Local VS Code Web Workbench
Validate VS Code workbench and Agents window changes in a local vscode.dev instance.
The skill confines activity to sibling local checkouts, a 127.0.0.1 service, and an optional local mock agent; it explicitly states that mock mode bypasses GitHub authentication and the hosts API, providing some data-flow transparency. Deductions: it does not require confirmation before running commands, define the permission boundaries of local WebSocket/file access and npm builds, or provide stop, cleanup, and recovery procedures; it still depends on the local workspace and browser tooling.
Working directory, ports, URLs, build prerequisites, an initial compilation race, cache behavior, and common failures are documented concretely and are mostly consistent. Deductions: it depends on an unprovided vscode-dev checkout, installed dependencies, and generated out/ artifacts; commands and logs were not executed in this static review, and failure coverage remains limited, so the score stays below the static ceiling of 10.
The audience and scenarios are clear, covering the workbench, Agents window, mock agent, mobile simulation, and cache refresh, with a stated prerequisite for Agents scenarios. Deductions: full non-fit boundaries, browser-tool limitations, and Chinese-language guidance are absent; normal installation/build flows may require overseas npm/GitHub resources. The mock interaction can work offline, but environment-fit evidence is incomplete.
The document is well organized, progressively covering layout, startup, URLs, interaction, cache clearing, console noise, troubleshooting, and the mock agent, with commands, parameters, expected logs, and a troubleshooting table. Deductions: the skill itself lacks versioning, changelog, maintenance ownership, update path, installation dependency notes, and FAQs. MIT licensing, Microsoft provenance, and repository governance are supported by shared material but are not a complete maintenance declaration in the skill.
It provides directly usable procedures for starting the server, opening URLs, browser interaction, cache invalidation, and mock-agent setup, with expected logs that make outcomes inspectable. Deductions: static evidence does not prove the vscode-dev routes, protocol version, mock binary, or browser tooling work; users still need a second checkout, generated artifacts, and a suitable environment, so the score remains near but below the static ceiling of 7.
It supplies reproducible layout assumptions, absolute-path guidance, commands, URLs, environment variables, expected logs, and named source files, creating useful but limited auditability. Deductions: there are no committed dedicated tests, execution records, or independent corroboration, and the supplied CI material does not cover this skill's key paths; the score is therefore limited below the static ceiling of 5.
- Running npm, the Node agent host, and the browser service creates local processes, build artifacts, and listening endpoints; confirm the directories, ports, and workspace contain no sensitive data, and add explicit stop and cleanup steps.
- The skill depends on a sibling vscode-dev checkout, generated vscode out/ artifacts, and browser tooling; it cannot run directly if any of these are missing.
- Dependency installation or the first build may access overseas npm/GitHub resources; mainland-China users should prepare caches, mirrors, or offline dependencies and should not generalize mock-mode offline behavior to the full installation flow.
What it does & when to use it
This skill runs the vscode.dev server locally and loads the web workbench or Agents window from a neighboring microsoft/vscode checkout. It covers starting the development server, using vscode-quality=dev URLs, browser-driven interaction, and optionally connecting a local mock agent host. The vscode-dev and vscode folders must be siblings under the same work root. Adopt it if you need fast validation of web-only workbench changes; it is not a general installation guide for desktop VS Code.
Serves the VS Code web workbench or the /agents window from the neighboring microsoft/vscode sources; runs npm run dev from the vscode-dev folder and exposes the service at https://127.0.0.1:3000; selects local development sources through URLs containing ?vscode-quality=dev; drives the UI with browser tools, including chat input and the Send button; unregisters service workers and clears caches after rebuilds; and optionally runs agentHostServerMain.js from the vscode checkout to expose a WebSocket mock agent host with seeded or scripted sessions.
- A VS Code web-workbench developer wants to validate local source changes before shipping an Insiders build.
- An engineer testing the Agents window wants a local mock agent host that bypasses GitHub authentication and tunnel discovery.
- A browser-automation workflow needs to exercise chat input, message sending, or workbench behavior.
- A developer editing vscode sources needs to recover from stale UI caused by service-worker caching.
Pros & cons
- Loads the local vscode sources directly into a local web workbench, making it useful for rapid web-only validation.
- Covers both the main workbench and the /agents Agents window.
- The local mock agent host can bypass GitHub authentication and the /agents/api/hosts endpoint, enabling offline scenarios.
- Includes concrete procedures for chat sending, cache invalidation, mobile emulation, and common failures.
- Requires both vscode-dev and vscode checkouts arranged as sibling folders; the VS Code checkout alone is insufficient.
- The dev command must run from vscode-dev; running it from vscode fails because that repository has no dev script.
- Agents testing generally needs a second mock-agent-host process, and the required agent-host build output must already exist in vscode/out.
- The integrated browser cannot provide a fully unconstrained mobile viewport, and the source does not document a standalone installer, version requirements, or a test suite.
How to install
The source does not document a standalone installer or installation command for this skill. Prepare a microsoft/vscode checkout and a vscode-dev checkout with this layout: <workRoot>/vscode/ and <workRoot>/vscode-dev/. The skill file is at microsoft/vscode/.github/skills/vscode-dev-workbench/SKILL.md. The README states that the microsoft/vscode repository is licensed under MIT; a separate installation procedure for this skill is not specified.
How to use
Run the development server from vscode-dev, not vscode:
cd /path/to/vscode-dev
npm run dev
You can also use:
pushd /absolute/path/to/vscode-dev >/dev/null && pwd && npm run dev
When ready, open https://127.0.0.1:3000/?vscode-quality=dev for the main workbench or https://127.0.0.1:3000/agents?vscode-quality=dev for the Agents window. For Agents scenarios, start the mock host in a second terminal from vscode:
cd /path/to/vscode
node out/vs/platform/agentHost/node/agentHostServerMain.js --enable-mock-agent --quiet --without-connection-token --port 8765
Then open https://127.0.0.1:3000/agents?vscode-quality=dev&mock-agent-host=ws://localhost:8765&vscode-log=trace. If the first start reports Cannot find module './indexes', wait for the watcher to finish compiling and restart automatically. If agentHostServerMain.js is missing, run the VS Code - Build task or npm run watch first.