What Shall We Build Next?

Describe

Describe your task

Refine

Refine the plan

SubAgents

Review all agents

Deploy

Deploy your agent

Let me break down the ScraperAgent workflow in detail:

A) SUBAGENT SUMMARY: 
A web scraping agent that visits Tracklib's new arrivals page, extracts detailed information about the 20 most recent songs, and structures this data for filtering by other agents.

B) FINAL TASK OUTPUT:
A structured JSON array containing 20 song entries, each with:
- Title
- Artist
- Release date
- Genre/style
- Sample category/price tier
- Tracklib URL
- Duration
- BPM (if available)
- Key (if available)

C) SUBAGENT INPUT:
- Base URL: "https://www.tracklib.com/music/new-arrivals"
- Number of songs to extract (20)

E) SUBAGENT TASK SUMMARY:
1. Initial Website Analysis:
   #174 (Take screenshot of website) > 
   #176 (Analyze Image with GPT Vision) to verify page layout and content structure

2. Data Extraction:
   #226 (Extract Structured Data From 1x URL) with specific instructions to extract:
   - Song titles
   - Artists
   - Release dates
   - URLs
   - Additional metadata

3. Data Enrichment:
   #224 (Oracle Ask A Question) to fill in any missing metadata by querying specific song URLs

F) SILOS:
SILO 1: INITIAL PAGE CAPTURE
- Input: Tracklib URL
- Skill: #174 (Screenshot website)
- Output: PNG of page

SILO 2: PAGE ANALYSIS
- Input: PNG from Silo 1
- Skill: #176 (GPT Vision analysis)
- Output: Text describing page structure and content locations

SILO 3: DATA EXTRACTION
- Input: Original URL + structure information from Silo 2
- Skill: #226 (Extract structured data)
- Output: Initial song data

SILO 4: DATA ENRICHMENT
- Input: Song URLs from Silo 3
- Skill: #224 (Oracle questions for missing data)
- Output: Final enriched JSON data structure with all 20 songs

This workflow ensures robust data extraction with verification steps and enrichment, providing clean, structured data for the next subagent in the chain.

SubAgent #1 - Diagram

Expand Diagram

Here's my complete analysis and workflow for SUBAGENT 2 (FilterAgent):

A) SUBAGENT SUMMARY:
A filtering system that takes 20 recently scraped songs and user requirements, then uses an LLM to analyze and rank them, outputting the 5 most relevant matches.

B) FINAL TASK OUTPUT:
A structured JSON array containing exactly 5 songs, each with fields for {title, artist, tracklib_url, relevance_score, matching_criteria}. The relevance_score will be a float between 0-1, and matching_criteria will explain why this song matches the user's requirements.

C) SUBAGENT INPUT:
1. An array of 20 songs (each containing title, artist, tracklib_url) from ScraperAgent
2. User's descriptive query text (e.g., "Looking for 1970s funk tracks with prominent horn sections")

D) SUBAGENT TASK SUMMARY:
Input > #223 (Initial Analysis) > #223 (Detailed Scoring) > #223 (Final Ranking) > Output

Detailed Flow:
1. First #223 call:
- INPUT: 20 songs + user query
- PROMPT: "Analyze these songs against the user requirements. For each song, generate initial matching criteria."
- OUTPUT: Initial analysis text for all 20 songs

2. Second #223 call:
- INPUT: Initial analysis + user query
- PROMPT: "Score each song from 0-1 based on matching criteria. Explain scoring rationale."
- OUTPUT: Scored list with explanations

3. Third #223 call:
- INPUT: Scored list
- PROMPT: "Select and rank the top 5 songs. Format output as JSON with {title, artist, tracklib_url, relevance_score, matching_criteria}"
- OUTPUT: Final JSON array of 5 songs

E) SILOS:
SILO 1: INITIAL ANALYSIS
- Purpose: First-pass analysis of all 20 songs
- Input: 20 songs + user query
- Skill: #223
- Output: Initial matching criteria for each song

SILO 2: SCORING
- Purpose: Detailed scoring of each song
- Input: Initial analysis from Silo 1
- Skill: #223
- Output: Scored list with explanations

SILO 3: FINAL SELECTION
- Purpose: Select and format top 5 matches
- Input: Scored list from Silo 2
- Skill: #223
- Output: Final JSON array of 5 best matches

This workflow ensures:
1. Thorough analysis of all songs
2. Consistent scoring methodology
3. Clear explanation of why each selected song matches
4. Structured output ready for the next subagent
5. No unnecessary skills/steps
6. Clear separation of concerns in silos

SubAgent #2 - Diagram

Expand Diagram

Let me break down the DownloaderAgent workflow in detail:

A) SUBAGENT SUMMARY:
A specialized agent that locates and downloads MP3 versions of five pre-selected songs from YouTube, ensuring high-quality audio retrieval for further processing.

