feat: transcript-grounded, on-brand metadata authoring (#43)

[jdwit]

Implements Phase-1 of #43. Gives an agent the two raw materials it needs to author good metadata - the spoken content of a video and a per-channel house style - and lets the existing videos update path apply the result. The agent writes; the CLI supplies grounding and applies changes.

What's new

Transcript grounding (src/ytstudio/commands/videos.py)

• ytstudio videos captions <id> [-o table|json] - list caption tracks (captions.list).
• ytstudio videos transcript <id> [--lang xx] [-o text|json] - download a track as clean plain text (captions.download, tfmt=srt); -o json adds track metadata. Surfaces a clear message when a track exists but is not downloadable (restricted ASR) instead of failing opaquely.

Per-channel brand voice (src/ytstudio/config.py, src/ytstudio/commands/profile.py)

• Stored at profiles/<name>/brand.md, owner-only (0o600 in the 0o700 profile dir).
• ytstudio profile brand show | set --file <path> | edit (edit opens $EDITOR and seeds a template). --profile/-p defaults to the active profile.

Skill guidance (skills/ytstudio/SKILL.md)

• New authoring section: brand show -> transcript -> videos show -> draft -> videos update dry-run -> --execute, with the rules that the transcript is the source of truth for claims, the brand file for tone, and the agent never fabricates.

No new OAuth scope (youtube.force-ssl is already requested). External ASR/STT and bulk authoring stay out of scope for later phases.

How it was built

Designed and reviewed with a multi-agent workflow: parallel readers mapped the relevant subsystems and synthesized a code-grounded plan, then an adversarial review pass (5 dimensions, each finding independently verified) ran over the diff. Six confirmed findings were all fixed before this PR:

• SRT parsing is now block-based, so cue text that is itself numeric (a countdown, a year like "2026") is no longer mistaken for a sequence index and dropped.
• Consecutive-duplicate collapsing is gated to ASR tracks; a legitimately repeated line in a human caption ("Run! Run!") is preserved.
• Caption download distinguishes a restricted/ASR 403 from a quotaExceeded 403, so quota exhaustion reaches the real quota handler instead of telling the user to "pick another track".
• Tightened tests so the standard-over-ASR preference and the post-editor re-chmod are actually exercised (they previously could pass even if the logic were removed).
• Corrected the quota cost documented for caption list/download (~50 / ~200 units, not ~1).

Tests

tests/test_captions.py (new) covers captions list (table/json/empty), transcript text/json, language selection, standard-vs-ASR preference, numeric-cue preservation, ASR dedup vs human no-dedup, not-downloadable, and quota-not-misreported; srt_to_text has direct unit tests (multiline join, tag stripping, empty). tests/test_profile.py adds brand show/set/edit including the owner-only re-chmod. Full suite: 338 passed, ruff clean, bundled skill reference regenerated.

[codecov]

Codecov Report

❌ Patch coverage is 98.57143% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 85.03%. Comparing base (a179e78) to head (1636bf5).

Files with missing lines	Patch %	Lines
src/ytstudio/commands/profile.py	97.50%	0 Missing and 1 partial ⚠️
src/ytstudio/commands/videos.py	98.88%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #52      +/-   ##
==========================================
+ Coverage   84.28%   85.03%   +0.75%     
==========================================
  Files          15       15              
  Lines        2520     2660     +140     
  Branches      372      393      +21     
==========================================
+ Hits         2124     2262     +138     
  Misses        264      264              
- Partials      132      134       +2     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.