coroboros/agent-skills / skills/suno-produce
suno-produce
VerifiedCreate and revise Suno v5.5 prompt artifacts from a song, track or album brief. Produces TRACK.md, ALBUM.md for explicit multi-track work, and optional ARTIST.md identity. Validates prompt fields and preserves supplied lyrics. Artifact-only; audio generation stays in Suno. Pure lyric writing and audio post-production use other skills.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 18.8 KB |
| evals/evals.json | config | 3.7 KB |
| references/genre-templates.md | markdown | 9.7 KB |
| references/rights-and-deprecation.md | markdown | 7.6 KB |
| references/sliders-and-personalization.md | markdown | 9.2 KB |
| references/songwriting-craft.md | markdown | 5.3 KB |
| references/style-and-lyrics.md | markdown | 18.6 KB |
| references/track-schema.md | markdown | 15.2 KB |
| scripts/validate.pyexecutable | script | 37.0 KB |
| steps/create.md | markdown | 11.3 KB |
| steps/revise.md | markdown | 7.5 KB |
| steps/validate.md | markdown | 6.1 KB |
references/track-schema.md
View on GitHub# TRACK / ALBUM / ARTIST schema
The three-layer artifact schema for Suno v5.5 production work. Every file produced by this skill conforms to one of the three.
The split is deliberate: each layer reads the layer below it, never above. `TRACK.md` reads `ALBUM.md` (sibling) for tracklist context. `ALBUM.md` reads `ARTIST.md` (referenced via `-f`) for artist defaults. `ARTIST.md` is a leaf — it reads nothing.
**Filename rationale.** The artist-identity layer is named `ARTIST.md`, not `MUSIC.md`, because the contents describe an artist persona (voice, custom model, recurring instrumentation, rights posture) — not music. The rename was a deliberate evolution of the open standard; the community Suno-prompt formats (Spidy88, Blake Crosley, Stoke McToke, suno.wiki) do not have an equivalent layer at all.
## Project folder layout
### Standalone single (one track, no album)
```
{track-slug}/
TRACK.md
versions/
v1.md, v2.md, … # archived prior takes (created by revise)
audio/ # optional, gitignored — user drops Suno exports here
v1-takeA.wav, …
.gitignore # the skill emits this if missing — ignores audio/
```
### Album / EP
```
projects/{album-slug}/
ALBUM.md
tracks/
01-{track-slug}/
TRACK.md
versions/
audio/
02-{track-slug}/
TRACK.md
…
.gitignore
```
The track-slug is kebab-case from the title, max 5 words. The leading two-digit prefix (`01-`, `02-`) matches the tracklist order in `ALBUM.md`. Reordering renames the folders; the skill handles this in `revise` when the album order changes.
`ARTIST.md` lives wherever the user keeps it — typically a brand workspace root (`~/<artist>/ARTIST.md`) — and is referenced from any project folder via `-f /path/to/ARTIST.md`. The skill never creates `ARTIST.md` on its own; it only consumes one when passed.
## TRACK.md
The unit of Suno generation. One file per track. Always emitted.
### Frontmatter
```yaml
---
suno_version: v5.5
title: <human title> # required, ≤ 100 chars
bpm: 92 # integer, soft guidance — drift ±4 is normal
key: D minor # optional, soft guidance — honour rate ~60%
length_target: "3:30" # MM:SS or "long-form" / "sketch"
voice_profile: null # set from bound ARTIST.md when applicable
custom_model: null # set from bound ARTIST.md when applicable
genre_primary: indie folk # one phrase, drives the load-bearing tag
era: 2010s bedroom # optional but recommended
mood: nostalgic # one or two adjectives
created: 2026-05-09 # ISO date, set on initial create
revised: null # ISO date, updated by every revise
---
```
`suno_version` is the durable lineage marker. The validator emits a YELLOW when this differs from the running Suno version, signalling that the prompt may not reproduce identically against a newer model.
### Body — six ordered sections
1. **Title** — `# <title>`. The song title repeated for human readability.