B) FINAL TASK OUTPUT:
5x MP3 audio files (URLs) of complete songs, each saved to the server with consistent audio quality (typically 128-320kbps, stereo), with durations matching the original songs.

C) SUBAGENT INPUT:
- List of 5 songs (each containing artist name, song title)
- Optional: Release year or additional metadata to ensure correct version

E) SUBAGENT TASK SUMMARY:
For each of the 5 songs, the following chain occurs:

1. Query Formation:
#223 (LLM) to format search query: "{song title} {artist} official audio youtube"

2. URL Discovery:
#224 (Oracle) to find exact YouTube URL using formatted query, ensuring:
- Matches artist/title exactly
- Preferably official channel
- Full song (not remix/live)

3. Download:
#193 (YouTube-to-MP3) to download complete audio

This chain repeats 5 times (once per song)

F) SILOS:
The subagent operates in 5 parallel silos, one for each song:

SILO 1: Song 1
- Input: Song 1 metadata
- #223: Format search query
- #224: Find YouTube URL
- #193: Download MP3
- Output: Song 1 MP3 URL

SILO 2: Song 2
[Same process as Silo 1]

SILO 3: Song 3
[Same process as Silo 1]

SILO 4: Song 4
[Same process as Silo 1]

SILO 5: Song 5
[Same process as Silo 1]

Each silo operates independently and can run concurrently, with all 5 final MP3 URLs collected at the end for passing to the next subagent.

This structure ensures reliable, high-quality audio retrieval while handling each song independently to prevent errors in one download from affecting others.

SubAgent #3 - Diagram

Expand Flow

Let me break down the StemExtractorAgent following these guidelines:

A) SUBAGENT SUMMARY: 
A specialized audio processing agent that takes 5 MP3 files and separates each into vocal and instrumental stems, producing 10 distinct audio files while maintaining audio quality and file organization.

B) FINAL TASK OUTPUT:
10 MP3 files (2 per original song):
- 5x vocal stem MP3s
- 5x instrumental stem MP3s
Each file properly labeled with original song name + stem type (vocal/instrumental)

C) SUBAGENT INPUT:
- 5x MP3 URLs from the DownloaderAgent
- Original song metadata (titles/artists) for proper file labeling

D) SUBAGENT TASK SUMMARY:
Input (5x MP3s) > #197 Extract Instrumental & Vocal (repeated 5 times, once per song) > Output (10x MP3 stems)

E) SILOS:
The agent operates in 5 parallel silos, one for each song:

SILO 1 (Song 1):
1. Input: Song 1 MP3
2. Action: #197 Extract Instrumental & Vocal
3. Output: 2x MP3s (Song1_Vocal.mp3, Song1_Instrumental.mp3)

SILO 2 (Song 2):
1. Input: Song 2 MP3
2. Action: #197 Extract Instrumental & Vocal
3. Output: 2x MP3s (Song2_Vocal.mp3, Song2_Instrumental.mp3)

SILO 3 (Song 3):
1. Input: Song 3 MP3
2. Action: #197 Extract Instrumental & Vocal
3. Output: 2x MP3s (Song3_Vocal.mp3, Song3_Instrumental.mp3)

SILO 4 (Song 4):
1. Input: Song 4 MP3
2. Action: #197 Extract Instrumental & Vocal
3. Output: 2x MP3s (Song4_Vocal.mp3, Song4_Instrumental.mp3)

SILO 5 (Song 5):
1. Input: Song 5 MP3
2. Action: #197 Extract Instrumental & Vocal
3. Output: 2x MP3s (Song5_Vocal.mp3, Song5_Instrumental.mp3)

Each silo operates independently and in parallel, using skill #197 to process one song at a time. The final output collection of all 10 stems is passed to the next subagent (TempoAnalyzerAgent) for further processing.

This is one of the simpler subagents in the workflow, as it primarily relies on a single skill (#197) repeated 5 times in parallel. The main complexity comes from managing the parallel processing and ensuring proper file organization/labeling of the resulting stems.

4 Template & Links

Expand Flow

Let me break down the TempoAnalyzerAgent according to your format:

A) SUBAGENT SUMMARY:
A specialized agent that processes 10 audio stems (5 instrumentals, 5 vocals) to extract precise tempo and beat marker data, enabling accurate 16-beat sample extraction in later stages.

