Ad Campaign Pack Generator

I want to build an AI agent that creates complete mini ad campaigns for small businesses. The user provides: business name, niche/industry, target audience, and product/service description. Step 1: The agent researches top competitor ads in the niche and extracts winning angles. Step 2: The agent generates ad copy – 5 headlines, 5 sub-headlines, 3 short forms, 3 long forms. Step 3: The agent generates 3–5 ad creatives (images) in the style of trending niche ads. Step 4 (optional): The agent creates a 15–30 second video ad with script, AI voiceover, and visuals. Step 5: The agent outputs all assets in a structured zip file with suggested captions, hashtags, and a quick-start deployment guide.

First, pre-entered info...

final-output2

A single compressed archive (a .zip file) that, when extracted, gives the user every asset needed to launch the mini-ad campaign.

subagents-extracted2

OVERVIEW OF SUBAGENTS 1. Insight & Copy Builder – mines competitor ads, writes summary + full ad copy set. 2. Visual Creative Designer – turns the best angles/headlines into 3–5 static image creatives. 3. Video Ad Producer (conditional) – writes VO script, produces 15–30 s vertical video. 4. Deployment & Packaging Assistant – crafts captions/hashtags + quick-start guide, zips everything. SUBAGENT 1 – Insight & Copy Builder • Name: "Insight & Copy Builder" • Final Output (before summariser): – research/competitor_summary.md – copy/ad_copy.json • Skills & Sequence 1) #224 Oracle – gather top-performing ads ➜ raw data. 2) #185 Write Text – produce competitor_summary.md and ad_copy.json. 3) #223 LLM – create insight_copy_manifest.txt. • SUBAGENT REQUIRED INPUT: {business-name}, {industry}, {target-audience}, {product-description} • SUBAGENT FINAL OUTPUT: [insight-copy-bundle] (bundle = competitor_summary.md, ad_copy.json & insight_copy_manifest.txt) SUBAGENT 2 – Visual Creative Designer • Name: "Visual Creative Designer" • Final Output (before summariser): images/creative_01.png … creative_03.png (±2) • Skills & Sequence 1) Loop 3–5 × → #222 Make Image (With Text) from angles/headlines. 2) #223 LLM – create images_manifest.txt. • SUBAGENT REQUIRED INPUT: {business-name}, [insight-copy-bundle] • SUBAGENT FINAL OUTPUT: [image-creatives] (folder of PNGs + images_manifest.txt) SUBAGENT 3 – Video Ad Producer (optional) • Name: "Video Ad Producer" • Final Output (before summariser): video_script.txt, video_ad.mp4, voiceover.mp3 (opt), storyboard.json • Skills & Sequence 1) #171 Write Voice-Over Script → video_script.txt 2) #201 Generate Vertical Text-Slide Video → video_ad.mp4 (+ extras) 3) #223 LLM – create video_manifest.txt • SUBAGENT REQUIRED INPUT: {business-name}, {include-video}, [insight-copy-bundle] • SUBAGENT FINAL OUTPUT: [video-ad-assets] (only produced if {include-video}=true) SUBAGENT 4 – Deployment & Packaging Assistant • Name: "Deployment & Packaging Assistant" • Final Output (before summariser): captions_and_hashtags.json, quick_start_guide.md, and the zipped archive • Skills & Sequence 1) #186 Create MicroCopy → captions_and_hashtags.json 2) #185 Write Text → quick_start_guide.md 3) Internal Zip & Save → mini_ad_campaign_{business-name}.zip 4) #223 LLM – delivery_manifest.txt • SUBAGENT REQUIRED INPUT: {business-name}, [insight-copy-bundle], [image-creatives], [video-ad-assets] • SUBAGENT FINAL OUTPUT: [packaged-mini-ad-campaign] (= the .zip file + delivery_manifest.txt)

subagentnumber

[subagentnumber]

Then, agent response...

