper than addressing PR comments. - **review-plan before implementing.** Plan-mode produces a plan; review-plan stress-tests it. Catching a bad assumption before coding saves hours. - **monitor-pr to stay in flow.** Push the PR, start the monitor, move to the next task. The agent tells you when something needs attention. - **ban-ast for rules the team agrees on.** Instead of documenting "don't use X" in a wiki nobody reads, we encode it as a hook that blocks the edit and explains why. ## Evaluations The repository uses [Vally](https://microsoft.github.io/vally/) to evaluate the eight skills declared under `evals` in [marketplace.json](marketplace.json). Node.js 22.12 or newer is required. Install dependencies with `npm ci`, then run `npm run eval:lint` to validate the specs without model credentials. Set `COPILOT_GITHUB_TOKEN` to run live evaluations. All eight eval specs use `gpt-6-luna` by default. `npm run eval:pr` runs the trigger suite used by pull requests; `npm run eval` runs the full suite, including capability cases. Live results are written to `vally-results/`. CI posts live scores to the PR without blocking on grader failures, so failing cases remain visible for skill improvements. Spec validation and execution errors still fail CI. Promote consistently passing cases to a gated smoke suite when their behavior is stable. ## Contributing We'd love contributions. Each local plugin is an [Agent Plugins 1.0](https://agent-plugins.org/) package. See [AGENTS.md](AGENTS.md) for the package layout and authoring rules. Run `npm run test:packages` to validate manifests, component layout, marketplace metadata, and changelog versions. A few guidelines: - **New plugin?** Open an issue first to discuss scope. We keep the collection focused, so a quick conversation helps. - **Improving an existing plugin?** PRs welcome — bug fixes, better docs, new eval cases. - **Have a workflow idea but not sure how to build it?** Open an issue describing the problem. This repo is set up for AI contributions, so a clear description often goes further than you'd expect. ## Plugin combos and overlap Plugins work independently, but some are better together: - **review-areas → monitor-pr** — Review your code, push the PR, monitor CI and Copilot's review in one flow. - **review-plan → review-areas** — Review the plan before coding, review the code after. Catches issues at both stages. - **ban-ast + review-areas** — Ban patterns enforce rules on every edit; review-areas catches what slips through. - **handy-commands + monitor-pr** — `/pr-comments` to address feedback, then monitor the resulting CI run. **review-areas vs. model-council:** Both do code review. review-areas uses one model with parallel area-focused subagents. model-council sends the same review to GPT, Claude, and Gemini for consensus. Use review-areas daily; reach for model-council on high-stakes changes.

Chargement du README du dépôt…

Skills dans ce dépôt

Développeurs de logiciels Analystes en assurance qualité des logiciels et testeurs Administrateurs de réseaux et de systèmes informatiques

3 catégories métier · 88% classifié

skill

métier

description

mis à jour

generate-presentation

non classé

Make the shortest slide deck (a 16:9 PDF with speaker notes, at most 12 slides) that fully explains a GitHub pull request, in one white Microsoft look: one idea per slide, drawn as diagrams, with screenshots of the real app when the change is visible, every…