Add Cloud Agent development environment

[anurag-roy]

Summary

Adds a .cursor/environment.json so Cloud Agents get a fully working Speak-O development environment.

Speak-O is a self-contained WXT + React MV3 Chrome extension. It has no backend, database, or external service required for development — the test suite uses fakes and never needs an ElevenLabs Provider Credential. The environment therefore only needs the Node/npm toolchain (already present in the base image) plus project dependencies.

{
  "name": "Speak-O",
  "install": "npm ci"
}

• install: npm ci performs a clean, lockfile-pinned dependency install and runs the postinstall hook (wxt prepare) that generates WXT types. It is idempotent by construction.
• No start/terminals: there are no long-running services required per boot. The dev server (npm run dev) launches Chrome interactively and is developer-initiated, so it does not belong in automatic startup.

Validation

Performed on the Cloud Agent VM (Node v22.14.0, npm 10.9.7):

Check	Result
npm ci (install)	Passed; idempotent (run twice)
npm run typecheck (tsc --noEmit)	Passed
npm test (vitest)	40/41 passed
npm run build (wxt build)	Passed; emits .output/chrome-mv3
npm run zip	Passed; emits speak-o-0.1.0-chrome.zip
npm run audit:zip	Passed; 18 files, permissions accepted
Draft environment build (npm ci)	SUCCEEDED

End-to-end demonstration

The environment-built extension (.output/chrome-mv3) was loaded in Chrome and exercised on a local article:

• Loaded in chrome://extensions with no errors.
• Injected its content script, extracted a real article, and rendered the first-use onboarding prompt.
• Rendered the floating reader control bar on the article after choosing Chrome Voice.
• The settings UI (a built React app) renders, navigates all sections, and persists a preference ("Preferences saved locally").

Speak-O loaded in chrome://extensions with no errors
Speak-O onboarding prompt on the extracted article
Speak-O reader control bar rendered on the article
Speak-O settings UI
speako_settings_ui_walkthrough.mp4

Audible narration / synchronized highlighting could not be exercised because Chrome's chrome.tts finds no system voices in the headless VM (Web Speech getVoices() returns 0 even after installing espeak-ng/speech-dispatcher; Chrome cannot reach the speech backend without a session D-Bus). This is a runtime-media limitation of the headless environment, not a defect in the npm ci dev environment, which needs no audio to build, type-check, test, or develop.

Pre-existing failing test (not an environment issue)

tests/extraction/x-articles.test.ts has one failing assertion on main (empty diff from main reproduces it). mappedBlockFromElement in src/extraction/dom-mapping.ts concatenates text-node slices verbatim, so a paragraph split across fixture lines keeps interior newlines/indentation instead of collapsing them to single spaces. This is deterministic application behavior independent of Node/npm/jsdom versions, so it is out of scope for environment setup and left unchanged.

_{To show artifacts inline, enable in settings.}