subagent1refined-new-json-core

I'll add section G) TASKFLOW SUMMARY to SUBAGENT 1's specification, extracting all tasks/skills from sections E) and F) and formatting them as requested: G) TASKFLOW SUMMARY Task 1: - ID: #218 - Title task: Brainstorm Related Keywords for Ad Research - Description of the task: Analyzes business inputs to generate 8-12 highly relevant search terms for competitive ad research, focusing on industry-specific and audience-targeted keywords - Input(s) involved: {business-name}, {industry}, {target-audience}, {product-description} - Output token name: [keywords_list] Task 2: - ID: #224 - Title task: Oracle Competitive Ad Research - Description of the task: Performs live searches using generated keywords to collect top-performing ads across multiple platforms (Facebook, Instagram, Google), analyzing real-world ad examples in the specified niche - Input(s) involved: [keywords_list], {industry}, {target-audience}, {product-description} - Output token name: [raw_ads_research] Task 3: - ID: #223 - Title task: LLM Analysis of Competitive Research - Description of the task: Processes raw ad research data to extract key marketing angles, hooks, offer types, CTA styles, and creative motifs, compiling findings into a structured markdown report - Input(s) involved: [raw_ads_research] - Output token name: [competitor_summary] Task 4: - ID: #185 - Title task: Generate Structured Ad Copy Set - Description of the task: Creates complete ad copy package including 5 headlines, 5 sub-headlines, 3 short-form copies, and 3 long-form copies, formatted as JSON, based on competitive insights - Input(s) involved: [competitor_summary], {business-name}, {industry}, {target-audience}, {product-description} - Output token name: [ad_copy_json] Task 5: - ID: #223 - Title task: Create Bundle Manifest - Description of the task: Generates a manifest file listing all produced assets, their purposes, and file sizes, creating a comprehensive index of the insight and copy bundle - Input(s) involved: [competitor_summary], [ad_copy_json] - Output token name: [insight_copy_manifest] The rest of the subagent specification (sections A through F) remains unchanged as per the original summary provided.

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[User Input] --> B[business-name] A --> C[industry] A --> D[target-audience] A --> E[product-description] B & C & D & E --> F[Task 1: Brainstorm Keywords] F --> G[keywords_list.txt] G --> H[Task 2: Oracle Research] H --> I[raw_ads_research.txt] I --> J[Task 3: LLM Analysis] J --> K[competitor_summary.md] K & B & C & D & E --> L[Task 4: Generate Ad Copy] L --> M[ad_copy.json] K & M --> N[Task 5: Create Manifest] N --> O[insight_copy_manifest.txt] K & M & O --> P[insight-copy-bundle] subgraph Silo 1: Research F G H I J K end subgraph Silo 2: Copy & Package L M N O P end </div> <script> mermaid.initialize({ startOnLoad: true }); </script> </body> </html>

Mermaid Image - https://static.aiz.ac/1757837145-mermaid/mermaid1.png

prompts-required

[prompts-required]