B) FINAL TASK OUTPUT:
JSON object containing 10 entries (one per stem), each with:
- Stem ID (1-10)
- Stem Type (instrumental/vocal)
- BPM (float number)
- Beat Markers Array (array of timestamps in seconds)
- Downbeat Markers Array (array of timestamps in seconds)
- Total Beat Count
- Audio Duration (seconds)

C) SUBAGENT INPUT:
- Array of 10 MP3 URLs (5 pairs of instrumental/vocal stems)
- Stem metadata (identifying which is instrumental/vocal)

D) SUBAGENT TASK SUMMARY:
For optimal processing, we need to:
1. Convert MP3s to WAV for better analysis:
   Input: 10x MP3 URLs > #178 (Convert MP3s to WAV)

2. For each WAV file:
   WAV URL > #180 (Extract Beatpoints & Tempo)

3. Generate visual confirmation:
   WAV URL > #179 (Create Visual Waveform)

4. Verify waveform accuracy:
   Waveform URL > #176 (Analyze Image with GPT Vision)

E) SILOS:
The agent operates in two parallel silos, each processing 5 files:

SILO 1: INSTRUMENTAL STEMS (5x files)
For each instrumental stem:
1. #178 - Convert MP3 to WAV
2. #180 - Extract beat/tempo data
3. #179 - Generate waveform
4. #176 - Analyze waveform to verify beat detection
Output: JSON object with tempo/beat data for instrumental

SILO 2: VOCAL STEMS (5x files)
For each vocal stem:
1. #178 - Convert MP3 to WAV
2. #180 - Extract beat/tempo data
3. #179 - Generate waveform
4. #176 - Analyze waveform to verify beat detection
Output: JSON object with tempo/beat data for vocal

Each silo processes its 5 files in parallel, then the results are combined into the final JSON output containing all 10 entries. This parallel processing ensures efficiency while maintaining organized data structure between instrumental and vocal stems.

The final output JSON structure ensures that the next subagent (ChopAndPackAgent) has precise timing data to create accurate 16-beat samples.

5 Template & Links

Expand Flow

Let me break down SUBAGENT 6 (ChopAndPackAgent) according to the guidelines:

A) SUBAGENT SUMMARY:
Takes 10 audio stems (5 instrumentals, 5 vocals) plus their beat/tempo data, extracts precise 16-beat segments from each, and packages them into a single organized ZIP file.

B) FINAL TASK OUTPUT:
A single ZIP file containing 10 WAV audio files (named consistently like "song1_instrumental_16beats.wav" and "song1_vocal_16beats.wav"), each precisely 16 beats in length, cut exactly on beat markers.

C) SUBAGENT INPUT:
- 10 MP3 URLs (5 pairs of instrumental/vocal stems)
- Beat/tempo data for each stem (containing exact beat markers and BPM)

E) SUBAGENT TASK SUMMARY:
Input > #178 (Convert MP3s to WAV) > #219 (Cut audio pieces) > ZIP creation > Output

Detailed flow:
1. Convert all MP3s to WAV format (#178) for precise cutting
2. For each of the 10 stems:
   - Calculate exact duration needed for 16 beats using BPM data
   - Use #219 to cut precise 16-beat segment starting from first beat marker
3. Package all WAV files into ZIP (external function needed)

F) SILOS:
This subagent naturally breaks into three distinct silos:

SILO 1: FORMAT CONVERSION
• Purpose: Convert all MP3s to WAV format for precise cutting
• Skill Used: #178 - Convert 1-20 MP3s to WAV
• Input: 10 MP3 URLs
• Output: 10 WAV URLs

SILO 2: AUDIO CHOPPING (runs 10 times, once per stem)
• Purpose: Extract precise 16-beat segments
• Skill Used: #219 - Cut WAV/MP3 Audio into Multiple Pieces/Samples
• Input: 
  - 1 WAV URL
  - Beat marker data
  - BPM data
• Output: 1 chopped WAV file (16 beats exactly)
• Process:
  1. Calculate duration needed (16 beats ÷ BPM × 60 = duration in seconds)
  2. Extract segment starting at first beat marker for calculated duration

SILO 3: PACKAGING
• Purpose: Create final ZIP file
• Skill Required: New ZIP creation function
• Input: 10 chopped WAV files
• Output: Single ZIP file containing all samples
• Process:
  1. Rename files consistently
  2. Create ZIP with organized structure
  3. Return final ZIP URL

This subagent will need to process each audio file sequentially through silos 1 and 2, collecting all outputs, before finally running silo 3 to package everything together.