Music Samples from Text Description

I want to extract the latest samples from the website tracklib and grab samples from the most recent 3 songs, based on a rough description of the samples I want. User will enter a brief description (the only input) and the agent visits tracklib and then grabs my samples. To do this we will need to crawl https://www.tracklib.com/music/new-arrivals and extract, say, the latest list of 20 songs, then an LLM will filter them based on my description and find the 3 best matches on youtube, download them and chop them up into samples. we then extract (say) 16 beats worth of audio - beginning from start of the song. We will need to analyse beatpoints and tempo, and then chop the sampes. this will then give us 3x song audio files to chop. then we could cut out samples from those points. Finally it will return the audio samples chopped up in zip files. If there are too many steps here, resulting in too many files/silos/subagents then we can chop fewer samples. I will let you decide, but please be decisive in your plan.

First, pre-entered info...

final-output2

The final output would be an LLM response summarising all the info and files collected for the 3 songs, pulling in the outputs of previous subagents into one single response.

subagents-extracted2

SEQUENCE 1. Subagent 1 (#226) → returns up to 20 latest arrivals (title, artist, link). 2. Subagent 2 (#223) → filters down to 3 songs that match user description. 3. Subagent 3 (#224, #193) → for each filtered track, find a YouTube link and download the MP3. 4. Subagent 4 (#180, #219) → detect tempo/beat info, chop out a 16-beat snippet for each MP3. 5. Subagent 5 (#223) → return summary of all information from sessions 2, 3, 4. SUBAGENT #1 - SCRAPERAGENT • Purpose: Crawl Tracklib's "New Arrivals" page and extract up to 20 song metadata items (title, artist, link). • Skill Used: – #226 (Extract Structured Data From 1x URL) • INPUT: – URL: https://www.tracklib.com/music/new-arrivals (this is can be preset and fixed within the agents preprompt, it will not change) – Instruction: "Extract song title, artist, tracklib_url for up to 20 new arrivals." (user can enter this - input will be [variable1] • OUTPUT: – A structured list of up to 20 items, each with {title, artist, tracklib_url} • SUBAGENT FINAL OUTPUT: [latest-track-metadata] SUBAGENT #2 - FILTERAGENT • Purpose: Based on the meta data collection, filter the song entries based on the user's brief text description, selecting exactly 3 that best match. Prompt could be slightly customized to say this (although provided it includes the below inputs - [variable1] and [latest-track-metadata] - a simple prompt should suffice here. • Skill Used: – #223 (Powerful LLM Prompt-to-Text Response) • INPUT: – The user's description (e.g. "Looking for soulful strings and anything that is multitrack") - [variable1] – The complete list of scraped tracks - [latest-track-metadata] • OUTPUT: – A short text listing of the top 3 chosen {title, artist, tracklib_url}, perhaps returned as JSON for clarity • SUBAGENT FINAL OUTPUT: [filtered-track-list] SUBAGENT #3 - AUDIOLOCATORAGENT x3 • Purpose: For each of the up to 3 filtered entries, find a suitable YouTube URL, then download the audio as MP3. • Skills Used: – #224 (Oracle Ask A Question) to locate a YouTube link by searching for "<title + artist> official" or similar. – #193 (Download Entire YouTube Video As MP3) to download the audio track. – These will need to be run for each of the mp3s (so 6 skills in total used in this subagent) • INPUT: – For each track: query about best YouTube link, then supply that link to #193. • OUTPUT: – Exactly 3 MP3 URLs (one for each selected song). • SUBAGENT FINAL OUTPUT: [mp3-download-urls] SUBAGENT #4 - BEATCHOPPERAGENT x3 • Purpose: Analyze tempo/beat, extract a 16-beat sample for each of the 3 MP3s. • Skills Used: – #180 (Extract Beatpoints & Tempo of MP3) to get BPM and beat offsets. – #219 (Cut Wav/mp3 Audio into Multiple Pieces/Samples) using the offsets from #180 to isolate the first 16 beats. – These will need to be run for each of the mp3s (so 6 skills in total used in this subagent) • INPUT: – Each MP3 file from Subagent 3 - [mp3-download-urls] – Tempo/beat information from #180 (to pass to skill #219) • OUTPUT: – Up to five short MP3/WAV clip URLs (each containing a 16-beat snippet). • SUBAGENT FINAL OUTPUT: [16-beat-clip-urls] SUBAGENT #5 - SUMMARY AGENT x1 • Purpose: Return a summary of everything completed in the session. A very simple task that just returns and formats the 3 select tracks, mp3 URLs and beat sample URLs - into a single simple list. • Skills Used: – #223 (Powerful LLM Prompt-to-Text Response) • INPUT: – A short text listing of the top 3 chosen [filtered-track-list] – Each MP3 file from Subagent 3 - [mp3-download-urls] – URLs of all clipped beat samples - [16-beat-clip-urls] • OUTPUT: – Text list/summary of the mp3 URLs and samples. • SUBAGENT FINAL OUTPUT: [final-summary]

subagentnumber

[subagentnumber]