subagent1-json-array

[ { "taskID": 218, "semanticTitleOfTask": "Generate Niche-Specific Research Keywords", "taskDescription": "Transforms core business details into 8–12 highly targeted search terms that will surface the most relevant, top-performing competitor ads across major ad platforms. These keywords serve as the seed for all downstream competitive research.", "inputDescription": "Receives four plain-text strings supplied by the user: business name, industry or niche, primary target audience description, and a short product or service description. The task parses these details to understand category, benefits, pain points, and audience traits before producing research terms.", "inputRequired": ["business-name", "industry", "target-audience", "product-description"], "outputDescription": "keywords_list – a simple text list (one keyword phrase per line) containing 8-12 carefully crafted search terms combining industry, audience, product benefits, and platform-specific ad terminology.", "outputName": "keywords_list", "promptInstruction": "You are a digital marketing strategist. Given a BUSINESS_NAME, INDUSTRY, TARGET_AUDIENCE, and PRODUCT_DESCRIPTION, brainstorm 8-12 distinct keyword phrases that will uncover high-performing competitor ads. Include mixes such as:\n• Industry + audience descriptor (e.g., \"dentist ads moms\")\n• Product type + benefit (e.g., \"teeth whitening instant results\")\n• Service category + pain point (e.g., \"family dental no insurance\")\n• Platform modifiers (\"facebook ad\", \"instagram ads\")\nReturn the phrases line-by-line, no numbering or extra text." }, { "taskID": 224, "semanticTitleOfTask": "Fetch Top Competitor Ads with Oracle Search", "taskDescription": "Uses the Oracle search/API tool to run multiple live queries with the generated keyword list and business context, retrieving raw text, image links, and meta-data for the top ads currently performing in the specified niche.", "inputDescription": "Requires the full keyword list produced by Task #218 plus the original industry, audience, and product details to refine query parameters. The task issues ~5 platform-specific searches (Facebook, Instagram, Google, etc.) and aggregates the returned ad snippets into one raw text file.", "inputRequired": ["keywords_list", "industry", "target-audience", "product-description"], "outputDescription": "raw_ads_research – a UTF-8 text file containing every ad example returned, including headlines, body copy, CTAs, impressions/engagement metrics when available, and any image or video URLs.", "outputName": "raw_ads_research", "promptInstruction": "no instruction" }, { "taskID": 223, "semanticTitleOfTask": "Analyze Raw Ads to Produce Competitor Summary", "taskDescription": "Processes the unstructured ad research file to extract marketing angles, hooks, offers, CTAs, and creative patterns, delivering a 1,000–2,500-word markdown report that distills actionable insights for copy creation.", "inputDescription": "Takes the single text blob produced by Task #224. No additional data is needed. The LLM scans the ads, clusters recurring themes, and selects representative examples for each insight category before composing the markdown report.", "inputRequired": ["raw_ads_research"], "outputDescription": "competitor_summary – a well-formatted markdown document saved to /research that details dominant angles, hooks, offers, CTAs, visual motifs, copy patterns, and concludes with key takeaways.", "outputName": "competitor_summary", "promptInstruction": "You are an advertising analyst creating a competitor research summary. Review the provided RAW_ADS_RESEARCH and extract:\n1. Dominant Marketing Angles (3-5)\n2. Most Effective Hooks (3-5)\n3. Popular Offer Structures (2-3)\n4. Common CTA Patterns (2-3)\n5. Recurring Visual/Creative Elements (3-4)\n6. Key Copywriting Patterns (2-3)\nFor each element give concrete ad examples. Format as markdown with clear H2 headings, bullet lists, and a final **Key Takeaways** section summarizing the most actionable findings." }, { "taskID": 185, "semanticTitleOfTask": "Generate Structured Ad Copy Variations", "taskDescription": "Leverages the competitor insights and full business details to craft a complete, ready-to-use ad copy set: 5 headlines, 5 sub-headlines, 3 short-form bodies, and 3 long-form bodies, output in strict JSON format.", "inputDescription": "Requires the competitor_summary markdown plus the four original user inputs (business name, industry, target audience, product description). The task references angles, hooks, and CTA styles from the summary while tailoring messaging to the client.", "inputRequired": ["competitor_summary", "business-name", "industry", "target-audience", "product-description"], "outputDescription": "ad_copy_json – a valid JSON document saved to /copy with keys: \"headlines\", \"sub_headlines\", \"short_form_copy\", \"long_form_copy\", each containing the specified number of copy variants ready for deployment.", "outputName": "ad_copy_json", "promptInstruction": "You are an ad copywriter. Using COMPETITOR_SUMMARY and the client info (BUSINESS_NAME, INDUSTRY, TARGET_AUDIENCE, PRODUCT_DESCRIPTION), create:\n• 5 headlines (≤65 chars)\n• 5 sub-headlines (≤90 chars)\n• 3 short-form copy blocks (≤100 words)\n• 3 long-form copy blocks (≤300 words)\nEnsure messaging is benefit-driven, aligns with successful angles, and speaks directly to the target audience. Return ONLY a JSON object with keys \"headlines\", \"sub_headlines\", \"short_form_copy\", \"long_form_copy\"." }, { "taskID": 223, "semanticTitleOfTask": "Compile Insight & Copy Bundle Manifest", "taskDescription": "Creates a plain-text manifest that enumerates every file in the insight-copy bundle, listing filename, byte size, one-line content description, and last-modified timestamp to aid downstream packaging and auditing.", "inputDescription": "Consumes the competitor_summary markdown and ad_copy_json files. The LLM inspects file metadata (size and timestamps) and drafts a concise manifest describing each asset’s purpose within the bundle.", "inputRequired": ["competitor_summary", "ad_copy_json"], "outputDescription": "insight_copy_manifest – a .txt file saved at the root of the bundle that documents competitor_summary.md, ad_copy.json, and itself, providing clear reference information for later agents or human users.", "outputName": "insight_copy_manifest", "promptInstruction": "You are a technical documentation specialist preparing a manifest for the Insight & Copy bundle. List each file: competitor_summary.md, ad_copy.json, insight_copy_manifest.txt. For each, include: Filename | File size in bytes | One-sentence description | Last modified ISO timestamp. Format as clean plain text with one file per line." } ]

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 Start[Start] --> Input[Input: business name, industry, audience, product] Input --> Task1[Task 218: Brainstorm Keywords] Task1 --> keywords_list[keywords_list.txt] keywords_list --> Task2[Task 224: Oracle Ad Research] Input --> Task2 Task2 --> raw_ads_research[raw_ads_research.txt] raw_ads_research --> Task3[Task 223: LLM Analysis] Task3 --> competitor_summary[competitor_summary.md] competitor_summary --> Task4[Task 185: Ad Copy Generation] Input --> Task4 Task4 --> ad_copy_json[ad_copy.json] competitor_summary --> Task5[Task 223: Bundle Manifest] ad_copy_json --> Task5 Task5 --> insight_copy_manifest[insight_copy_manifest.txt] competitor_summary --> Bundle[Insight Copy Bundle] ad_copy_json --> Bundle insight_copy_manifest --> Bundle Bundle --> End[End] </div> <script> mermaid.initialize({ startOnLoad: true }); </script> </body> </html>