music

name	music
description	Use when a bot or assistant needs to satisfy music listening requests, song search, 点歌, 听歌, broad artist requests like “xxx 的歌”, Meting API, NetEase music lookup, selecting a song from search results, downloading a song URL, converting audio with ffmpeg, or sending an mp3/audio file to a user.

Music Bot via Meting API

Overview

Use this skill to implement a conversational music request flow: infer a song query, search NetEase Music, show up to 10 choices, accept a follow-up selection by number/name/random intent, resolve the selected song through a Meting API backend, download it, convert it to MP3 with ffmpeg, and send the resulting audio/file.

The Meting API endpoint resolves song IDs to playable resources; it is not the search endpoint. Search NetEase first, then call Meting with the selected song ID.

When to Use

Use this when the user asks a bot to play or find music, for example:

• “我想听 alanwalker 的歌”
• “点一首周杰伦”
• “搜一下 faded，给我听第二首”
• “随便选一首你喜欢的” after a song list was shown
• Any bot feature that needs api.injahow.cn/meting, Meting API, NetEase song IDs, ffmpeg conversion, mp3 output, or voice/audio sending

Do not use this for music theory, lyrics explanation, playlist writing, or recommendations that do not need actual audio retrieval.

Conversation State

Keep a short-lived per-user or per-channel state after search:

interface MusicSearchState {
  query: string
  results: Array<{ id: number; name: string; artists: string; albumName: string; duration: number }>
  createdAt: number
}

Expire the state after 2-5 minutes. Every follow-up selection must refer to the latest unexpired state for that user/channel.

Request Modes

Classify the user request before deciding whether to play directly or show choices:

User message	Mode	Bot behavior
“我要听 Faded Alan Walker”	Specific song + artist	Search, choose high-confidence match, play directly
“我要听 The Spectre 2.0”	Specific title with version qualifier	Play only an exact/qualified match; otherwise show choices
“我要听 alanwalker 的歌”	Broad artist request	Must show up to 10 choices and wait for user selection before resolving/downloading
“随便来首 alanwalker”	Random request	Search and pick one result

Broad artist/style requests are not specific enough to auto-play. Phrases such as “xxx 的歌”, “xxx 歌手”, “来点 xxx”, or “我想听 xxx 的歌” should run format-list, send the list to the user, store the latest result state, and stop until the user replies with a number/name/random intent.

2. Search NetEase Music

Use the NetEase search API, not Meting, to get candidate song IDs:

GET http://music.163.com/api/search/get/web?csrf_token=hlpretag=&hlposttag=&s=<encoded query>&type=1&offset=0&total=true&limit=10

Map each result to:

{
  id: song.id,
  name: song.name,
  artists: song.artists.map(a => a.name).join('/'),
  albumName: song.album.name,
  duration: song.duration
}

Show at most 10 results:

我搜到了这些：
1. Faded - Alan Walker
2. Alone - Alan Walker
3. The Spectre - Alan Walker
你想听哪首？可以回复序号、歌名，或者说“随便”。

3. Auto-play exact song requests or resolve selection

Only auto-play when the request has exactly one safe match: an exact title match with no competing same-title results, or a title+artist match that preserves every requested qualifier. If the query only loosely matches the top result, or if the same title appears with multiple artists, show choices instead of playing the first result.

If the request contains version/variant qualifiers such as 2.0, remix, live, DJ, 伴奏, or cover, do not drop those words during matching. The original song is not a high-confidence match for a qualified request unless the title itself includes the requested qualifier.

For broad requests such as “我要听 alanwalker 的歌”, do this instead:

1. Run search "alanwalker" --limit 10 to get JSON candidates with numeric song IDs for internal state.
2. Run format-list "alanwalker" --limit 10 and send the returned plain text list to the user.
3. Save the search JSON as the latest per-user/per-channel state so numbers and titles map back to song IDs.
4. Wait for the user to reply with a number, title, or random intent.
5. Only after the selection, call download <songId> --out /download/<file.mp3> or resolve the chosen id and convert it.

If the initial user message is broad and does not include “随便/你选/随机/都行”, asking the user to choose is the correct behavior even if the search result has a clear first item.

If the initial user message contains a specific song title, search first and then decide whether to ask for confirmation:

Search result situation	Behavior
One safe match: exact title match with no competing same-title result, or title plus artist match that preserves every requested version/variant qualifier	Play it directly without asking another question
No exact/safe match, only a likely top search result	Show up to 10 candidates and ask the user to choose
Multiple likely versions: original/live/remix/DJ/伴奏/cover, or several same-title songs	Show up to 10 candidates and ask the user to choose
Query is only an artist or broad style, such as “alanwalker 的歌”	Show candidates and ask the user to choose, unless the user asked for random
User says “随便/你选/随机” in the initial request	Search and pick one candidate, preferably not always index 0

When auto-playing, still send a short confirmation before audio processing:

我找到最匹配的是：Faded - Alan Walker，开始处理音频。

For follow-up selection after a list, accept these forms:

Input	Behavior
1, 第一个, 第二首	Select by ordinal
exact or partial song title	Select the best matching result by title/artist text
随便, 你选, 随机, 都行	Pick one result, preferably not always index 0

If ambiguous, ask one clarification instead of guessing. If no active search state exists, ask the user what they want to hear.

