MDE AI-DLC Harness for VS Code

This extension is the graphical client for the mde-aidlc-harness Python package. It guides the user through creating a GitLab personal access token, validates that token against the internal GitLab host, clones the selected harness source into VS Code global storage, and uses the source checkout's Python adapter to browse and export workflows, agents, tools, and skills.

Two surfaces

The dashboard is available in two places, both driven by the same controller so they always show identical state:

Editor tab — MDE: Open Dashboard (or the status bar item, or the $(link-external) button in the sidebar). This is the full experience: the catalog grid, filters, and the detail drawer. Because it lives in an editor tab, the Explorer stays available in the sidebar at the same time.
Sidebar view — a compact panel with connection status, the current source revision and target project, plus the shortcuts used most often.

Prerequisites

VS Code 1.85 or newer
Git available on PATH
Python 3.10 or newer with the harness dependencies available
A GitLab personal access token from git3.fsoft.com.vn
The token must include read_user and read_repository scopes

The dashboard opens the configured mdeAiaHarness.gitlab.tokenCreationUrl when the user chooses to create a token. The default repository is the internal mde-aidlc-harness repository on its main branch. Both the repository URL and the branch can be edited before syncing.

Features

Onboarding stepper covering sign-in, source sync, and target selection.
Catalog grid with search, per-kind tabs, and category/author filter chips.
Detail drawer showing the exact files an export will write, which of them already exist, and any missing source assets. Existing files open on click.
Install badges: Installed when the tracked files are present and match the current source revision and platform, Update available otherwise.
Multi-select installs, confirmed once for the whole batch and reported through a single cancellable progress notification.
Target project and platform persist in workspaceState across window reloads.
English and Vietnamese, following the VS Code display language or the mdeAiaHarness.language setting.

Development

npm install
npm run check     # type-check the extension host and the webview separately
npm run build     # esbuild bundles out/extension.js and media/webview.js
npm run watch     # rebuild both bundles on change

Press F5 from this folder to launch an Extension Development Host.

media/codicon.css, media/codicon.ttf, and media/webview.js are build outputs and are not committed.

Layout

src/
  extension.ts              activation, commands, status bar
  controller.ts             all dashboard state and adapter calls
  protocol.ts               types shared by the host and the webview
  l10n/strings.ts           English and Vietnamese bundles
  state/workspaceStore.ts   target, platform, and the install ledger
  dashboard/                webview host, editor panel, sidebar view
  webview/                  browser-side code, bundled to media/webview.js

Security boundary

The personal access token is stored only in VS Code SecretStorage. Repository URLs are normal settings. Git operations use a non-interactive temporary askpass bridge and never put credentials in clone URLs, process arguments, or logs. The source checkout is stored under ExtensionContext.globalStorageUri; only exported resources are written into the user's selected target project.

The webview runs under a strict Content-Security-Policy: no remote origins, no inline scripts beyond a per-load nonce, and local resources restricted to media/.

The extension manages resource files. It does not execute the exported AI agents or workflows.

MDE AI-DLC Harness