2. **Suno prompt block** — copy-paste payload. Three fenced blocks under one `## Suno prompt block` heading. Each is `text` fenced (triple-backtick + text language tag) so Markdown rendering does not interfere with the verbatim copy:
```text
### Style of Music
```text
<descriptor stack, comma-separated, ≤ 1000 chars, 4–7 descriptors>
```
### Lyrics
```text
[Intro]
[Fingerpicked acoustic, room mic]
[Soft male tenor, breathy delivery]
The hallway light still flickers
The dog still waits at the door
…
```
### Exclude Styles
```text
<0–3 tags, comma-separated>
```
```
3. **Sliders** — bullets:
```markdown
## Sliders
- **Weirdness**: 40 — keeps phrasing within indie-folk convention
- **Style Influence**: 75 — the descriptor stack is the identity, lock it in
- **Audio Influence**: 78 — Voice attached, hold resemblance
```
Audio Influence appears only when `voice_profile` or `custom_model` is set. Each slider gets a one-clause rationale.
4. **Voice / Custom Model** — declared if ARTIST.md is bound; "None." otherwise. When set, repeat the bound names so the file is self-contained when copied without ARTIST.md context:
```markdown
## Voice / Custom Model
- **Voice**: `Studio-A` (from ARTIST.md at ~/artists/studio-a/ARTIST.md)
- **Custom Model**: none
```
5. **Rationale** — prose. Why these descriptors, why this BPM, why this slider profile. Anchors the next `revise` decision. Aim 100–250 words.
6. **Iteration log** — a single bulleted list. Append-only. Each entry: `- v{N} ({YYYY-MM-DD}): <change axes>; <one-line summary>; <kept | archived>`. The first entry on `create` is `v0`. Every `revise` appends.
### Worked example — indie folk single
```markdown
---
suno_version: v5.5
title: Midnight Letter
bpm: 92
key: D minor
length_target: "3:30"
voice_profile: null
custom_model: null
genre_primary: indie folk
era: 2010s bedroom
mood: nostalgic
created: 2026-05-09
revised: null
---
# Midnight Letter
## Suno prompt block
### Style of Music
```text
Indie folk, 2010s bedroom, intimate and bittersweet, fingerpicked acoustic guitar, soft male tenor, slight lo-fi tape warmth, room mic, 92 BPM
```
### Lyrics
```text
[Intro]
[Fingerpicked acoustic, room mic]
[Soft male tenor, close-mic]
[Verse 1]
The hallway light still flickers
The dog still waits at the door
I haven't moved your jacket
It still hangs there, the same as before
[Pre-Chorus]
(building intensity)
And every quiet morning sounds the same
[Chorus]
[Doubled vocal, light harmony stack]
Waiting on the weather to change
Waiting on a reason to stay
[Verse 2]
…
[Bridge]
[Solo guitar, whispered]
…
[Final Chorus]
[Layered harmonies, fuller mix]
…
[Outro]
[Solo fingerpicked, fade]
```
### Exclude Styles
```text
808s, heavy distortion, EDM drops
```
## Sliders
- **Weirdness**: 40 — keeps phrasing within indie-folk convention
- **Style Influence**: 75 — the descriptor stack is the identity, lock it in
- **Audio Influence**: n/a — no Voice attached
## Voice / Custom Model
None.
## Rationale
Indie folk leads the Style stack because era + mood read as 2010s-bedroom and that informs vocal pocket more than instrumentation does. 92 BPM sits in the genre's centre band; the song breathes at that tempo. Excluded 808s and heavy distortion because the genre rewards space, and the brief specifically asked for "intimate". Vocal descriptor stays in Style — no Voice attached. Pre-chorus parenthetical `(building intensity)` is the one signature performance cue; everything else is structural.
## Iteration log
- v0 (2026-05-09): initial draft from brief — "intimate indie-folk track about absence, 92 BPM"
```
More worked examples (cinematic, melodic techno, melodic trap, alt rock, ambient drone, indie pop, ritual industrial, lo-fi hip-hop) live in [`./genre-templates.md`](./genre-templates.md). Adapt those by adding the YAML frontmatter and the Rationale + Iteration log sections.
## ALBUM.md
The album / EP concept and arc. One file per album folder. Created only when album mode is detected.
