[ConalMullan]

Publish kit: tools/publish_kit.py

The render's audio is transcribed once and aligned to the scene scripts, which gives:

• out/publish/video.srt: one subtitle file for the whole video, spelled as the scripts spell it and timed to the render (no scene offsets to add up).
• out/publish/chapters.txt: a chapter at each scene's cut (the middle of the pause before its first word), titled from project.json scenes[].title or the file names, under YouTube's rules (first at 0:00, three or more, 10 s each; a short scene joins the one before).

Every part is optional: no scene scripts means raw-transcript subtitles and no chapters; --no-captions / --no-chapters skip one. Re-runs reuse the saved transcript, so renaming chapters is free. Cost: Scribe at $0.40 an audio hour (about $0.01 for 90 s), or --engine whisper locally.

youtube_upload.py --check

Read-only, about 1 quota unit: the channel a login belongs to, its scopes, and whether the Data and Analytics APIs are on, with the fix for each problem (a login command, or the Cloud Console page that switches the API on). youtube_stats.py prints the same page instead of a bare accessNotConfigured.

/publish runs the check first, then the kit, and passes the SRT as --captions and the chapters into the description.

Tested

• 10 new tests (chapter rules, titles, script order, project number, enable link); 209 OK.
• On Made with itself (93 s): 100% of script words heard, $0.0103; subtitles within ~40 ms of the hand-made file; every chapter within 0.5 s of the real cut (the first word alone ran 1-3 s late, so the kit uses the pause).
• --check live on two logins: one ready (channel, four scopes, both APIs on), one with an expired token (reported with the login command); after re-login it showed both logins are the same channel.

🤖 Generated with Claude Code