fix(transcribe): env pass-through for yt-dlp cookies and JS runtimes

[Azeem1985]

Fixes #3049.

download_audio() hardcodes its embedded YoutubeDL options, so on server IPs where YouTube answers Sign in to confirm you're not a bot (yt-dlp's FAQ prescribes browser cookies for this), the user's exported cookies file has no way in — and neither does --js-runtimes node, which yt-dlp's YouTube extractor wants for full format coverage.

This mirrors the module's existing env convention (GRAPHIFY_WHISPER_MODEL, GRAPHIFY_WHISPER_PROMPT):

• GRAPHIFY_YTDLP_COOKIES → cookiefile (same as yt-dlp --cookies)
• GRAPHIFY_YTDLP_JS_RUNTIMES → js_runtimes (same as yt-dlp --js-runtimes), e.g. node

With neither set, the options dict is byte-identical to before. 4 new tests; tests/test_transcribe.py passes (15). Applies clean on v8 (0.9.49).

🤖 Generated with Claude Code

[graphify-labs]

Graphify reviewed this change.

Looks safe to merge — no coupling regressions and no blocking issues, checked against the code graph (not a self-assessment).

Formal verification. No changes could be formally verified in this run.

---

Graphify review — findings

Adds _ytdlp_env_opts so download_audio reads GRAPHIFY_YTDLP_COOKIES and GRAPHIFY_YTDLP_JS_RUNTIMES from the environment and forwards them to yt-dlp as cookiefile and js_runtimes, letting the embedded API supply browser cookies and a JS runtime to get past YouTube's bot check. With neither var set the options stay empty, so existing behaviour is unchanged.

No blocking issues surfaced. 1 lower-confidence candidate did not survive cross-model review.

Analysis details — impact, health, verification

Impact & health

Graphify review

Impact — 205 functions depend on the 48 functions this change touches.

Health — this change adds coupling hotspots:

• new: dispatch_command() — 2 callers, 122 callees
• new: transcribe() — 5 callers, 5 callees
• new: ingest() — 3 callers, 8 callees
• new: download_audio() — 4 callers, 3 callees
• new: test_poisoned_manifest_is_healed() — 0 callers, 6 callees

Verification — 205 functions in the blast radius were not formally verified this run (proofs are advisory here).

Gate & verification

graphify gate

PASS — objectively clean (no health regressions, tests not run — proofs not run this pass (advisory)). Grounded, not self-assessed.

Advisory (not blocking):

• verification_scope: 51 function(s) in the blast radius were not formally verified this run

Formal verification

Could not verify: Could not verify download\_audio.

The verifier did not have enough to check download\_audio, so it is saying so rather than guessing. No false assurance is the whole point.

Guarantee: No guarantee either way, this is an honest abstention, not a pass.

Note: Reason: parameter `output_dir` is annotated `Path` — outside the synthesizable primitive/collection set

· 1 grounded finding(s) anchored inline below; 4 more finding(s) on lines outside this diff (see the check run).

[graphify-labs]

⚠️ Health regression — download_audio()

high coupling complexity (Ca·Ce = 12).

Grounded coupling-delta finding (deterministic), not an LLM guess.

[Azeem1985]

Thanks for the review — glad to see the gate come back clean. For what it's worth, our fork (TelB-io/graphify) has been running this change in production: with the env vars unset it's a no-op as described, and with them set it gets yt-dlp past YouTube's bot check reliably. Happy to adjust anything if the maintainers want changes before merge.