4. Get playable URL via Meting

For the selected id, request:

GET https://api.injahow.cn/meting/?type=url&id=<songId>

The response may be either a plain URL or JSON containing url. Accept only http: or https: URLs.

Other useful Meting calls:

GET https://api.injahow.cn/meting/?type=song&id=<songId>
GET https://api.injahow.cn/meting/?type=pic&id=<songId>
GET https://api.injahow.cn/meting/?type=lrc&id=<songId>

5. Download and convert to MP3

Download the playable URL to a temp file, then convert with ffmpeg:

ffmpeg -y -hide_banner -loglevel error -i "input_file" -vn -acodec libmp3lame -b:a 192k "output.mp3"

The reusable helper first checks ffmpeg in PATH. If it is missing on Linux, it downloads an ffmpeg build into music/.cache/ffmpeg/ and reuses that cached binary for later conversions.

Use timeouts and file size limits appropriate for the bot platform. Always clean up temporary files after sending or failure.

All final audio files must be written under /download. Pass an output path such as /download/song.mp3; the helper rejects paths outside /download. Before each download/play command, the helper deletes files under /download that are older than 24 hours.

6. Send result

Send a short confirmation and attach or upload the converted MP3:

给你放：Faded - Alan Walker

If the bot platform supports voice messages only for specific codecs, convert again as required by that adapter; otherwise send the MP3 as an audio/file attachment.

Reusable Helper Script

If you can run Python, use scripts/meting_music.py for deterministic operations instead of rewriting the flow:

python scripts/meting_music.py doctor
python scripts/meting_music.py search "alanwalker" --limit 10
python scripts/meting_music.py format-list "alanwalker" --limit 10
python scripts/meting_music.py choose-best "Faded Alan Walker" --limit 10
python scripts/meting_music.py resolve 36990266
python scripts/meting_music.py download 36990266 --out /download/song.mp3
python scripts/meting_music.py play "Faded Alan Walker" --out /download/song.mp3

resolve only returns the playable direct URL and the Meting API that worked. Use download or play when the bot needs an actual MP3 file.

Use search together with format-list for broad artist/style requests: keep the JSON from search internally for song IDs, send only the format-list text to the user, then stop and wait for selection. Do not immediately call play for “xxx 的歌” unless the user also says “随便/随机/你选”. It returns plain chat text, not JSON:

我搜到了这些和「alanwalker」相关的歌曲：
1. Faded - Alan Walker (3:32)
2. Alone - Alan Walker (2:41)
你想听哪首？可以回复序号、歌名，或者说“随便”。

Use play only when the user gave a specific enough song request, or explicitly asked for random selection. play will still return an ambiguous: true JSON response instead of writing a file when the search only has a loose top result, multiple same-title songs by different artists, or a requested qualifier is missing. Send only the prompt text to the user in that case. If it is not ambiguous, it writes the MP3 file and returns the file path.

For command-runner bot backends, set the command timeout to at least 120 seconds for download or play. The first Linux run may need extra time to download ffmpeg, and normal song download + conversion can exceed 30 seconds. If the backend cannot change timeout, tell the user you are processing audio before running the command and tolerate a delayed file publish step instead of treating a 30-second timeout as failure when JSON output and the MP3 path are present.

resolve, download, and play use multiple preset Meting APIs as fallback. To override or add priority APIs:

python scripts/meting_music.py resolve 36990266 --api https://api.injahow.cn/meting/ --api https://api.qijieya.cn/meting/

doctor checks Python, ffmpeg auto-setup, NetEase search, and Meting API availability before deploying the bot.

Error Handling

Failure	Bot response
Search returns empty	“没搜到相关歌曲，换个关键词试试？”
Search/API timeout	“音乐服务暂时没响应，稍后再试。”
Selection invalid	“这个序号/歌名不在刚才的列表里，请重新选。”
Meting URL missing	“这首歌暂时拿不到播放链接，换一首？”
All Meting fallback APIs fail	“音乐直链服务暂时不可用，稍后再试。”
ffmpeg auto-setup failed	“音频转换环境自动配置失败，需要手动安装 ffmpeg。”
Download/ffmpeg failed	“音频处理失败，换一首或稍后再试。”
Song too long or file too large	“这首太长了，可能发不出去，换一首短一点的？”

Common Mistakes

• Do not call ?type=url&id=<keyword> with a search keyword. Meting URL lookup requires a numeric song ID.
• Do not show more than 10 results in chat; long lists make follow-up selection unreliable.
• Do not lose the search state before the user chooses; selection by number needs the previous result list.
• Do not assume the user will only reply with a number; support song name and random choice.
• Do not send arbitrary remote URLs as final output when the requirement is an MP3/audio file; download and convert first.
• Do not auto-play the default top search result; if there is no exact/safe match, show the candidate list and wait for selection.
• Do not treat a 30-second command timeout as failure if the command output contains valid JSON with a generated MP3 path.