agentleFS
frankxaifrankxai/creator-skills/skills/music/suno-ai-mastery/SKILL.md

suno-ai-mastery

Skill6 starsChanged 47 days agoMITView on GitHub

Design professional song and instrumental direction for the newest stable Suno model using genre, harmony, rhythm, arrangement, vocal persona, production, dynamics, and controlled iteration. Use for Suno composition strategy, arrangement, production troubleshooting, covers, extensions, and remasters.

  • Reads credentials

1 other project uses a version of this, 1 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 7 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use the live MuAPI model schema before generation.

    line 36
  2. Very specific:

    Use the live model catalog and schema rather than assuming model availability.

    line 60
  3. Distinctive:

    Use this order so the model receives priorities clearly:

    line 78
  4. Distinctive:

    Do not write "in the style of", "sounds like", or named-voice references.

    line 88
  5. Distinctive:

    Avoid "perfect" or celebrity voice language.

    line 109
  6. Distinctive:

    Use only high-value tags in lyrics:

    line 123
  7. Distinctive:

    Keep 1–3 cues per section.

    line 128
  8. Distinctive:

    Use Exclude Styles to protect the thesis.

    line 132
  9. Distinctive:

    Do not use exclusions to negate half the prompt.

    line 141
  10. Very specific:

    Read the relevant modules in …

    line 169

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Suno AI Mastery

Freshness rule

Suno changes quickly. Do not hardcode an old model as current. When operating the UI, capture it and select the newest stable model available to the logged-in account unless the user requests another. Treat exact character limits and feature names as UI facts to verify, not memory.

Suno, Atlas Cloud, and MuAPI are paid services. Use Suno's product UI for its full editing suite. When API automation is needed, Atlas Cloud and MuAPI are optional routes to Suno generation; they are never required for prompt design. Confirm the model and price in the live provider catalog before submitting, and get user approval before any paid generation.

Optional MuAPI Generation

The sibling muapi_music.py helper maps a song idea and style into MuAPI's asynchronous Suno API. It is dry-run by default:

python3 <skill-directory>/muapi_music.py \
  "Late-night city lo-fi piano with rain" \
  --style "lo-fi hip-hop, mellow piano" \
  --instrumental

Review the JSON payload first. To make exactly one paid submission, set MUAPI_API_KEY and add --submit. The helper never retries the generation POST; it only performs bounded GET polling for the returned request.

python3 <skill-directory>/muapi_music.py \
  "A hopeful song about finding your way home" \
  --style "indie pop, warm guitar, layered vocals" \
  --title "Way Home" \
  --submit

Use the live MuAPI model schema before generation. The current catalog lists this generation at $0.09 per request; confirm the price before submission. The completed result includes generated audio in outputs.

Optional Atlas Cloud Generation

The sibling atlas_music.py helper maps this skill's inspiration or custom-lyrics prompt into Atlas Cloud's asynchronous audio API. It is dry-run by default:

python3 <skill-directory>/atlas_music.py \
  "Late-night city lo-fi piano with rain" \
  --instrumental

Review the JSON payload first. To make exactly one paid submission, set ATLASCLOUD_API_KEY and add --submit. The helper never retries the generation POST; it only performs bounded GET polling for the returned prediction.

python3 <skill-directory>/atlas_music.py \
  "[Verse]\nNeon streets on a rainy night\n[Chorus]\nDrive on, drive on" \
  --custom \
  --title "Midnight Drive" \
  --style "synthwave, female vocal, cinematic" \
  --vocal-gender Female \
  --submit

Use the live model catalog and schema rather than assuming model availability. For suno/chirp-v4-5-all, inspiration mode uses the prompt as a song description, while --custom uses it as lyrics. A completed prediction returns audio URLs in outputs.


Translate intent into seven musical layers

  1. Form: song length, section sequence, bar/energy proportions.
  2. Rhythm: BPM range, meter, pocket, swing/syncopation, density.
  3. Harmony: tonal center/mode, harmonic tension, cadence character, rate of change.
  4. Melody: register, contour, motif behavior, hook interval/mouthfeel.
  5. Timbre: focal instruments, supporting textures, negative space.
  6. Performance: vocal persona, articulation, dynamics, ensemble behavior.
  7. Production: era without trademarks, spatial image, saturation/clarity, transient character, dynamic arc.

A prompt is not a keyword dump. It is an arrangement brief compressed into natural language.

Style prompt order

Use this order so the model receives priorities clearly:

Genre/fusion; emotional state and energy; tempo/meter/pocket; focal instrumentation; vocal persona and delivery; section-to-section dynamic journey; production and spatial character; ending behavior.

Name 3–6 important instruments. Over-specification makes arrangements muddy and generic.

No named-artist shortcuts

Do not write "in the style of", "sounds like", or named-voice references. Translate a reference into attributes:

  • decade or production era;
  • vocal range, grain, intimacy, articulation;
  • drum pocket and bass behavior;
  • harmonic vocabulary;
  • arrangement density;
  • spatial and saturation choices;
  • emotional/dynamic movement.

This produces more original work and avoids imitation/voice-clone requests.