### Frontmatter
```yaml
---
title: <album title>
artist: <artist name | "TBD">
length_target_total: "28:00"
track_count: 7
release_format: EP # EP | Album | Mixtape | Single (when single is part of a series)
created: 2026-05-09
---
```
### Body — five ordered sections
1. **Concept** — what the album is about, in 3–5 sentences. The overarching theme. Tone and posture across the whole arc.
2. **Arc** — the dramatic shape. Four-line label minimum:
```
- Opening: <vibe / energy>
- Development: <what shifts>
- Climax: <peak track or moment>
- Closing: <resolution or fade>
```
3. **Tracklist** — numbered list. Each line: `nn. Title — BPM — key — feel`. The `nn-` prefix matches the per-track folder name.
4. **Transitions** — bullets. Each bullet describes the move between two adjacent tracks: key change, BPM ramp, cross-fade or hard cut, narrative bridge. This is what makes a tracklist an album rather than a playlist.
5. **Rights and distribution** — placeholder section with declared posture: licensed audience, intended platforms, any cover-vs-original status, ARTIST.md binding if any. Cross-reference [`./rights-and-deprecation.md`](./rights-and-deprecation.md) when the artist has a `ARTIST.md`.
### Worked example — short EP
```markdown
---
title: Quiet Rooms
artist: TBD
length_target_total: "18:00"
track_count: 5
release_format: EP
created: 2026-05-09
---
# Quiet Rooms
## Concept
A short EP about absence as a daily routine — what stays in a house after someone leaves. Five tracks, all in minor keys, all under 4 minutes. Indie folk register throughout, with one production shift at the centre. The arc is one continuous evening.
## Arc
- Opening: still and intimate, single fingerpicked acoustic, no drums
- Development: brushed drums enter, harmony stacks introduced
- Climax: track 3 — full band moment, the only loud song on the record
- Closing: returns to solo guitar, fades to room tone
## Tracklist
1. Hallway Light — 88 BPM — D minor — sparse, voice and guitar only
2. The Garden — 92 BPM — A minor — brushed drums, upright bass enters
3. Midnight Letter — 96 BPM — D minor — full band, the loudest track
4. Still Waiting — 90 BPM — F# minor — pulls back, layered harmonies
5. Morning Tone — 78 BPM — D minor — solo guitar, room tone, fades to silence
## Transitions
- 1 → 2: D minor to A minor, key kinship; +4 BPM ramp; soft cross-fade
- 2 → 3: A minor to D minor, downward fifth; +4 BPM ramp; clean cut on the kick
- 3 → 4: D minor to F# minor, modal shift, mood drops; -6 BPM; long cross-fade
- 4 → 5: F# minor to D minor; -12 BPM, full pull-back; cross-fade through room tone
## Rights and distribution
Original songs, no covers, no sampled audio. No `ARTIST.md` bound for this EP — single-project artifact. See [`rights-and-deprecation.md`](./rights-and-deprecation.md) for the export-WAV reminder before model deprecation lands.
```
## ARTIST.md
The artist identity layer. Optional. Artist-scoped — one file referenced from many album folders. The skill never auto-creates one; the user keeps it wherever their workspace lives and binds it via `-f /path/to/ARTIST.md`.
The Voice and Custom Model fields encode the platform-side personalization. Voice input audio is 15 s – 4 min (Suno selects the best 2 min after voice-print verification). Custom Model training requires ≥ 6 stylistically-consistent tracks; up to 3 Custom Models per Pro/Premier account; training takes 2 – 5 minutes. None of these are enforced by the validator (they are platform-side facts) — they are recorded in the body so the user has the rules at hand when registering identity.
### Frontmatter
```yaml
---
artist: <artist name>
voice_profile: <name registered in Suno UI | null>
voice_consent: <"self" | "licensed: <source>" | "public-domain: <source>" | null>
custom_model: <name registered in Suno UI | null>
custom_model_training_set: <"self-authored" | "licensed: <source>" | null>
recurring_instruments: ["acoustic guitar", "upright bass", "Rhodes"]
slider_bias:
weirdness_default: 40
style_influence_default: 75
audio_influence_default: 78
rights_posture: license-only # license-only | licensed | public-domain | unknown
created: 2026-05-09
---
```
### Body — four ordered sections
1. **Artist identity** — 2–3 sentences. Era, register, recognisable signatures. The sound someone would describe in one breath.
