What Shall We Build Next?

Describe

Describe your task

Refine

Refine the plan

SubAgents

Review all agents

Deploy

Deploy your agent

Let me break this down for SUBAGENT 1: "Bestseller Finder"

A) SUBAGENT SUMMARY:
A crawler and data processor that accepts an Amazon product category, finds its bestseller page, extracts product data, and identifies the highest-priced item, returning structured product information.

B) FINAL TASK OUTPUT:
A JSON object containing:
- product_name (string)
- price (float)
- affiliate_link (URL string)
- details (string with product specifications)
All data saved as a .json file on the server.

C) SUBAGENT INPUT:
- Amazon product category name (string)
- Amazon marketplace identifier (optional, defaults to .com)

E) SUBAGENT TASK SUMMARY:
1. Use Skill #224 (Oracle Ask A Question) to convert category name into correct Amazon bestseller URL
   Input: category name
   Output: Amazon bestseller page URL

2. Use Skill #225 (Crawl 1x Amazon URL) to extract all product data from bestseller page
   Input: Amazon bestseller URL
   Output: Structured text data of all products

3. Use Skill #223 (Powerful LLM Prompt-to-Text Response) to process the structured data and identify highest-priced item
   Input: Structured product data
   Output: Single product data in JSON format

F) SILOS:
SILO 1: URL FINDER
- Input: Category name
- Skill #224: Query oracle to find exact Amazon bestseller URL
- Output: Amazon URL string

SILO 2: DATA EXTRACTION
- Input: Amazon URL from Silo 1
- Skill #225: Crawl Amazon page
- Output: Raw structured product data

SILO 3: DATA PROCESSING
- Input: Raw structured product data from Silo 2
- Skill #223: Process data to identify highest-priced item and format as JSON
- Prompt: "Analyze the provided product data. Identify the item with the highest price. Return a JSON object with the following fields: product_name, price (as float), affiliate_link (insert affiliate ID), and details. Include only the single highest-priced item."
- Output: Final JSON object

Each silo operates independently but sequentially, with the output of each silo feeding directly into the input of the next. The final output (JSON object) contains all required data for subsequent subagents in the main workflow.

SubAgent #1 - Diagram

Expand Diagram

I'll analyze and break down the Review Researcher subagent with specific attention to inputs, outputs, and workflow steps.

A) SUBAGENT SUMMARY: 
A research-focused subagent that gathers, analyzes, and synthesizes product reviews from multiple sources (YouTube and Google) to create a structured summary of product feedback and features.

B) FINAL TASK OUTPUT:
A JSON object containing structured review data with the following components:
```json
{
  "sources": {
    "youtube_reviews": [
      {"url": "...", "key_points": "...", "rating": "..."},
      // 2-3 more YouTube reviews
    ],
    "written_reviews": [
      {"url": "...", "key_points": "...", "rating": "..."},
      // 2-3 more written reviews
    ]
  },
  "consolidated_summary": {
    "pros": ["...", "...", "..."],
    "cons": ["...", "...", "..."],
    "key_features": ["...", "...", "..."],
    "overall_sentiment": "..."
  }
}
```

C) SUBAGENT INPUT:
- Product name (string)
- Product category (string)
- Product URL on Amazon (string)

E) SUBAGENT TASK SUMMARY:
The workflow divides into two parallel research tracks (YouTube and Google) that then merge for final analysis:

SILO 1 - YouTube Research:
1. Research YouTube reviews:
   - Input: Product name + "review"
   - Skill #224 (Oracle Ask A Question) to find top 3 YouTube review videos
   - Output: List of YouTube URLs and basic metadata

2. Analyze each video:
   - Input: YouTube URLs
   - Skill #193 (Download YouTube Video as MP3) x3
   - Skill #198 (Get Transcription of MP3) x3
   - Output: Transcriptions with timings

3. Summarize video content:
   - Input: Transcriptions
   - Skill #223 (Powerful LLM) to analyze transcriptions
   - Output: Structured summary of video reviews

