fix: preprocess music cover lyrics flow

[RyanLee-Dev]

Summary

• run the music cover preprocess API before music generation whenever users provide custom lyrics for mmx music cover
• send the returned cover_feature_id in the follow-up generation request instead of raw reference audio
• add typed preprocess response support and cover command regression coverage for the two-step flow

Testing

• bun test test/commands/music/cover.test.ts
• bun run typecheck

[github-actions]

Summary

Adds a cover preprocess step to music cover — when --lyrics is provided, the CLI now calls /v1/music_cover_preprocess first to obtain a cover_feature_id, then uses it in the main music generation request. This allows lyrics to be passed to the cover flow without requiring manual ASR extraction.

---

Checks

Command	Result	Note
bun run typecheck	✅ Pass	No TypeScript errors
bun test	✅ Pass	354 tests, 0 failures
bun run lint	⚠️ Warnings + 2 pre-existing errors	The 2 errors (test/auth/oauth.test.ts, test/sdk/image.test.ts) are pre-existing on main — confirmed by running lint on ${PR_BASE_SHA} before the PR changes
bun run build:dev	✅ Pass	236KB (mmx) + 54KB (sdk)
Live smoke: music cover --dry-run --lyrics ...	✅ Pass	preprocess_request + request body correctly structured
Live smoke: music cover --dry-run (no lyrics)	✅ Pass	audio_url included directly, no preprocess call

---

Changes by file

src/commands/music/cover.ts

• When --lyrics is supplied, calls musicCoverPreprocessEndpoint before the main music generation request.
• Sets cover_feature_id from the preprocess response into the main request body.
• In dry-run mode, prints both preprocess_request and request so the full flow is visible.

One minor observation: audioSource (the audio_url / audio_base64 pick) is computed before the needsPreprocess branch — this is fine, but worth a quick mental verification that readFileSync is only called when audio-file is used (not audio-url). It is — no issue.

src/client/endpoints.ts

• Added musicCoverPreprocessEndpoint → POST /v1/music_cover_preprocess.

src/types/api.ts

• Added CoverPreprocessRequest and CoverPreprocessResponse types. Looks correct — cover_feature_id is the primary output used downstream.

test/commands/music/cover.test.ts

• Updated dry-run assertion to verify preprocess body shape, cover_feature_id placeholder, and that audio_url is absent from the main request when lyrics are provided.

---

Notes / non-blocking

• The test covers the audio-url + lyrics path. The audio-file + lyrics path (which exercises audio_base64 in preprocess) isn't explicitly tested — but the code path is symmetric and the logic is straightforward.
• The lint errors in test/auth/oauth.test.ts and test/sdk/image.test.ts are pre-existing on main and unrelated to this PR.

---

✅ Approve — all required checks green, build succeeds, smoke tests pass, and no blocking issues introduced.

[TumCucTom]

this is the piece #159 was missing — switching to cover_feature_id after the preprocess call is the right shape, and the fact that the API was always two-step and the CLI was just skipping the preprocess call is a useful data point.

one concern: if preprocess succeeds but the second music_generation call fails, the user is left with a charged preprocess step and a failed cover. worth a comment in the code noting the two-step failure mode, and maybe a --no-preprocess escape hatch for retrying generation without re-charging preprocess.

when this lands, #159 should be closeable. happy to drop a follow-up comment on that issue confirming the fix if useful.

nice work.