test(api/chat): add DI seam for OpenAI client + mocked route tests

[sergiopesch]

Why

The pure-function tests added in #16 cover every branch of POST /api/chat except the ones that actually call OpenAI. That means the normalize → validate → sanitize pipeline (the bulk of the server-side logic) is exercised only through its individual unit tests, not end-to-end.

Rather than mock OpenAI at the module level (brittle, tsx-specific) or spin up a real API call in CI (flaky, needs secrets), introduce a minimal dependency-injection seam so tests can swap in a fake client.

What

app/api/chat/openai-client.ts (new)

export type ChatCompletionClient = {
  chat: { completions: { create: OpenAI['chat']['completions']['create'] } };
};
export function getChatClient(): ChatCompletionClient;
export function __setChatClientForTests(client: ChatCompletionClient | null): void;

• Prod: lazily constructs a real new OpenAI(), same as before.
• Tests: call __setChatClientForTests(fake) to inject, null to clear.
• The double-underscore prefix is intentional; production code must not reach for it.

app/api/chat/route.ts

One-line swap: getClient() → getChatClient(). import OpenAI from 'openai' is removed (no longer needed in this file — the concrete type lives behind the seam).

No behavior change on the happy path.

tests/chat-route-mocked.test.ts (new, 7 tests)

Drives the full handler with canned LLM outputs:

Test	What it covers
returns a validated spec card from the LLM	normalize + validate happy path
voice mode attaches voice.spokenSummary	voice-mode branch with VOICE_PROMPT_RULES
unparseable LLM output → FALLBACK_TEXT_RESPONSE	!normalized branch
partially-valid payload → rescued via toSafeTextResponse	!validated + text-recovery branch
AbortError from client → FALLBACK_TEXT_RESPONSE (not 500)	the timeout path added in #17
non-timeout upstream error → 500 with generic message	error-leakage guard
requestedImageCount > 1 overrides LLM's imageCount	post-validate override branch

The fake client also records calls, so tests can assert on the exact messages / model sent to OpenAI.

Test count

Branch	Before	After
#17 fix/chat-route-hardening	17	24 (+7)

Verification

• ✅ npm run validate
• ✅ npm test — 24/24
• ✅ npm run build — production build succeeds

Depends on #17 (uses __resetRateLimitMapForTests from there). Stacked onto fix/chat-route-hardening.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
clitronic	Ready	Preview, Comment	Apr 21, 2026 10:31am