Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Captures human source verification for tracks, timestamps it, and updates track files. Use when sources need human review before generation.
.claude/skills/bitwize-music-studio-verify-sources/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-11 | ✗→✓ | ▲ Improved | 31% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -23% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -20% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 260% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 87% | 0% |
Input: $ARGUMENTS (album name)
Guide the user through source verification for all tracks with pending sources in the specified album.
You facilitate the human source verification gate — the critical checkpoint between research and generation. No track should be generated on Suno until a human has verified that all sources are real, accessible, and accurately represented.
find_album(name) — fuzzy match by name, slug, or partialget_pending_verifications(album_slug="{album}") — returns pending tracks for the target albumget_pending_verifications() for all albumsIf no pending tracks:
All tracks in [album] have been verified. No action needed.If pending tracks exist, list them:
SOURCE VERIFICATION: [Album Title]
===================================
Tracks needing verification:
1. [track-slug] — [track-title]
2. [track-slug] — [track-title]
...
Total: X tracks pending verificationFor each pending track:
extract_links(album_slug, track_slug) — extracts markdown links from the track fileextract_links(album_slug, "SOURCES.md") — get the full citation listTRACK: [track-title]
--------------------
Sources referenced in this track:
1. [Source Name](URL) — [brief description of what it supports]
2. [Source Name](URL) — [brief description]
...
Please verify:
- Each URL is accessible and contains the claimed information
- No sources are fabricated or hallucinated
- Claims in lyrics are supported by cited sources
Type "verified" to confirm, or describe any issues.When user confirms verification for a track:
update_track_field(album_slug, track_slug, "sources-verified", "✅ Verified (YYYY-MM-DD)") — updates the field and auto-rebuilds state cache✅ [track-title] — Sources verified (2025-02-06)After processing all tracks, check if the album status should advance:
get_album_progress(album_slug) — check how many tracks are now verifiedResearch Complete:| **Status** | Research Complete | → | **Status** | Sources Verified |In Progress:rebuild_state() to ensure MCP server has fresh dataVERIFICATION COMPLETE
=====================
Album: [title]
Tracks verified: X/Y
Album status: [previous] → [new status]
Date: YYYY-MM-DD
All sources verified. This album is cleared for lyric writing.
Next step: /bitwize-music:lyric-writer [track] (write lyrics from verified sources)Partial verification report (if some tracks still pending):
VERIFICATION PROGRESS
=====================
Album: [title]
Tracks verified this session: X
Tracks still pending: Y
- [track-slug] — [reason if known]
Album status: unchanged ([current])
Resume verification later with /bitwize-music:verify-sources [album]If the user reports a problem with a source:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 7,665 | 5,415 | -29% | 1 | 1 | 0% | 1,126 | 1,645 | +46% | 0 | 0 | — |
case-02 | fail→fail | 6,268 | 4,881 | -22% | 1 | 1 | 0% | 1,037 | 1,532 | +48% | 0 | 0 | — |
case-03 | fail→fail | 5,771 | 5,145 | -11% | 1 | 1 | 0% | 901 | 1,639 | +82% | 0 | 0 | — |
case-04 | pass→fail | 16,630 | 15,672 | -6% | 1 | 1 | 0% | 2,685 | 1,575 | -41% | 0 | 0 | — |
case-05 | fail→fail | 7,250 | 5,630 | -22% | 1 | 1 | 0% | 298 | 1,737 | +483% | 0 | 0 | — |
case-06 | pass→fail | 3,492 | 6,580 | +88% | 1 | 1 | 0% | 608 | 1,656 | +172% | 0 | 0 | — |
case-07 | fail→fail | 7,044 | 5,737 | -19% | 1 | 1 | 0% | 1,294 | 1,595 | +23% | 0 | 0 | — |
case-08 | fail→fail | 4,137 | 6,506 | +57% | 1 | 1 | 0% | 647 | 1,727 | +167% | 0 | 0 | — |
case-09 | fail→fail | 5,508 | 4,509 | -18% | 1 | 1 | 0% | 881 | 2,012 | +128% | 0 | 0 | — |
case-10 | pass→pass | 7,982 | 5,255 | -34% | 1 | 1 | 0% | 1,217 | 2,119 | +74% | 0 | 0 | — |
case-11 | fail→pass | 9,169 | 3,131 | -66% | 1 | 1 | 0% | 1,423 | 1,869 | +31% | 0 | 0 | — |
case-12 | fail→pass | 11,898 | 1,688 | -86% | 1 | 1 | 0% | 1,945 | 1,506 | -23% | 0 | 0 | — |
case-13 | fail→fail | 7,185 | 3,324 | -54% | 1 | 1 | 0% | 1,110 | 1,832 | +65% | 0 | 0 | — |
case-14 | fail→fail | 8,877 | 3,986 | -55% | 1 | 1 | 0% | 1,416 | 1,855 | +31% | 0 | 0 | — |
case-15 | fail→fail | 13,428 | 1,786 | -87% | 1 | 1 | 0% | 1,988 | 1,471 | -26% | 0 | 0 | — |
case-16 | fail→pass | 12,771 | 2,458 | -81% | 1 | 1 | 0% | 2,087 | 1,667 | -20% | 0 | 0 | — |
case-17 | fail→pass | 3,085 | 3,447 | +12% | 1 | 1 | 0% | 500 | 1,801 | +260% | 0 | 0 | — |
case-18 | fail→fail | 8,368 | 7,781 | -7% | 1 | 1 | 0% | 1,300 | 1,693 | +30% | 0 | 0 | — |
case-19 | fail→pass | 7,444 | 5,141 | -31% | 1 | 1 | 0% | 1,123 | 2,103 | +87% | 0 | 0 | — |
case-20 | fail→fail | 8,298 | 9,833 | +18% | 1 | 1 | 0% | 1,301 | 1,787 | +37% | 0 | 0 | — |
case-21 | fail→pass | 6,201 | 1,334 | -78% | 1 | 1 | 0% | 974 | 1,435 | +47% | 0 | 0 | — |
case-22 | fail→pass | 6,005 | 3,441 | -43% | 1 | 1 | 0% | 958 | 1,769 | +85% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 12 counted toward the lift figure. The other 10 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +23 percentage points is the difference between those two pass rates over the 12 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.