beatra-ai/beatra-skills/beatra-ai-music-creator
Claude Code and agent skill by beatra-ai/beatra-skills
- community
Checking that the SKILL.md file is still in the repository.
Create original songs, AI-generated music, lyrics-to-song tracks, instrumentals, background music, video soundtracks, jingles, multilingual songs, and reference-led arrangements from a clear creative brief. This AI music generator and AI song maker develops genre, mood, structure, singable lyrics, vocal direction, and production style, then creates reviewable audio with Suno 6 or another model you explicitly choose. Use it as an AI music studio for songwriting, an AI lyrics writer, text-to-music creation, BGM generation, brand music, podcast themes, game music, bilingual songs, or focused new versions of reference audio. Review vocals, pronunciation, duration, loop points, and arrangement fit after generation, then refine the strongest result.
Install
npx vskill@latest install beatra-ai/beatra-skills/beatra-ai-music-creatorSKILL.md
The same SKILL.md was published earlier as beatra-ai/beatra-skills/beatra-ai-music-creator.
AI Music Creator
Develop a theme, scene, lyric fragment, complete lyric, functional music brief, or reference recording into one coherent musical direction and, when the user wants audio, one intentional generation request. The deliverable may be a song, instrumental, BGM, soundtrack, jingle, multilingual track, or reference-led new arrangement. Treat singer identity, pronunciation, loop points, melody carryover, and mastering as qualities to review after generation rather than guaranteed properties. A length is a control when this account has the music tools (see music editing and audio).
Use the bundled scripts/mcp_client.py for every Beatra operation. Do not configure or use a host Beatra Connector. Never use REST/OpenAPI as a fallback. For ordinary calls, run python3 scripts/mcp_client.py call <tool-name> and pass one JSON object on stdin. For local files, use only the dedicated upload command described below. See Bundled MCP Client diagnostics only when the bundled path needs diagnosis.
Shape the direction first
At the studio entrance, split first: original song, cover, gift song, or instrumental BGM. This package defaults to an original song. Route a cover or re-arrangement of an existing recording to ai-song-cover-studio, a gift or occasion song to personalized-song-maker, and pure instrumental BGM to short-video-bgm-studio.
Reuse every settled choice in the conversation. Infer ordinary creative details from the destination when that is safe. The hard inputs are a theme or lyrics plus the intended use. If genre is empty, ask before writing the production card. Ask only when a missing answer changes the lyrics, reference direction, vocal or instrumental route, exact paid payload, or another high-impact user choice.
Follow this golden path: lock structure, write an executable caption, generate a limited set of takes, then make one focused revision. Never extend automatically or repeatedly; each revision is its own confirmed call.
Before generation, prepare one compact production card:
- destination and listener;
- title, premise, emotional arc, primary genre, tempo feel, instrumentation,
structure, mix direction, ending or loop direction, and exclusions;
- vocal or instrumental route;
- full lyrics, language and register, and vocal direction when the track has
vocals;
- selected model and model-specific controls;
- reference intent and the qualities that should carry over or change when a
reference is involved; and
- the fact that the next action is one billable generation.
For instrumentals, omit lyrics and make room for dialogue or other functional needs in the creative direction. For vocal music, finish and show material lyric changes before generation. Lyrics writing, creative planning, model discovery, prompt preparation, and critique do not require a paid music call. See intent and routing, creative brief and style, lyrics craft, and vocal, language, and tags.
When a brief uses artist shorthand, translate it into actionable genre, era, tempo, instrumentation, harmony, vocal texture, phrasing, and mix attributes.
Choose and validate the model
Set model: "suno-6" for this package's ordinary generation. Never omit the model and never silently use auto. Before every production card, call beatra.models.list with a copyable payload:
When the route uses a reference recording, also call it with:
Read the pricing of the chosen model's entry (suno-6 unless the user named another eligible model) and show those credits on the card per slot and in total. Never a remembered number. Call it again when the user requests another model or needs a model-specific capability. Do not silently substitute a model.
Length: when this account has the music tools (the bundled client's tools command lists beatra.music.edit and beatra.audio.process; see music editing and audio), send duration_seconds when the user names a length (not with a reference recording; fit a reference-guided track afterwards with trim or loop). A song with lyrics gets duration_seconds of 30 or more that the lyrics fit, without duration_fit: a shorter target, or exact, cuts the singing at the target, so cut afterwards with beatra.audio.process trim only when the user accepts losing the end. An instrumental that needs exact seconds adds duration_fit: "exact", and a bed that must run longer or loop (generated with duration_seconds of 30 or more and no duration_fit; see the shared reference) is lengthened with beatra.audio.process. Creative brief and style covers background music. Without the tools, keep the length as direction in the prompt and do not add a duration argument. Always report the real returned length.
