[codex] Improve YouTube transcript extraction

[Talljack]

What changed

• add resilient YouTube caption extraction through mobile player clients, watch-page tracks, timed-text tracks, and the existing package fallback
• support watch, Shorts, live, embed, and youtu.be URLs with strict host validation
• normalize JSON3 captions and preserve the existing import API response contract
• cap direct extraction at six YouTube requests with 300ms, 600ms, and 1200ms backoff for retryable failures
• stop extraction after aborts and add focused route/parser/fallback/budget tests

Why

The previous route relied only on youtube-transcript, so videos available through YouTube's other caption sources frequently returned no transcript. Unbounded fallback attempts could also amplify failures and trigger YouTube network protection.

Impact

YouTube imports recover more manual and automatic captions while limiting request bursts during rate limits or upstream failures. Existing API consumers keep the same response shape.

Validation

• pnpm test src/lib/youtube-transcript.test.ts src/app/api/import/youtube/route.test.ts — 22/22 passed
• pnpm typecheck — passed
• pnpm lint — passed
• pnpm build — passed
• real public-video smoke test recovered 61 English segments / 2089 characters before later network throttling
• full suite: 720/721 passed; the sole failure is the pre-existing daily-plan.test.ts:283 assertion

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
echo-type	Ready	Preview, Comment	Jul 13, 2026 1:45pm