What Shall We Build Next?

Describe

Describe your task

Refine

Refine the plan

SubAgents

Review all agents

Deploy

Deploy your agent

A) SUBAGENT SUMMARY  
This “DataExtractor” subagent receives a single product listing URL (Amazon/Etsy) and returns a clean JSON-like structure containing the title, bullet points, description, and main image URL.

B) FINAL TASK OUTPUT  
A JSON-like structure in this exact shape:  
{  
  "title": "...",  
  "bullets": ["..."],  
  "description": "...",  
  "image_url": "..."  
}

C) SUBAGENT INPUT  
{productListingUrl}

E) SUBAGENT TASK SUMMARY  
1) Use Skill #226 - Extract Structured Data From 1x URL  
   • Input: {productListingUrl}, plus an instruction such as:  
     "Please extract the product title, bullet points, product description, and the main image URL from this listing."  
   • Output: A text-based summary (structured or semi-structured) containing the extracted information.

2) Use Skill #223 - Powerful LLM Prompt-to-Text Response  
   • Input: The output from Step 1, plus a directive to format the extracted information into the final JSON-like structure:  
     "Here is the extracted data about the product. Please return it in JSON form with the keys 'title', 'bullets', 'description', and 'image_url' exactly."  
   • Output: The final JSON-like structure.

F) SILOS  
• Silo 1: Data Extraction  
  - Step 1: (#226) Crawl and gather raw structured data from {productListingUrl}.

• Silo 2: JSON Formatting  
  - Step 2: (#223) Convert the raw structured data into the final JSON-like output.

SubAgent #1 - Diagram

Expand Diagram

A) SUBAGENT SUMMARY  
This subagent (BrandGuard) identifies potential trademark issues in the extracted text (title, bullets, description) and checks the original image for any visible brand or logo references. It returns a concise report of any flagged risks.

B) FINAL TASK OUTPUT  
[trademark-analysis] → A short text report listing any flagged trademarked words/phrases and an indication of whether the image may contain protected/trademarked branding or logos.

C) SUBAGENT INPUT  
[extracted-data], which is a JSON-like structure of the form:  
{  
  "title": "...",  
  "bullets": ["..."],  
  "description": "...",  
  "image_url": "..."  
}

D) SUBAGENT OUTPUT  
[trademark-analysis], which is a single text output containing any flagged trademark terms from the text plus an assessment of the image.

E) SUBAGENT TASK SUMMARY (Detailed Steps)

1) EXTRACT KEYWORDS FROM TEXT  
   • Purpose: Parse the combined text (title + bullets + description) to identify candidate words/phrases that warrant trademark checks.  
   • Skill Used: #223 (Powerful LLM Prompt-to-Text Response)  
   • Input to Skill:  
     – Prompt:  
       “Below is the text from an Amazon or Etsy listing. Identify relevant keywords and phrases that might be trademarked or brand references:  
        Title: {title}  
        Bullets: {bullets}  
        Description: {description}  
        Return them as a simple list.”  
     – Data from [extracted-data]: {title}, {bullets}, {description}  
   • Output from Skill: A concise list of keywords/phrases (e.g., [“dog treat”, “brandX”, …]).

2) CHECK FOR TRADEMARKED TERMS WITH THE ORACLE  
   • Purpose: For each keyword from step 1, determine if it is trademarked by searching “[keyword] trademark.”  
   • Skill Used: #224 (Oracle Ask A Question)  
   • Input to Skill: A single query or multiple queries (depending on approach). For example:  
     – “Check if any of the following keywords are listed as registered trademarks. For each, do a search with ‘[keyword] trademark.’ Then summarize which are possibly trademarked.”  
   • Output from Skill: A text-based analysis specifying which keywords have potential conflicts and any relevant notes.

3) IMAGE ANALYSIS FOR LOGOS OR BRAND REFERENCES  
   • Purpose: Look for visible brand references or trademarked logos in the main product image.  
   • Skill Used: #176 (Analyze An Image With GPT Vision & Return Text)  
   • Input to Skill:  
     – image_url from [extracted-data], plus instruction:  
       “Analyze this image for any logos, brand names, or trademarked designs. Describe any text or brand markers you see.”  
   • Output from Skill: A text summary reporting any brand/logos discovered or whether it appears free of such references.

4) COMPILE A FINAL TRADEMARK REPORT  
   • Purpose: Merge the text trademark check findings (step 2) with the image analysis (step 3) into a concise final trademark risk assessment.  
   • Skill Used: #223 (Powerful LLM Prompt-to-Text Response)  
   • Input to Skill:  
     – The summary from step 2 (text trademark results)  
     – The summary from step 3 (image analysis)  
     – Instruction: “Combine these findings into a short text report. List each term flagged and note if it is or might be trademarked. Then provide a separate note about whether the image includes any trademark references.”  
   • Output from Skill → [trademark-analysis]: A short text write-up indicating flagged terms, any relevant disclaimers, and granting a “safe” or “caution” recommendation regarding the image.

F) SILOS (If Needed)  
• Text Trademark Check Silo: Steps 1–2 (Identifying text keywords, using Oracle to confirm trademark status).  
• Image Trademark Check Silo: Step 3 (Analyzing the image for brand references).  
• Compilation Silo: Step 4 (Combining text-based and image-based findings into final [trademark-analysis]).

This completes the final workflow for Subagent 2: TRADEMARKANALYSISSUBAGENT (BrandGuard).