Then, agent response...

subagent1refined-new-json-core

G) TASKFLOW SUMMARY SKILL 1: - ID: #226 - Title task: Extract Structured Song Data from Tracklib New Arrivals Page - Description of the task: Crawls the Tracklib new arrivals webpage, parses the HTML/DOM content, and extracts structured data for up to 20 most recent songs including title, artist, and tracklib URL for each entry - Input(s) involved: * [url] = "https://www.tracklib.com/music/new-arrivals" * [instruction] = "Extract song title, artist, tracklib_url for up to 20 new arrivals." - Output token name: [latest-track-metadata] This is the only skill/task needed for this particular subagent, as it handles the entire process of: 1. Accessing the webpage 2. Parsing the content 3. Extracting the structured data 4. Returning it in the required JSON format The output [latest-track-metadata] will contain a JSON array like: ```json { "tracks": [ { "title": "Song Title 1", "artist": "Artist Name 1", "tracklib_url": "https://www.tracklib.com/..." }, // ... up to 20 entries ] } ``` This single skill (#226) comprises the entire taskflow for this subagent since it's designed specifically for structured data extraction from a single URL source. The resulting output token [latest-track-metadata] will then be passed to subsequent subagents in the larger workflow.

subagent1mermaid

<!DOCTYPE html> <html> <head> <title>Mermaid Diagram Example</title> <script src="https://cdnjs.cloudflare.com/ajax/libs/mermaid/9.4.3/mermaid.min.js"></script> </head> <body> <div class="mermaid"> graph TD A[Input: URL and Instruction] --> B[Skill #226] B -->|Extract Data| C[Parse HTML/DOM] C -->|Process Content| D[Structure Data] D -->|Format Results| E[Output: latest-track-metadata] subgraph SUBAGENT1[ScraperAgent] B C D end </div> <script> mermaid.initialize({ startOnLoad: true }); </script> </body> </html>

Mermaid Image - https://static.aiz.ac/1738090396-mermaid/mermaid1.png

prompts-required

[prompts-required]

subagent1-json-array

[ { "taskID": 226, "semanticTitleOfTask": "Scrape Tracklib New Arrivals Page to Extract Structured Track Data", "taskDescription": "This task uses skill #226 to scrape the Tracklib New Arrivals webpage, parse the HTML content, and collect up to 20 of the most recent songs, returning their title, artist, and tracklib_url as structured JSON.", "inputDescription": "We must pass the fixed URL (https://www.tracklib.com/music/new-arrivals) and a text instruction specifying which data fields to extract (title, artist, tracklib_url) for up to 20 new arrivals. The skill then fetches the webpage and uses built-in data extraction to parse and return the relevant content.", "inputRequired": ["page-url", "extraction-instruction"], "outputDescription": "After scraping, the skill produces a structured JSON array with up to 20 items, each containing 'title', 'artist', and 'tracklib_url'. This ensures only valid data with complete fields is included. If fewer songs are available, only those are returned.", "outputName": "latest-track-metadata", "promptInstruction": "Analyze the Tracklib “New Arrivals” webpage content and extract data for up to 20 most recent songs. For each, retrieve 'title', 'artist', and the full 'tracklib_url'. Format these entries into a JSON array with valid syntax. If fewer than 20 songs are listed, return only those available. Exclude any tracks lacking complete data. Ensure each URL is fully formed. Return the final JSON result." } ]

subagent1-json-array-mermaid

<!DOCTYPE html> <html> <head> <title>Mermaid Diagram Example</title> <script src="https://cdnjs.cloudflare.com/ajax/libs/mermaid/9.4.3/mermaid.min.js"></script> </head> <body> <div class="mermaid"> graph TD A[Start] --> B[Initialize SubAgent] B --> C[Input: Tracklib URL] B --> D[Input: Extraction Instruction] C --> E[Skill #226] D --> E E --> F[Web Scraping] F --> G[Parse HTML] G --> H[Extract Track Data] H --> I[Format JSON Output] I --> J[Validate Data Fields] J --> K[Filter to 20 Tracks] K --> L[Create latest-track-metadata] L --> M[Return JSON Array] M --> N[End] </div> <script> mermaid.initialize({ startOnLoad: true }); </script> </body> </html>