Vitest unit-test infrastructure + 90% coverage for components and core modules

[cliffhall]

Background

AGENTS.md mandates 90% test coverage for all code: "Ensure all code has corresponding tests. Ensure test coverage for each file is at least 90%." The Phase 1-5 component refactor explicitly deferred unit tests as out-of-scope, deferring the full coverage push to a single follow-up effort once the interface dust settled.

There is currently no Vitest config, no test runner, no test files in clients/web/. The root package.json has a placeholder "test": "echo \"Error: no test specified\" && exit 1". Coverage is effectively zero.

Scope

Test infrastructure:

• Add Vitest, @testing-library/react, @testing-library/jest-dom, jsdom (or happy-dom) to clients/web/package.json devDependencies.
• Add vitest.config.ts configured for React + TypeScript with the existing tsconfig.app.json.
• Add npm run test and npm run test:coverage scripts to clients/web/package.json and surface them at the root.
• Configure c8 / built-in coverage reporter with the 90% threshold per AGENTS.md, failing the script when uncovered.

Component tests (62 components):

• Each component under clients/web/src/components/{elements,groups,screens,views}/ has a <Name>.test.tsx file.
• Tests exercise: render with minimal props, render with each documented prop variant, callback firing on user interaction, accessible role/label assertions where relevant.
• Per AGENTS.md: "In unit tests that expect error output, suppress it from the console."

Core module tests (under core/):

• core/mcp/types.ts — type-only file; covered by usage in component tests, not separate.
• core/mcp/elicitationCreateMessage.ts — schema validation tests.
• core/mcp/taskNotificationSchemas.ts — schema validation tests.
• clients/web/src/utils/jsonUtils.ts, schemaUtils.ts — pure-logic tests against AJV-validated fixtures.
• clients/web/src/components/groups/historyUtils.ts — extractMethod happy-path and synthetic-response branch.

Acceptance criteria

• npm run test runs the full suite headlessly and exits 0 on a clean tree.
• npm run test:coverage reports ≥90% line coverage on every file under clients/web/src/components/ and core/mcp/.
• CI (.github/workflows/main.yml) is updated to run npm run test:coverage on every PR; failure blocks merge.
• AGENTS.md is updated to point at the now-real npm run test command (no longer "aspirational").

Test plan

• Run npm run test:coverage locally; manually inspect a sample of generated test files for quality (not just coverage padding).
• CI run on a draft PR confirms the coverage gate fires when coverage drops.
• npm run format && npm run lint && npm run build continues clean.

Out of scope

• E2E / browser tests (Playwright or similar) — would be a third tier.
• Storybook interaction tests — already exercised manually.
• Performance / load tests.
• Tests for the v2 core/ hook layer (its own tests live with the hook layer issue).

Independent of

• This effort can run in parallel with the v2 core hook layer effort. Tests for existing dumb components don't depend on hooks landing.

Related

• AGENTS.md § "Always test new or modified code"
• Plan reference: specification/v2_ux_interfaces_plan.md § "Out of this plan, but adjacent"
• Project history: PR Phase 5 audit: annotate spec, close HistoryControls partial #1236 review noted the missing npm run test command in the test plan checklist

[cliffhall]

Closed by #1269 (merged a1b1289).