TDD & feedback loops

Principles: outrun your headlights; bad code is more expensive with AI because speed amplifies entropy.

Default loop

1. Scope one behavior or interface delta (see interface-first-delegation).
2. Red — failing test or observable check (Jest for units; Playwright for UI when behavior is visual).
3. Green — minimal implementation.
4. Refactor — only while green; prefer module boundaries (deep-modules-architecture).

Expand scope only when the last step is green (types, lint, tests).

Test design (decide explicitly)

• Unit: pure logic (lib/billing/entitlements, lib/tasks-schema) — no mocks.
• Module interface: storage managers, API guards — mock IO at the boundary.
• Component: React Testing Library — user-visible testids; mock hooks/storage as needed.
• E2E: Playwright for cross-route flows. tests/app/ runs signed in against a local Supabase stack on 8800 (the config builds and starts the app itself); tests/marketing/ runs against the marketing dev server on 3100.

Do not add tests that only assert implementation details.

SpokenLeaf commands

Command	Use
npm test -- --testPathPattern=<suite>	Fast subset
npm run test:ci	Prettier + Jest CI
npm run test:e2e:app	App E2E, signed in (needs supabase start)
npm run test:e2e:marketing	Marketing E2E

Anti-patterns

• Landing 500+ line component changes without slice tests.
• Ignoring failing suites in jest.config.js instead of fixing or deleting stale tests.
• Skipping browser visibility for layout/regression UI work when tools are available.

Chain to sibling skills

• Testable structure → deep-modules-architecture.
• Plans name what to test → interface-first-delegation.
• Domain terms in test names → ubiquitous-language.