SubAgent #2 - Diagram

Expand Diagram

A) SUBAGENT SUMMARY:  
This “TShirtDesignSubagent (DesignGenerator)” takes the trademark-safe text, style preferences, and user instructions, then generates multiple on-theme T-shirt designs of the required dimensions, returning an array of newly created image URLs.

B) FINAL TASK OUTPUT:  
An array (list) of 12 unique PNG image URLs, each sized 4500×5400 pixels at 300 dpi.

C) SUBAGENT INPUT:  
• {extracted-data} – containing the original title, bullets, description, image URL, etc.  
• [trademark-analysis] – summary specifying which terms are safe or flagged.

E) SUBAGENT TASK SUMMARY:  
1) Receive the inputs ({extracted-data}, [trademark-analysis]).  
2) Generate a safe design prompt:  
   • Use skill #223 (Powerful LLM Prompt-to-Text Response) to combine user style requests, safe keywords from [trademark-analysis], color/theme guidelines, and any required text from {extracted-data}. Instruct the LLM to incorporate 300dpi, 4500×5400 size, same text/niche, and avoidance of flagged terms.  
   • Output: A single short text prompt that will be used to generate T-shirt designs.

3) Create 12 T-shirt design variations at once (or in a loop):  
   • For each of the 12 variations:  
     a) Use skill #222 (Make Image [Especially With Text]) with the design prompt.  
     b) Receive a PNG URL.  
     c) Use skill #191 (Resize Image) on each PNG URL, specifying width=4500, height=5400 to ensure correct dimensions.  
   • Collect all 12 resized image URLs into an array.

4) Return the array of 12 PNG URLs:  
   • [new-shirt-design] = [url1.png, url2.png, …, url12.png]

F) SILOS:

• Silo 1: Prompt Refinement (Skill #223)  
  – Inputs: {extracted-data}, [trademark-analysis], user instructions  
  – Output: Single textual prompt that includes style directives (dimensions, color/theme if any), original slogan/text (minus trademark issues), niche, etc.

• Silo 2: Image Generation & Resizing (Skills #222, #191) (Repeated 12×)  
  – Step a) Create T-shirt design from the approved prompt (#222).  
  – Step b) Resize each resulting image to exactly 4500×5400 (#191).  
  – Collect final URLs in an array of 12.

• Final Output: [new-shirt-design], an array of 12 PNG URLs.

SubAgent #3 - Diagram

Expand Flow

A) SUBAGENT SUMMARY:  
This subagent (ListingCreator) takes the original listing’s extracted data and a report on trademark risks, then creates a revised Amazon-compliant product listing (title, bullet points, description) and provides a final compliance note (including a confidence score).

B) FINAL TASK OUTPUT:  
The ListingCreator outputs a structured JSON-like object containing:  
{  
  "title": "new title (max 80 chars)",  
  "bullets": ["new bullet 1 (max 200 chars)", "new bullet 2 (max 200 chars)", ...],  
  "description": "new description (max 1000 chars)",  
  "compliance_note": "...",  
  "confidence_score": "..."  
}

C) SUBAGENT INPUT:  
• [extracted-data]: Object with original title, bullets, description, image_url  
• [trademark-analysis]: Object/text detailing flagged (trademark) terms and safe terms

E) SUBAGENT TASK SUMMARY (Step-by-Step Workflow):  
1) Receive [extracted-data] and [trademark-analysis].  
2) SKILL #190 - “Write or rewrite text based on instructions”:  
   • Input to skill #190: A prompt that instructs the LLM to rewrite the original listing text:  
     – Remove or replace any flagged trademark terms from [trademark-analysis].  
     – Maintain original theme/niche but ensure new text is unique and within length limits:  
       – Title: up to 80 characters.  
       – 5 bullet points, each up to 200 characters.  
       – Description: up to 1000 characters.  
     – Incorporate relevant safe keywords (if [trademark-analysis] suggests them).  
   • Output from skill #190: The newly revised listing text (title, bullets, description).

3) SKILL #223 - “Powerful LLM Prompt-to-Text Response” (Compliance Check):  
   • Input to skill #223:  
     – The newly created listing text.  
     – Amazon’s print-on-demand rules (supplied as reference text/URL or summarization).  
     – Ask the LLM to:  
       1) Confirm the text does not violate Amazon’s guidelines.  
       2) Provide a “compliance_note” and a “confidence_score” for trademark compliance.  
   • Output from skill #223:  
     – The final compliance assessment (compliance_note)  
     – The confidence_score (e.g., “8/10” or “High Confidence”)

4) Combine the result into a final JSON-like structure:  
   {  
     "title": "...",  
     "bullets": [...],  
     "description": "...",  
     "compliance_note": "...",  
     "confidence_score": "..."  
   }

F) SILOS:  
• SILO 1: Draft Listing Text  
  – Input: [extracted-data], [trademark-analysis]  
  – Action: #190 to create a revised title, bullets, description  
  – Output: “draft” listing text

• SILO 2: Compliance Validation  
  – Input: “draft” listing text (from SILO 1) + summary of Amazon guidelines  
  – Action: #223 to generate “compliance_note” + “confidence_score”  
  – Output: final updated listing text + compliance details

• Final Output:  
{  
  "title": "<80 chars>",  
  "bullets": [5 bullets, each <200 chars],  
  "description": "<1000 chars>",  
  "compliance_note": "...",  
  "confidence_score": "..."  
}