* fix(youtube/transcript): scope timedtext URL match to current videoId
YouTube watch-page is an SPA — page.goto between watch URLs preserves
performance.getEntriesByType('resource') entries from prior videos.
findTimedtextUrl filtered only by lang, so a previously-viewed
same-language video's timedtext URL could be picked up by the polling
loop before the current video's fetch hook captured a fresh one,
returning the wrong video's captions to the caller.
Fix: require URLs to contain v=<currentVideoId> across all three paths:
- in-page findTimedtextUrl (resource-buffer scan)
- in-page isJson3TimedtextUrl (fetch/XHR hook)
- Node-side extractSegmentsFromNetworkCapture (CDP capture)
Most likely to hit callers that reuse a single daemon tab to fetch
many transcripts back-to-back (e.g. ml-scout). Confirmed in the wild:
a Fox News Ukraine clip got Whisper Flow promo captions written to
its row when the prior call on the same tab pulled an English
Whisper Flow video.
Adds one source-contract assertion (both in-page sites use a shared
videoIdMarker) and one behavioral test (CDP capture buffer with a
stale 'v=prev' entry alongside the current 'v=abc' returns only the
current video's captions).
* fix(youtube): exact-match transcript timedtext video id
---------
Co-authored-by: ml-scout <ml-scout@anthropic.com>
Co-authored-by: jackwener <jakevingoo@gmail.com>
* fix(youtube): request srv3 format for caption URLs (#1420)
YouTube may return empty responses when caption URLs lack an explicit format
parameter. This adds fmt=srv3 (standard YouTube XML caption format) to the
caption URL when no fmt parameter is already present, with a fallback to the
original URL if srv3 also returns empty.
Also adds HTTP status checking before reading the response body, preventing
silent failures on non-200 responses.
Fixes#1420
* fix(youtube): preserve caption fetch failures
---------
Co-authored-by: jackwener <jakevingoo@gmail.com>
Fixes#1376 — YouTube transcript command failed with `No captions available for this video` for all videos.
## Root cause
Transcript adapter used InnerTube `/youtubei/v1/player` API with Android client context (`clientName: 'ANDROID'`, version `20.10.38`) to retrieve caption track URLs. YouTube has restricted/deprecated this approach; the Android client no longer reliably returns captions data.
## Fix
Replace Step 1 (caption track retrieval) with watch page HTML bootstrap parsing — fetch `/watch?v=...` with cookies and extract `ytInitialPlayerResponse.captions.playerCaptionsTracklistRenderer`. This is the same approach used by sibling `clis/youtube/video.js`, so it's an alignment to existing site-local stable pattern, not a new invention.
## 2 head iteration
- `cf77f5e8` initial fix (Step 1 caption retrieval switch + 18/18 unit tests)
- `bb30788c` lead test hardening — source-contract regression test in `transcript.test.js`:
- **positive lock**: must fetch `/watch?v=...`, parse `ytInitialPlayerResponse`, read `playerCaptionsTracklistRenderer`
- **negative lock**: must NOT use `/youtubei/v1/player` or `clientName: 'ANDROID'` (prevents regression)
- stale Android-InnerTube file header comment also updated
## Better-solution evaluation
- Official YouTube Data API captions surface (`developers.google.com/youtube/v3/docs/captions/download`) is owner-authorized API, NOT a public transcript replacement
- yt-dlp also relies on watch-page bootstrap path
- Existing `youtube/video.js` already uses the same `ytInitialPlayerResponse` extraction → this PR aligns transcript with stable site-local pattern instead of inventing a new path
## Typed failure / no-silent-empty boundaries
- watch HTML HTTP failure / missing `ytInitialPlayerResponse` / no `captionTracks` → `CommandExecutionError` (typed fail)
- Empty parsed XML → `EmptyResultError` (existing path, preserved)
- `Strategy.COOKIE` matches YouTube adapter family + `video.js`; cookies/session/consent unavailable → typed fail not silent empty success illusion
## Diff containment
Runtime change limited to Step 1 caption track discovery. XML fetch, segment parsing, chapters, raw/grouped formatting all unchanged.
## Verification
Local: YouTube adapter tests `19/19` (+1 from new test), `npm run build`, typed-error-lint `192/192`, silent-column-drop `103/103`, doc coverage `140/140`, `docs:build`, listing-id advisory unchanged `13`, `git diff --check`, merge-tree clean.
GitHub: build × 3 OS, unit × 2 shards, bun-test, adapter-test, audit, doc-coverage, docs-build all SUCCESS. PR CLEAN/MERGEABLE.
Author: kagura-agent (fork). Lead: codex-mini0. Aux: First-principles-0. Coordination: pr-monitor.
* refactor(adapters): convert adapter layer from TypeScript to JavaScript
Core framework stays TypeScript; adapter layer moves to JS-first.
Adapters are essentially "executable config + browser scripts" that
barely use TS features — this simplifies the build/distribution pipeline
by removing the dist/clis/ intermediate compilation step.
Changes:
- Convert all 753 adapter files in clis/ from .ts to .js
- Update tsconfig to exclude clis/ from compilation
- Simplify build-manifest to scan clis/*.js directly (no dist/clis/)
- Update discovery, main, fetch-adapters to load JS adapters from clis/
- Update generate-verified to output .js artifacts
- Update package.json files field: dist/clis/ → clis/
- Fix all test files for the .ts → .js transition
* fix(main): use findPackageRoot for BUILTIN_CLIS path
The previous relative path (../../clis from __dirname) only worked for
dist/src/main.js but broke dev mode (tsx src/main.ts) where __dirname
is <repo>/src — resolving to /clis instead of <repo>/clis.
Use findPackageRoot() which works for both dev and prod paths.
Ocean