SILO 2 - Written Review Research:
1. Find written reviews:
   - Input: Product name + review query
   - Skill #224 (Oracle Ask A Question) to search for top review sites
   - Output: List of review URLs

2. Extract review content:
   - Input: Review URLs
   - Skill #226 (Extract Structured Data from URL) x3
   - Output: Raw review content

3. Analyze written content:
   - Input: Raw review content
   - Skill #223 (Powerful LLM) to analyze content
   - Output: Structured summary of written reviews

SILO 3 - Final Synthesis:
1. Combine all research:
   - Input: Both YouTube and written review summaries
   - Skill #223 (Powerful LLM) to synthesize all findings
   - Output: Final JSON structure with consolidated review data

F) SILOS:
As detailed above, this subagent operates in three distinct silos:
1. YouTube Research Track
2. Written Review Research Track
3. Final Synthesis Track

Each silo operates independently until the final synthesis, allowing for parallel processing and clear separation of concerns. The final synthesis silo ensures all data is properly structured and normalized before being passed to the next subagent in the larger workflow.

SubAgent #2 - Diagram

Expand Diagram

Let me break down the Image Finder subagent in detail:

A) SUBAGENT SUMMARY: 
A specialized image collection agent that searches for and validates high-quality product images suitable for an Amazon product review article.

B) FINAL TASK OUTPUT:
An array of 3 validated, high-quality product image URLs, each being a JPG/PNG file of the specific Amazon product, verified as accurate and relevant, with dimensions suitable for web article embedding (ideally 800-1200px width).

C) SUBAGENT INPUT:
- Product name (string)
- Product ASIN/Amazon ID (string)
- Product category (string)
- Any specific image requirements (optional string)

E) SUBAGENT TASK SUMMARY:
1. Search Phase:
   Input > Skill #187 (Find Relevant Images with Brave Search) 
   - Input prompt will be specifically crafted to include product name, model number, and "product photo" specification
   - Output: List of 5 initial image URLs

2. Validation Phase:
   Image URLs > Skill #176 (Analyze An Image With GPT Vision & Return Text)
   - Input: Each image URL + prompt "Verify if this is an authentic product photo of [product name]. Check for watermarks, proper lighting, and professional quality"
   - Output: Analysis text for each image

3. Selection Phase:
   Analysis text > Skill #223 (Powerful LLM Prompt-to-Text Response)
   - Input: All image analyses + prompt to select top 3 most suitable images
   - Output: Final selection of 3 best image URLs with reasoning

F) SILOS:
SILO 1: IMAGE COLLECTION
- Purpose: Initial broad image gathering
- Skill: #187 (Find Relevant Images)
- Input: Product details
- Output: 5 image URLs

SILO 2: IMAGE VALIDATION
- Purpose: Verify image authenticity and quality
- Skill: #176 (GPT Vision Analysis)
- Input: 5 image URLs from Silo 1
- Output: 5 analysis texts

SILO 3: FINAL SELECTION
- Purpose: Choose best 3 images
- Skill: #223 (LLM Selection)
- Input: Analysis texts from Silo 2
- Output: Final 3 validated image URLs

This structured approach ensures we get high-quality, relevant images while avoiding incorrect products, poor quality photos, or watermarked images that wouldn't be suitable for the final article.

SubAgent #3 - Diagram

Expand Flow

I'll analyze Subagent 4 ("Review Writer") and break it down according to your requirements:

A) SUBAGENT SUMMARY:
A specialized content generation subagent that transforms product research data and review summaries into a comprehensive, original 300-500 word product review with a structured format and affiliate call-to-action.

B) FINAL TASK OUTPUT:
A single text file containing a 300-500 word product review, structured with introduction, main features, pros/cons, and conclusion with affiliate call-to-action. The review must maintain a friendly, informative tone while incorporating key findings from the research data.

