music-recommender

name	music-recommender
description	Analyze NetEase Cloud Music (网易云音乐) playlist and recommend songs matching their taste. Use when user asks for music recommendations, wants a daily playlist, says '推荐音乐', '今日歌单', 'music', or shares a NetEase playlist/album link. Recommend once per day, never repeat previously recommended songs. Supports free Bilibili links.

Music Recommender

Analyze a user's NetEase Cloud Music playlist, profile their taste, and recommend songs with clickable Bilibili links (free, no membership required).

Workflow

Step 1 — Parse Playlist

Extract playlist ID from user's link. Supported formats:

• https://music.163.com/playlist?id=XXXXX
• https://music.163.com/#/playlist?id=XXXXX

Run the fetch script:

python3 {baseDir}/scripts/fetch_playlist.py <playlist_id> > /tmp/playlist_<id>.json

Output: JSON array of {name, artists, album} objects.

Step 2 — Analyze Taste

Read the JSON output. Profile the user's taste:

1. Top artists — count occurrences, identify top 10-20
2. Language mix — estimate Chinese/English/Japanese/Korean ratio from song titles
3. Genre tags — infer from artists and song names (e.g. 气声唱法, 90s怀旧, indie folk, dream pop)
4. Era — identify decade distribution
5. Mood — upbeat/melancholic/dreamy/energetic based on song names and artists

Summarize the taste profile in 3-5 bullet points.

Step 3 — Recommend

Based on the taste profile, recommend 10 songs that:

• Match the user's preferences (similar artists, genres, mood)
• Are NOT already in their playlist
• Are diverse — mix of Chinese and foreign, different sub-genres
• Include both well-known and lesser-known picks

For each recommended song, search Bilibili for a playable link:

python3 {baseDir}/scripts/search_bilibili.py "<artist> <song> 官方MV"

Output: BV_ID|TITLE|URL

Step 4 — Format Output

Present the recommendations as a plain text list (NOT HTML/markdown links) for Telegram compatibility:

🎵 Vulpis 今日推荐歌单

**华语女声：**
1. 陈粒 — 奇妙能力歌
https://www.bilibili.com/video/BVxxxxx

2. ...

**欧美梦幻：**
6. ...

Rules for Telegram formatting:

• Use bold for section headers, NOT markdown links [text](url)
• Put URL on its own line after the song name
• Group by genre/language (华语/欧美/日语 etc.)
• Keep descriptions short (5-10 words)

Step 5 — Record & De-duplicate

IMPORTANT: Only recommend ONCE per day. Before recommending:

1. Check if today's recommendation file exists:

   ~/.openclaw/workspace/music-history/YYYY-MM-DD.json
   

   If it exists, reply with today's list and say "今天已经推荐过了". Do NOT generate a new one.

2. If not, load the full history to avoid repeats:

   python3 {baseDir}/scripts/history.py show
   

   This outputs all previously recommended songs across all days.

3. When generating recommendations, exclude any song that appears in the history.

4. After generating, save today's recommendations:

   python3 {baseDir}/scripts/history.py save
   

   Pipe in JSON array: [{"name":"...","artists":"...","bvid":"...","url":"..."}]

History Storage

~/.openclaw/workspace/music-history/
├── 2026-03-29.json
├── 2026-03-30.json
└── ...

Each file: JSON array of recommended songs for that day.

Step 6 — (Optional) Extra Save

If user wants to save elsewhere, offer to:

• Write to Notion (Content Calendar or a Music DB)
• Generate an HTML page in the workspace
• Create a text file in the workspace

Notes

• NetEase API endpoint: https://music.163.com/api/v6/playlist/detail?id=<ID>&n=1000
• Required headers: User-Agent: Mozilla/5.0, Referer: https://music.163.com/, Cookie: os=pc;
• Artist field is ar (not artists) in NetEase API response
• Bilibili search API: https://api.bilibili.com/x/web-interface/search/all/v2?keyword=<query>
• Required headers for Bilibili: User-Agent: Mozilla/5.0, Referer: https://www.bilibili.com/
• Default recommendation count: 10 songs
• Always use Bilibili links (free, no membership) instead of NetEase links