What Shall We Build Next?

Describe

Describe your task

Refine

Refine the plan

SubAgents

Review all agents

Deploy

Deploy your agent

A) SUBAGENT SUMMARY:  
“ProfileBuilder” converts the raw user-supplied business details (name, services, service areas, optional URL) into a well-formed UTF-8 JSON file called business_profile.json.

B) FINAL TASK OUTPUT:  
business_profile.json – a single UTF-8 JSON file with the following schema  
{  
  "business_name": string,  
  "services": [string, …],        // trimmed, lowercase, deduplicated  
  "service_area": [string, …],    // trimmed, lowercase, deduplicated  
  "website": string | null        // null if not supplied  
}

C) SUBAGENT INPUT (from end-user):  
{variable1} = business_name (string)  
{variable2} = services (comma-separated string)  
{variable3} = service_area (comma-separated string)  
{variable4} = website_url (optional string, may be blank)

E) SUBAGENT TASK SUMMARY (skill chain):

1. {variable1-4} → #223 Powerful LLM  
   Prompt:  
   “Take the following raw inputs and return ONLY valid JSON that matches the exact schema  
   {business_name, services[], service_area[], website}.  
   • Split any comma-separated lists (services, service_area) into arrays.  
   • Trim whitespace, make each item lowercase.  
   • Deduplicate each array.  
   • If website_url is blank, set value to null.  
   INPUTS:  
   business_name: {variable1}  
   services: {variable2}  
   service_area: {variable3}  
   website_url: {variable4}”

Output: structured JSON text.

2. JSON text → #185 Write Text  
   Instruction to skill:  
   “Save the incoming JSON exactly as received into a UTF-8 file named business_profile.json.”

Output: business_profile.json (file saved on server).

F) SILOS:  
No additional silos required – the subagent is a single linear flow.

SubAgent #1 - Diagram

Expand Diagram

A) SUBAGENT SUMMARY:  
SeedGenerator reads the previously-created business_profile.json and turns the services × service-area combinations into a clean, de-duplicated list (≤ 25) of local-SEO “seed” search phrases, then saves them as a one-column CSV file.

B) FINAL TASK OUTPUT:  
seed_keywords.csv – UTF-8, one column named keyword, up to 25 rows (not counting header), each row a single local search phrase.

C) SUBAGENT INPUT:  
[business-profile-json]  → the JSON file output by ProfileBuilder that contains:  
• business_name (string)  
• services (array of strings)  
• service_area (array of strings)  
• website (string, optional)

E) SUBAGENT TASK SUMMARY (skill chain):

1. #223 – Powerful LLM Prompt-to-Text Response  
   INPUT:  
   └─ Prompt: “Load the following JSON of a local business. Return two separate, alphabetically-sorted plain-text lists without bullets:  
        List-A = all ‘services’ values;  
        List-B = all ‘service_area’ values (cities/areas).  
      Ensure each item is on its own line.  
      JSON: ```<contents of business_profile.json>```”  
   OUTPUT:  
   └─ Plain text block containing List-A lines followed by List-B lines.

2. #223 – Powerful LLM Prompt-to-Text Response  
   INPUT:  
   └─ Prompt (feeds in the two lists from step 1):  
      “Using the Services list and Cities list below, generate local-intent search phrases following these patterns until you have no more than 25 UNIQUE items:  
        1) “[service] in [city]”  
        2) “best [service] [city]”  
        3) “[service] near me [city]”  
        4) “[city] [service]”  
      • Keep all text lowercase.  
      • Remove any duplicate phrases.  
      • Output ONLY the phrases, one per line, no numbering.”  
   OUTPUT:  
   └─ Plain text list (≤ 25 lines) of finished seed phrases.

3. #185 – Write Text (Or Copy) From Inputted Text  
   INPUT:  
   └─ Text: “keyword\n<line-1>\n<line-2>\n…<line-n>” (adds CSV header ‘keyword’ above the list from step 2).  
   Additional instruction inside prompt: “Save exactly this text as a UTF-8 CSV file named seed_keywords.csv.”  
   OUTPUT:  
   └─ seed_keywords.csv (stored on server, UTF-8, one column ‘keyword’).

F) SILOS:  
This subagent operates in a single silo – Seed Phrase Generation – comprising the three sequential skills above; no parallel branches are necessary.

SubAgent #2 - Diagram

Expand Diagram

A) SUBAGENT SUMMARY:  
KeywordAnalyzer takes the one-column seed_keywords.csv, expands each seed into a larger but capped list of candidate queries, then enriches every query with an estimated local search volume and a 0-100 difficulty score, outputting everything as keyword_metrics.csv.

B) FINAL TASK OUTPUT:  
keyword_metrics.csv (UTF-8, comma-separated, three columns in this exact order: keyword, avg_monthly_searches, difficulty_score). Maximum 100 data-rows, header included.

C) SUBAGENT INPUT:  
[seed_keywords.csv] ‑ a UTF-8 CSV file produced by SeedGenerator that contains one column headed “keyword”, ≤ 25 rows.

E) SUBAGENT TASK SUMMARY (skill chain)  
[seed_keywords.csv]  
 → Skill #223 (Powerful LLM Prompt-to-Text Response)  
  Prompt: “Below is the content of seed_keywords.csv.  
  1. For EACH seed keyword, create up to THREE closely related local search variants (examples: synonym + city, plural, ‘near-me’, ‘best … in …’).  
  2. Combine the original seeds and the new variants into ONE de-duplicated list, keep total ≤ 100 items.  
  3. Return the final list as one keyword per line.”  
  OUTPUT: plain-text list of ≤ 100 unique keywords (one per line).

(list from step 1)  
 → Skill #224 (Oracle Ask A Question)  
  Prompt / question:  
  “Using reliable keyword-research sources, provide the following for EACH keyword in the list:  
  • average monthly searches local to the geographic element in the query (or US city if none)  
  • an organic ranking difficulty score from 0–100 (higher = harder).  
  Return the answer as a three-column table in this exact order: keyword | avg_monthly_searches | difficulty_score, no extra commentary.”  
  OUTPUT: markdown-style or TSV/CSV-style table containing the three requested columns for every keyword.

(table from step 2)  
 → Skill #223 (Powerful LLM Prompt-to-Text Response)  
  Prompt: “Convert the previous table to clean CSV with header row keyword,avg_monthly_searches,difficulty_score.  
  If any value is missing, insert an estimated value based on similar rows.  
  Return ONLY the CSV text.”  
  OUTPUT: correctly formatted CSV text.

(CSV text)  
 → Skill #185 (Write Text)  
  Instruction: “Save the received CSV text to a UTF-8 file named keyword_metrics.csv.”  
  OUTPUT: [keyword-metrics-csv] (the file path/URL to keyword_metrics.csv).

F) SILOS

Silo 1 – Expansion  
 Input seed_keywords.csv → #223 → expanded keyword list

Silo 2 – Metric Enrichment  
 expanded list → #224 → raw table of metrics

Silo 3 – Normalise & Save  
 raw table → #223 (format/patch) → #185 (write file) → keyword_metrics.csv

SubAgent #3 - Diagram

Expand Flow

A) SUBAGENT SUMMARY  
PlanMaker converts raw keyword metrics plus the business profile into a ready-to-upload local-keyword page plan (one row = one page), choosing the best keyword per cluster and adding all required SEO metadata.

B) FINAL TASK OUTPUT  
local_keyword_plan.csv – UTF-8 CSV, comma-delimited, headed columns in this exact order:  
page_slug, primary_keyword, supporting_keywords, avg_monthly_searches, difficulty_score, location, intent

C) SUBAGENT INPUT  
[keyword-metrics.csv] – three columns: keyword, avg_monthly_searches, difficulty_score  
[business-profile.json] – keys: business_name, services[], service_area[], website (optional)

E) SUBAGENT TASK SUMMARY

1. Read both input files (they will be passed to the skill as raw text).  
2. Skill #223 – Powerful LLM Prompt-to-Text Response  
    • Prompt: “Cluster the supplied keyword list into ≤ 30 topical groups, maximising intra-cluster similarity and matching service_area tokens.  
     For each cluster choose the single best primary_keyword (highest volume / lowest difficulty trade-off).  
     Return JSON array where each item = {cluster_id, primary_keyword, supporting_keywords[], avg_monthly_searches, difficulty_score, location, intent (‘informational’ | ‘transactional’ | ‘navigational’ | ‘near-me’)}.”  
    • Output: clusters.json-like text.

3. Skill #223 – Powerful LLM Prompt-to-Text Response (second call)  
    • Input: clusters JSON from step 2 plus business_profile.json.  
    • Prompt: “Using the clusters JSON, generate rows for a CSV with columns page_slug, primary_keyword, supporting_keywords (semicolon-separated), avg_monthly_searches, difficulty_score, location, intent.  
     Create page_slug by slugifying ‘primary_keyword’ (spaces→-, remove punctuation) or, if business_profile.website contains a matching URL segment, reuse that segment.  
     Make sure supporting_keywords are deduplicated; keep max 8 per row.”  
    • Output: complete CSV body (including header line).

4. Skill #185 – Write Text (Or Copy) From Inputted Text  
    • Input: CSV body from step 3.  
    • Instruction: “Save exactly as UTF-8 CSV named local_keyword_plan.csv.”  
    • Output: [local_keyword_plan.csv]

F) SILOS

SILO 1 – CLUSTER & SELECT  
input files → #223 (clustering & selection) → clusters JSON

SILO 2 – ASSEMBLE CSV  
clusters JSON → #223 (format to CSV with slugs & intents) → CSV text → #185 (write file) → local_keyword_plan.csv