C) SUBAGENT INPUT:
1. Product details JSON object (name, price, affiliate link) from Bestseller Finder
2. Structured review research data (containing pros, cons, features) from Review Researcher

E) SUBAGENT TASK SUMMARY:
The workflow requires multiple LLM operations to ensure quality and structure:

Input > #223 (Process research data into key points) > #190 (Generate initial review draft) > #223 (Refine and structure review) > #185 (Final polish and formatting) > Output

Specifically:
1. #223 (Powerful LLM) - Process and organize research data into structured writing points
2. #190 (Write text based on instructions) - Generate initial 300-500 word review draft
3. #223 (Powerful LLM) - Review and enhance content structure, ensuring all components present
4. #185 (Write/Copy from Input) - Final polish pass focusing on tone and affiliate integration

F) SILOS:
SILO 1: DATA PREPARATION
- Input: Raw research data
- Skill: #223 (Powerful LLM)
- Purpose: Organize research into structured writing points
- Output: Structured writing outline

SILO 2: INITIAL DRAFT
- Input: Structured writing outline
- Skill: #190 (Write text based on instructions)
- Purpose: Generate first draft of review
- Output: Initial review draft

SILO 3: STRUCTURE ENHANCEMENT
- Input: Initial review draft
- Skill: #223 (Powerful LLM)
- Purpose: Ensure proper structure and component inclusion
- Output: Structured review

SILO 4: FINAL POLISH
- Input: Structured review
- Skill: #185 (Write/Copy from Input)
- Purpose: Polish tone, style, and affiliate integration
- Output: Final review text

This multi-silo approach ensures each aspect of the review (organization, content, structure, and polish) receives focused attention while maintaining a logical flow toward the final output.

4 Template & Links

Expand Flow

I'll break down the HTML Assembler subagent according to your guidelines:

A) SUBAGENT SUMMARY:
A specialized HTML generator that combines product information, review content, and images into a properly formatted, complete HTML document with consistent styling and proper metadata.

B) FINAL TASK OUTPUT:
A single .html file containing:
- Complete HTML5 document structure
- Embedded CSS styling
- Product name, price, and affiliate link in structured format
- Full product review text with proper formatting (headers, paragraphs)
- 2-3 product images with proper alt tags and responsive sizing
- Meta tags for SEO
- Schema.org product markup

C) SUBAGENT INPUT:
1. JSON object containing:
   - product_name
   - price
   - affiliate_link
   - product_details
2. Review text (formatted article, 300-500 words)
3. Array of image URLs (2-3 product images)

E) SUBAGENT TASK SUMMARY:
The flow breaks down into three main steps:

1. Initial HTML Structure Creation:
   Input JSON + review text > Skill #223 (Powerful LLM Prompt-to-Text Response)
   - Prompt: "Generate complete HTML5 structure with embedded CSS styling for a product review page. Include schema.org product markup and meta tags. Use the following product data: [JSON data]"
   Output: Base HTML structure

2. Content Integration:
   Base HTML + review text > Skill #190 (Write or rewrite text based on instructions)
   - Prompt: "Format the provided review text into proper HTML structure with h1, h2, p tags etc. Maintain proper semantic structure and include the product details from the JSON in appropriate locations"
   Output: HTML with formatted content

3. Final Assembly:
   HTML with content + image URLs > Skill #223 (Powerful LLM Prompt-to-Text Response)
   - Prompt: "Complete the HTML by adding responsive image tags with proper alt text and sizing for the following image URLs. Ensure all HTML validates and closes properly."
   Output: Complete, validated HTML file

F) SILOS:
SILO 1: STRUCTURE
- Purpose: Create base HTML5 document structure
- Input: Product JSON data
- Skill: #223
- Output: Base HTML template

SILO 2: CONTENT
- Purpose: Format and integrate review content
- Input: Review text + base HTML
- Skill: #190
- Output: HTML with formatted content

SILO 3: MEDIA
- Purpose: Add and format images
- Input: HTML with content + image URLs
- Skill: #223
- Output: Final complete HTML file

This structure ensures each component (structure, content, media) is handled separately but builds logically toward the final output, with each silo's output feeding into the next silo's input.