rapper（rapper 开发）

SenseAudio AI Rapper

Use this skill for SenseAudio music-generation tasks focused on rap, hip-hop, and other vocal-song creation requests.

Read First

• references/music.md

Load the reference only when you need exact endpoint, parameter, polling, or response-shape details.

When To Use

Use this skill when the user wants to:

• generate rap or hip-hop lyrics from a prompt,
• turn lyrics into a full song with vocals,
• control song style, title, vocal gender, or instrumental mode,
• poll async lyric or song generation tasks,
• debug errors around SenseAudio music-generation requests.

Default Workflow

1. Confirm the target output:

• lyrics only,
• full song from user-provided lyrics,
• or full song with platform-generated lyrics.

1. Start with the minimal valid request:

• lyrics generation requires prompt and provider: sensesong.
• song generation requires model: sensesong.

1. Treat both pipelines as potentially asynchronous:

• if create returns task_id, poll the matching pending endpoint,
• wait for status == SUCCESS before reading payload data,
• surface FAILED clearly instead of assuming empty data.

1. Add creative controls only when needed:

• lyrics, title, style, style_weight, negative_tags, vocal_gender, instrumental.

1. Return production-ready output:

• one minimal curl example,
• one requested-language example,
• polling logic,
• and safe response parsing for audio_url, cover_url, lyrics, and duration.

Rap-Specific Guidance

• For rap requests, make the prompt explicit about theme, tone, pacing, and era, such as trap, boom bap, drill, conscious rap, or melodic rap.
• If the user already has bars or a chorus, skip lyric generation and call song creation directly with lyrics.
• Use style for positive steering and negative_tags for exclusions.
• Use vocal_gender: "m" or "f" only when the user asks or when matching a clear creative brief.
• Use instrumental: true only for beat-only output; otherwise omit it.

Implementation Rules

• Keep API keys in environment variables; never hardcode secrets.
• Prefer curl first for reproducibility, then provide Python or JavaScript if useful.
• Do not invent extra request fields or undocumented endpoints.
• Do not assume lyric tags beyond what the docs examples show; if structure syntax matters, mirror the documented examples closely.
• When the docs show semicolon-separated section markers in lyrics, preserve that formatting unless the user provides a different valid lyric format.

Response Parsing Rules

• Lyrics polling success shape: response.data[0].text and optional response.data[0].title.
• Song polling success shape: response.data[0].audio_url, cover_url, duration, id, lyrics, title.
• Guard every array/object access because response or data may be absent before success.

Output Checklist

When producing an answer with this skill, include:

1. which endpoint(s) are used,
2. the minimal valid request body,
3. whether polling is required,
4. the final fields the caller should read,
5. and any rap-specific prompt/style suggestions relevant to the user’s brief.