Vocal persona

Specify a person-like performance, not only gender:

  • range: contralto, alto, tenor, baritone, etc.;
  • grain: clear, smoky, weathered, breath-edged, dry;
  • placement: close-mic, room-present, distant, choir-backed;
  • articulation: conversational, legato, clipped, melismatic, restrained;
  • arc: almost spoken → sung → open belt → fragile close;
  • harmony behavior: unison, thirds, octave doubles, call-response.

Avoid "perfect" or celebrity voice language. Human detail beats generic power.

Dynamic architecture

Map energy by section. A useful default is not a rule:

Intro 2 → Verse 4 → Pre 6 → Chorus 8 → Verse 5 → Bridge 3/7 → Final Chorus 9 → Outro 2

Contrast is the engine: sparse/dense, dry/wet, low/high register, solo/stacked, held/pulsed, silence/impact. The final chorus should change because the story changed.

Structural and performance tags

Use only high-value tags in lyrics:

  • [Intro], [Verse 1], [Pre-Chorus], [Chorus], [Bridge], [Outro];
  • one performance cue such as intimate, harmonized, spoken, stripped, full lift;
  • an instrumental cue only when arrangement depends on it.

Keep 1–3 cues per section. Contradictory or crowded tags reduce control.

Exclusions

Use Exclude Styles to protect the thesis. Typical exclusions:

  • unwanted vocal type or choir;
  • genre drift;
  • excessive distortion, trap hats, festival drops, cinematic swells;
  • spoken samples;
  • long intro/outro;
  • abrupt ending;
  • dense midrange for narration beds.

Do not use exclusions to negate half the prompt. If the positive direction is unclear, rewrite it.

Controlled iteration

  1. One Create action yields the initial take pair.
  2. Review both against the same brief; do not move the goalposts.
  3. Name the strongest 15–30 seconds and the single largest failure.
  4. Choose one operation: regenerate prompt, replace section, extend, cover, or remaster.
  5. Change one dominant variable at a time so learning compounds.

For extensions, restate the core genre, vocal persona, and ending target to reduce drift.

Troubleshooting

  • Generic output: strengthen premise, focal instrument, rhythmic pocket, and dynamic contradiction; remove adjectives.
  • Muddy vocal: reduce instrumentation, request close/dry vocal and restrained midrange, use a sparse verse.
  • Flat song: define section jobs and explicit energy contrast; add a silence or stripped bridge.
  • Wrong tempo/feel: state BPM plus pocket (half-time, swung, straight, syncopated).
  • Mispronunciation: spell out numbers/acronyms and respell the problem word phonetically.
  • Overlong intro: state immediate vocal entry or exact short intro behavior; exclude long intro.
  • Weak ending: specify hard stop, final held chord, unresolved cutoff, or long fade.

Output

Provide one recommended production direction and, only when useful, one deliberate alternate that changes a single thesis variable. Include why the recommendation best serves the emotional contract.

Shared fundamentals and provider boundary

Read the relevant modules in https://github.com/frankxai/agentic-music-producer-os/blob/18dc9bee02001befb3a2a3bd7747b78f798dff13/docs/MUSIC-FUNDAMENTALS.md, then https://github.com/frankxai/agentic-music-producer-os/blob/18dc9bee02001befb3a2a3bd7747b78f798dff13/docs/PROVIDER-CAPABILITIES.md and https://github.com/frankxai/agentic-music-producer-os/blob/18dc9bee02001befb3a2a3bd7747b78f798dff13/docs/MUSIC-FACTORY-CONTRACT.md when selecting tools or executing production. Canonical public source: https://github.com/frankxai/agentic-music-producer-os. The craft precedes the provider packet.

Suno v6 is the documented baseline checked 2026-10-01; capture the actual account model, settings, limits and credit cost before operating. No official public generation API was verified. Preserve the approved lyrics across adapters. Score/text analysis and a prompt review do not prove audio quality. Record real listening and measurements separately; keep unknown values null. Personal preferences and artist canon remain in their owner project.

Resolve docs/... paths from the repository root, not from the skill folder. If using only this skill outside a checkout, read the same named documents from the canonical public source.

Current craft and production integration

Checked 2026-10-01. Use the public music fundamentals, provider register and factory contract. They cover harmony/voice leading, groove, motif, prosody, hooks, form, vocals, arrangement, low end, mixing, measured delivery, provenance and durable jobs. Load only the modules needed for this task.

Keep local artist/persona canon, identity, preferences and account entitlements scoped to this project. Shared craft does not assign a singer or overwrite canon. Select the current account-visible Suno model (documented baseline v6); no official public Suno generation API was verified. Lyria, Eleven Music and fal MiniMax Music 3 have separate documented API routes and constraints. A skill or MCP tool is not a connected generation account.

Preserve approved lyrics across adapters. Prepare, authorize/reserve, submit, reconcile, archive, listen, measure, attest rights and stage release separately. Bind receipts to exact contract/asset hashes. Never infer audio quality, measured BPM, playback, rights or publication from a text review or completed job. Unknown paid submissions remain held for reconciliation; no blind rerender.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.