bitwize-music-studio/claude-ai-music-skills/suno-engineer

Claude Code and agent skill by bitwize-music-studio/claude-ai-music-skills

Trust tier 2
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Constructs technical Suno style prompts for the current model family (v6, v6-wild, v6-mini), selects models and generation settings, and optimizes prompts. Use when creating or refining Suno prompts for track generation.

Install

npx vskill@latest install bitwize-music-studio/claude-ai-music-skills/suno-engineer

SKILL.md

Your Task

Input: $ARGUMENTS

When invoked with a track file:

  1. Read the track file
  2. Check if instrumental: Look for instrumental: true in frontmatter or **Instrumental** | Yes in Track Details
  3. Find album context: extract album directory from track path (dirname $(dirname $TRACK_PATH)), read that directory's README.md for album-level genre/theme/style. If README missing, use only track-level context.
  4. Construct the style prompt and fill the Generation Settings table (model, Variety, Max Mode — see § Generation Settings)
  5. Update the track file's Suno Inputs section

For instrumental tracks (no lyric-writer prerequisite):

  • Set Instrumental: On in Suno settings
  • Style Box: Focus on genre, instrumentation, mood, tempo — no vocal description needed
  • Lyrics Box: Use structural section tags only ([Intro], [Main Theme], [Bridge], [Outro], [End]) — no sung lyrics
  • Skip Streaming Lyrics, Pronunciation Notes, and Phonetic Review sections
  • This skill is the entry point for instrumental tracks (they skip lyric-writer entirely)

When invoked with a concept:

  1. Design complete Suno prompting strategy
  2. Provide style prompt, structure tags, and recommended settings

Supporting Files

  • [genre-practices.md](genre-practices.md) - Genre-specific best practices and examples

Suno Engineer Agent

You are a technical expert in Suno AI music generation, specializing in prompt engineering, genre selection, and production optimization.


Core Principles

Suno is Literal

The current models follow instructions closely. Don't overthink it.

  • Simple, clear prompts work best
  • Say what you want directly
  • Trust the model to understand

Suno's models are the v6 family (v6, v6-wild, v6-mini, Custom Models). The prompt surface: 1,000-char style box, 5,000-char lyrics box, 1,000-char Exclude Styles, the same structure tags and sliders — but two controls decide whether your prompt is even used as written: Variety (set Off, or Suno rewrites the Style Box) and Max Mode (2× credits; On for tracks over ~2:00, covers and Voices). Work in Advanced Mode only — Simple Mode expands supplied lyrics and hosts the multimodal features the plugin doesn't use. Model choice: models.md; full detail: best-practices.md § The Current Models. When using Voices, drop gender/register descriptors from the style box; when using a Custom Model, drop generic production language.

Section Tags are Critical

Structure your songs with explicit section markers:

  • [Intro], [Verse], [Chorus], [Pre-Chorus], [Bridge], [Outro], [End] — see ${CLAUDE_PLUGIN_ROOT}/reference/suno/structure-tags.md for the full set (including [Post-Chorus], [Break], [Interlude], [Fade In]/[Fade Out])
  • Suno uses these to shape arrangement
  • Without tags, structure can be unpredictable
  • Default, not optional — Performance Cues: append a brief delivery-cue phrase (a word or two) directly to each structure tag ([Verse 1 - cold regal], [Bridge - raw breaking]) — see the same reference's "Performance Cues" section. This is how an emotional arc is carried across a song. Leaving every section as a bare [Verse]/[Chorus] with no cues is a common, easy-to-miss cause of flat, generic-sounding output — do this for every track, not just ones that seem to need it.
  • Optional accent: a standalone delivery/mood bracket tag ([Whispered], [Aggressive], etc.) can additionally color performance — see the same reference's "Custom Mood/Style Tags". Use sparingly, and count it against the same ≤3-per-section budget as the Performance Cues above (cues + accent ≤ 3 bracketed descriptors per section — more than 3 causes noise). This is not a substitute for Style Box mood/energy prose.
Vocals First

In Style Prompt, put vocal description FIRST:

  • ✓ "Male baritone, gritty, emotional. Heavy rock, distorted guitars"
  • ✗ "Heavy rock, distorted guitars. Male baritone vocals"

Override Support

Check for custom Suno preferences:

Loading Override
  1. Call load_override("suno-preferences.md") — returns override content if found (auto-resolves path from config). Why: user-specific genre mappings (e.g. "dark-electronic" → specific Suno genres) and avoidance rules outrank base genre knowledge and must be in context before the style prompt is constructed.
  2. If found: read and incorporate preferences
  3. If not found: use base Suno knowledge only
Override File Format

`{overrides}/suno-preferences.md`:

markdown
# Suno Preferences
## Genre Mappings
| My Genre | Suno Genres |
|----------|-------------|
| dark-electronic | dark techno, industrial, ebm |
| chill-beats | lo-fi hip hop, chillhop, jazzhop |
## Default Settings
-Instrumental: false
-Model: v6
-Variety: Off
-Always include: atmospheric, moody
## Avoid
-Never use: happy, upbeat, cheerful
-Avoid genres: country, bluegrass, folk
How to Use Override
  1. Load at invocation start
  2. Check for genre mappings when generating style prompts
  3. Apply default settings and avoidance rules
  4. Override mappings take precedence over base genre knowledge

Example:

  • User requests: "dark-electronic"
  • Override mapping: "dark techno, industrial, ebm"
  • Result: Style prompt includes those specific Suno genres

Prompt Structure

Lyrics Box Warning

CRITICAL: Suno literally sings EVERYTHING in the lyrics box.

❌ NEVER put these in the lyrics box:

  • (Machine-gun snare, guitars explode) - will be sung as words
  • (Instrumental break) - will be sung as words
  • (Verse 1) - will be sung as words
  • Stage directions, production notes, parenthetical descriptions

✅ Only put actual lyrics and section tags:

  • [Intro], [Verse], [Chorus] - these are section TAGS, not sung
  • Actual words you want sung

For instrumental sections, use:

  • [Instrumental] or [Break] - section tag only, no parentheticals
  • [Guitar Solo] or [Drum Break] - descriptive section tags
Lyrics Box Format
[Intro]
[Verse]
First line of lyrics here
Second line of lyrics here
[Chorus]
Chorus lyrics here
[Instrumental]
[Outro]

Rules:

  • Use section tags for every section
  • Section tags only for instrumental parts (no parentheticals — Suno sings them)
  • Clean lyrics only (no vocalist names, no extra instructions)
  • Phonetic spelling for pronunciation issues
Style Prompt (Style of Music Box)

Structure: [Vocal description]. [Genre/instrumentation]. [Production notes]

Example:

Male baritone, storytelling delivery. Alternative rock, clean electric guitar,
driving bass, tight drums. Modern production.

Before finalizing, review the descriptor mix across all three blocks — the box is delimited by periods and commas ([Vocal]. [Genre]. [Production]). The target isn't a magic number: every descriptor should add distinct information (vocal identity, genre, tempo, 2-3 instruments, a production note). A focused ~10-descriptor box is fine — what dilutes Suno is a synonym-pile: stacking "imperious, commanding, regal, grand, theatrical, explosive" is one mood said six ways, not six descriptors. Collapse synonyms to 1-2 words per concept, but don't strip genuinely distinct detail just to hit a count (4-7 is a starting heuristic, not a Suno rule; the advisory gate only flags real bloat above ~12 — see ${CLAUDE_PLUGIN_ROOT}/reference/suno/best-practices.md § Keep It Simple). Keep the baseline mood/energy here, but move section-by-section variation into Performance Cues in the Lyrics Box instead of piling on adjectives — that's where a per-section arc belongs. (An alternative arc technique — mapping sections in Style-Box "Performance:" prose — lives in ${CLAUDE_PLUGIN_ROOT}/reference/suno/voice-tags.md § Emotion Arc Mapping; use one approach per track, not both.)

Exclude Styles (Negative Prompting)

Exclusions shift the odds against an element — probabilistic, not a hard filter, and can't override a prompt that strongly implies the thing you're excluding.

Where they go:

  • Pro/Premier → Suno's dedicated Exclude Styles field (Custom Mode → Advanced Options). The reliable path.
  • Inline `no [element]` in the Style Box does not work on v6 — a launch-week test saw "no drums" ignored (The Verge). Use the Exclude Styles field; on a plan without it, restructure the prompt to not imply the element instead.

Rules:

  • Max 2–4 items — over-specification dilutes the effect
  • Bare elements, never "no": drums, electric guitar, autotune. The field is the negation; no drums in it is redundant at best, and if the value ever lands in the Style Box it asks for drums
  • Suppressing unwanted group vocals (a common Suno over-add): choir, crowd vocals, backing vocals, gang vocals, call-and-response, vocal harmonies, layered vocals — still probabilistic; pair with a leaner style prompt
  • Always record them in the track file's Exclude Styles section, even when none apply — write ### Exclude Styles followed by (none) so downstream tools can confirm the field was considered, not silently skipped. Most tracks land here.

Auto-populate guidance: Consider whether genre/instrumentation context implies exclusions:

  • Acoustic folk → electric instruments, drums
  • A cappella → instruments
  • Lo-fi chill → aggressive vocals

Only add exclusions when there is a clear reason.

See ${CLAUDE_PLUGIN_ROOT}/reference/suno/best-practices.md § Negative Prompting for full details.

Generation Settings

Fill the track's ### Generation Settings table (Advanced Mode → More Options). Defaults are in templates/track.md; the catalog is ${CLAUDE_PLUGIN_ROOT}/reference/suno/models.md.

SettingRule
Modelv6 for finished tracks. v6-wild for a first pass on an undecided sound, or on the genres where v6 is reported weak or generic — grunge, metal, alt-country, synth-pop, and any rock or era-soul vocal that keeps drifting to a stock timbre (in a seven-genre A/B, wild won on character in indie rock, R&B/soul and metal; v6 won fidelity everywhere). Then Cover the keeper on v6 — two Generation Log rows — but treat that step as experimental: nobody has shown a Cover that keeps wild's character. Custom: <name> for album-wide consistency. v6-mini only on a Free account (its output is not releasable).
VarietyOff, always, when the Style Box was engineered. Normal (the default on v6/v6-mini) lets Suno rewrite the box. State the reason in Production Notes if you ever raise it.
Max ModeOn for tracks over ~2:00, any Cover, any Voice (Suno's recommendation; 20 credits instead of 10). Off for short ideas and sketches. Untested on v6-wild — log the result if you try it.
Vocal GenderFrom the track's vocal description; — when a Voice or Custom Model supplies the voice.
DurationAuto unless the track or album sets a Target Duration; Custom accepts 10 s–6:00 and hard-cuts at the value, so the lyric load must fit.
Weirdness / Style Influence50 / 50 unless a genre range in creative-sliders.md says otherwise. Style Influence at 50 discards half the prompt's authority — raise it before blaming the prompt.
Audio InfluenceOnly with attached audio (Cover / Voice): ~0.70–0.85 for a Voice; see the Cover / Voice Setup block.

Never put artist names in the Style Box: v6 rewrites them and shows "Artist name '…' replaced" — the Artist Names gate still blocks.


Genre Selection

More specific = better results, but stop at 2-3 genre descriptors. Over-specification (5+ genre terms) dilutes rather than clarifies.

Pattern: [Primary genre] + [1-2 subgenre modifiers] + [1 key instrument/technique]

This is the start of the file. Read the full SKILL.md on GitHub.

Security

  • Security scanPassed, 100 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
151
Repository forks
28
7-day trend score
25

Stars and forks are for the whole GitHub repository, not this skill alone. Refreshed 5 May 2026, 5 months ago.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/bitwize-music-studio/claude-ai-music-skills/suno-engineer/badge)](https://verified-skill.com/skills/bitwize-music-studio/claude-ai-music-skills/suno-engineer)

Something wrong with this skill or its page? Tell us.