Hover — the review cockpit for your Vibe Testing suite
A place to see, run, and watch the Playwright suite your agent authored. Hover is an open-source Vibe Testing suite: you add its MCP server to the coding agent you already run (Claude Code, Cursor, …), and the agent explores your app and crystallizes each flow into a plain @playwright/test spec you own. This extension is the optional cockpit on top of that — it drives no agent and ships no engine; it's where you review the suite, its coverage, and its run health.
One Hover view in the Activity Bar, gated on Sign in to Hover Cloud — a miniature Hover Cloud in your editor. Signed in, it's a tabbed panel:
What it gives you
- Overview — the spec × run health matrix (pass / fail / flaky) with a per-spec flaky flag, and a Local ↔ Remote toggle: Local reads this checkout's
.hover/runs; Remote reads the CI runs Hover Cloud ingested for this repo. Run all specs or any single spec through your own Playwright.
- Heal — this repo's open heal queue: specs that drifted in CI. Click one to copy its
/mcp__hover__heal <slug> command; paste it into your coding agent, review the diff, and CI closes the item when the spec is green again.
- Environments — Local + remote targets (the roster commits to
.hover/environments.json). Each test account is a label + email + password — passwords stay in VS Code SecretStorage, never in a spec. Set the active target, and export an environment's HOVER_<LABEL>_USER/PASS vars to a local .env or your clipboard for CI secrets.
- Map — a coverage summary read from
.hover/hover-map.md; open the full business-flow graph in an editor panel.
- Run specs (F3) — a CodeLens on every
*.spec.ts runs it through your own Playwright in a terminal. No agent, no tokens — just @playwright/test.
How authoring works
Authoring happens in your coding agent via the MCP, not in this extension:
npm i -g @hover-dev/mcp
claude mcp add hover -- hover-mcp
Already installed? Update the MCP with npm i -g @hover-dev/mcp@latest, then reload your agent (no need to re-run claude mcp add). This extension updates itself from the Marketplace.
Then, in your agent: /mcp__hover__test_app explores your app and crystallizes specs into __vibe_tests__/. Come back here to review the Business Map, watch the Dashboard, and run the suite. The differentiator is record == replay — the agent acts through grounded tools, so the selector that drove a click is the exact one saved, and the saved tests run in your CI with zero AI.
The point of the suite is regression safety for AI-written code: you define a flow's behavior once, as a spec, and it guards that behavior on every future change — however much the AI rewrites the code underneath, the spec holds it to what you defined.
Requirements
- VS Code 1.85 or higher.
- A Hover-tested project — i.e. one where you've run
@hover-dev/mcp from your own coding agent. The extension reads the .hover/ wiki + __vibe_tests__/ specs it produces.
New to Hover?
Visit gethover.dev to get started, or read the docs.
Open source (Apache-2.0) — github.com/Hyperyond/Hover.