Use only controls returned or documented for the selected model family. Suno and MiniMax options are not interchangeable. Validate the exact payload using model routing and use the examples in music recipes as patterns rather than fixed claims.
Use reference audio
For a local FLAC, MP3, or WAV reference, run only:
The bundled command obtains and validates the upload grant, uploads exactly the file bytes, and returns the artifact to use as reference_audio. Do not replace it with host HTTP, a connector, REST, or a hand-written grant and PUT sequence. The general upload limit is 100 MB; the selected music model can impose a lower size or duration limit.
Describe what should carry over and what should change. A reference guides a new creative result; after generation, review melody, vocal character, energy, instrumentation, and arrangement against that direction.
Confirm the paid boundary once
Even a direct, well-specified request gets the production card below first. Approval of the complete card authorizes that exact request once; do not add a second confirmation. An estimate, comparison, lyric review, direction choice, or “do not generate yet” is not approval.
Show one current production card and wait:
- Work — what will be made in the user's words, a
beatra.music.generatecall,
plus any finishing step the user asked for with it.
- Credits — the live price of the chosen model (
suno-6by default) just
read from beatra.models.list, per slot and in total, with each finishing step as its own line at its live audio_process price; the total includes every line on the card. Never a remembered number.
- Count — one paid call for this generation, plus one per finishing line on
this card. A later revision is another slot.
- Identity — one new opaque
client_request_idper line listed on the card. Any changed prompt, lyrics,
file, model, or control mints a new ID.
- If we stop here — the production card and lyrics remain usable.
- If the balance is insufficient — relay the official message and its top-up
URL exactly. Translate the prose; keep the URL. Do not retry until the user says they have topped up.
Before a billable call, make model, title, lyrics or instrumental status, reference, important controls, and the paid calls the card lists visible. Create a stable 1..128-character client_request_id only after that payload is final. Any change to prompt, lyrics, instrumental flag, title, model, reference, model options, or any other argument accepted by the current or a future MCP generation schema is a new paid request with a new identity and confirmation. Fields outside the published MCP schema are ignored and do not change the request identity or task.
Let the bundled client perform its cached, best-effort, non-billable beatra.installations.register step automatically. Do not add manual registration to the creative workflow.
Submit beatra.music.generate once. Retain its task_id and poll the same task with beatra.tasks.get. Honor a returned deadline_at; otherwise stop active polling after 30 minutes, report the current state and how to resume, and do not resubmit. Cancel only when the user asks via beatra.tasks.cancel; if cancel conflicts with a terminal transition, return to the same task.
For an uncertain submission or lost task ID, search recent work with beatra.tasks.list using the matching music capability. List results do not contain the complete request: call beatra.tasks.get for each plausible candidate and compare its task.input, resolved model, reference, and options with the saved full payload. Only an exact identical retry may reuse the same client_request_id. Never create replacement paid work merely because a response was lost or a task is slow. See tasks and results and billing, errors, and recovery.
Deliver and review the music
On success, present every task.output.clips entry in returned order. Include the returned title and lyrics when present plus clip.audio.url, artifact ID, duration, MIME type, and size. Report factual task identity and billing.net_charged_credits. When the task returns task.links.assets, use that exact destination for asset management; do not invent a generic URL.
Review composition, lyrics, vocal performance, pronunciation, role assignment, arrangement, ending, and production only when the host can actually play the audio. Otherwise label listening review incomplete while still delivering the factual artifacts. Preserve successful qualities and turn the largest gap into one focused, newly approved step: a new generation or, when this account has the music tools (see Choose and validate the model), beatra.music.edit with operation: "replace_section" to fix one part of a chosen song or operation: "extend" to continue it. Each is its own paid call with its own card, the live price for that step, and its own client_request_id; never extend automatically or repeatedly. Follow review and iteration.
When the user asks how many credits remain, call beatra.wallet.get. When they ask what was charged, call beatra.wallet.ledger. Both are read-only. Do not invent an account-balance or top-up tool. Do not make wallet.get a required step before every paid submit.
References by task
This is the start of the file. Read the full SKILL.md on GitHub.
Security
- Security scanPassed, 95 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/beatra-ai-music-creator/SKILL.md
Popularity
- Repository stars
- 2
- 7-day trend score
- 10
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
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
[](https://verified-skill.com/skills/beatra-ai/beatra-skills/beatra-ai-music-creator)