2. **Voice and Custom Model** — what's registered, what's the consent posture, what's the training-set posture. Cross-reference [`./rights-and-deprecation.md`](./rights-and-deprecation.md) § *Voice-cloning consent*.
3. **Recurring instrumentation and texture** — the descriptor fragments that should bias every TRACK.md created under this identity. Plus textures (`tape warmth`, `room mic`, `vinyl crackle`) that are part of the artist's sound.
4. **Rights posture** — the artist's commercial-rights stance. Statement of what is original, what is licensed, what is public-domain. Reminder to export WAV stems before model deprecation. Cross-reference `rights-and-deprecation.md` for the active litigation context.
### Migrating from MUSIC.md
The artifact was renamed from `MUSIC.md` to `ARTIST.md` to align the filename with what the file describes. To migrate an existing artifact: rename the file (`mv MUSIC.md ARTIST.md`), update any `-f` references in your project workspace, and re-run the validator. The schema is identical — frontmatter and section names did not change. The `MUSIC.md` filename is no longer recognised by the validator; an explicit rename is required to clear stale references.
### Worked example — minimal ARTIST.md
```markdown
---
artist: Studio-A
voice_profile: studio-a-tenor
voice_consent: self
custom_model: null
custom_model_training_set: null
recurring_instruments: ["fingerpicked acoustic guitar", "upright bass", "brushed drums", "soft male tenor"]
slider_bias:
weirdness_default: 40
style_influence_default: 75
audio_influence_default: 78
rights_posture: license-only
created: 2026-05-09
---
# Studio-A
## Artist identity
Indie folk in the 2010s-bedroom register. Soft male tenor, intimate close-mic vocals, fingerpicked guitar as the spine of every arrangement. Brushed drums and upright bass when the song asks; otherwise sparse.
## Voice and Custom Model
- Voice profile `studio-a-tenor` registered in Suno UI on 2026-05-01. Cloned from self-recorded acapella, 90 seconds, low and high register. Consent: self.
- No Custom Model trained as of 2026-05-09. May add when six-track-minimum catalogue exists.
- See [`rights-and-deprecation.md`](rights-and-deprecation.md) § *Voice-cloning consent* for the consent contract this artist signs against.
## Recurring instrumentation and texture
- Hero instruments: fingerpicked acoustic guitar (always), soft male tenor (always)
- Supporting: upright bass, brushed drums (when arrangement asks), Rhodes (rare, special-occasion)
- Textures: room mic, slight lo-fi tape warmth, no auto-tune, no heavy reverb on lead vocal
## Rights posture
License-only — Suno's commercial assignment to a paid subscriber. No representation that copyright vests. No DMCA / Content ID enforcement available. Original songs, no covers. Voice cloned from self-recorded audio with consent. See [`rights-and-deprecation.md`](rights-and-deprecation.md) for the deprecation cliff and export-WAV reminder.
```
## Validator output schema
The JSON shape returned by [`../scripts/validate.py`](../scripts/validate.py):
```json
{
"verdict": "GREEN" | "YELLOW" | "RED",
"file": "tracks/01-midnight-letter/TRACK.md",
"errors": [
{"check": "<check_name>", "line": 16, "value": "<actual>", "expected": "<expected>", "fix": "<one-line fix>"}
],
"warnings": [
{"check": "<check_name>", "line": 16, "value": "<actual>", "expected": "<expected>", "fix": "<one-line fix>"}
],
"info": [
{"check": "<check_name>", "value": "<actual>", "running": "<runtime value>"}
]
}
```
`errors` empties on GREEN and YELLOW. `warnings` empties on GREEN. `info` always populates the `suno_version` running comparison, even on GREEN.
Stable contract for downstream consumers. Future `migrate` verb will read `info.suno_version` to decide what to rewrite.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |