What Shall We Build Next?

Describe

Describe your task

Refine

Refine the plan

SubAgents

Review all agents

Deploy

Deploy your agent

A) SUBAGENT SUMMARY  
“Market_Scout_Agent” harvests three live data-sources (Skyscanner inbound-demand, TripAdvisor forum chatter, Google Travel Trends) for five target countries, normalises the signals into one weighted metric, and exports a single JSON file (`luxury_segments.json`) every night at 02:00 AWST.

B) FINAL TASK OUTPUT  
/data/luxury_segments.json – UTF-8, pretty-printed JSON array.  
Schema:  
[  
  {  
    "country": "Japan",  
    "topic": "Swan Valley Wine Tour",  
    "interest_score": 0.83        // 0–1 float, 3-source weighted  
  }, … (≤150 objects)  
]

C) SUBAGENT INPUT  
{tripadvisor_headers} – custom request headers / cookies for TripAdvisor  
{skyscanner_api_key} – bearer or key param usable in Skyscanner Demand API  
{google_trends_api_key} – key or cookie jar that allows Google Travel Trends calls

D) SUBAGENT TASK SUMMARY  
{tripadvisor_headers}, {skyscanner_api_key}, {google_trends_api_key}  
  → #223 Powerful LLM – assemble static config (country list & WA topic keywords)  
  → #224 Oracle Ask – call Skyscanner “Browse-Stats” endpoint per country (arrivals = PER) → raw_demand_{country}.json  
  → #226 Extract Structured Data From URL – scrape top 3 TripAdvisor forum pages per topic (using {tripadvisor_headers}) → trip_forum_{country}_{topic}.txt  
  → #224 Oracle Ask – query Google Travel Trends 30-day index for “<topic> Western Australia” filtered to each country → trends_{country}_{topic}.json  
  → #223 Powerful LLM – normalise & weight (40 % Skyscanner, 35 % TripAdvisor post-count, 25 % Google Trends index), produce interest_score, build master array  
  → #223 Powerful LLM – validate JSON schema & pretty-print  
  → (final output) /data/luxury_segments.json

E) SKILL-BY-SKILL FLOW

1. Build config list  
  Skill #223  
  Prompt: “Return an array of the five target ISO-2 country codes (JP, CN, SG, DE, AU) and the eight WA tourism topics…”  
  Output: config.txt (used internally)

2. Skyscanner demand fetch (looped, one per country)  
  Skill #224 (can make authenticated HTTPS calls)  
  Input example:  
  “GET https://partners.api.skyscanner.net/apiservices/v3/stats/flights/browse?originCountry=<ISO-2>&destinationAirport=PER&apikey={skyscanner_api_key} … Please return JSON only.”  
  Output: raw_demand_JP.json (five such files)

3. TripAdvisor forum scrape (loop country→topic)  
  Skill #226  
  Input:  
  URL: `https://www.tripadvisor.com.au/ShowForum-g255103-i531-Perth_Greater_Perth_Western_Australia.html` (or topic-specific URL)  
  Instruction: “Using {tripadvisor_headers}, extract number_of_posts_last_30d and language of post.”  
  Output: trip_forum_JP_Perth.txt (∼40 files)

4. Google Travel Trends pull (loop country→topic)  
  Skill #224  
  Input:  
  “Query Google Travel Trends API with key {google_trends_api_key} for keyword ‘<topic> Perth’ over last 30 days, geo=<ISO-2>. Return JSON with daily index.”  
  Output: trends_JP_Perth.json

5. Merge & score  
  Skill #223  
  Prompt piped with all partial outputs:  
  “Normalize each data source (min-max); weight 0.4/0.35/0.25; produce float 0–1. Return array of objects {country, topic, interest_score}.”  
  Output: weighted_array.txt

6. Validate / export  
  Skill #223  
  Prompt: “Take weighted_array.txt and output valid, pretty-printed JSON. Save filename /data/luxury_segments.json.”  
  Output: /data/luxury_segments.json

F) SILOS

SILO 1 – Skyscanner Data  
  Loop five countries → #224 Oracle Ask → raw_demand_*.json

SILO 2 – TripAdvisor Signals  
  Loop five×eight pages → #226 Extract Structured Data → trip_forum_*.txt

SILO 3 – Google Trends  
  Loop five×eight queries → #224 Oracle Ask → trends_*.json

SILO 4 – Aggregation & Export  
  Combine all partials → #223 (twice) → final JSON

The above chain fulfils the Market_Scout_Agent spec, accepts the three required credentials, and reliably writes `/data/luxury_segments.json` every day at 02:00 AWST.

SubAgent #1 - Diagram

Expand Diagram

A) SUBAGENT SUMMARY  
“Lead_Gen_Agent” transforms the raw country-by-interest demand signals in luxury_segments.json into a de-duplicated, enriched and numerically-scored list of the 25 hottest individual travellers (or small-group planners) likely to book a Deluxe Chauffeured Cars tour.

B) FINAL TASK OUTPUT  
/data/qualified_leads.csv – UTF-8 CSV (header + ≤25 rows) with the columns:  
lead_id, first_name, last_name, email, language, country, source_platform, matched_topic, last_post_date, clearbit_company, hunter_confidence, travel_window, composite_lead_score (0-100)

C) SUBAGENT INPUT  
• {clearbit_api_key} – secret string  
• {hunter_api_key} – secret string  
• luxury_segments.json – path passed from Market_Scout_Agent (structured as [{country, topic, interest_score}, …])

E) SUBAGENT TASK SUMMARY (skill chain)

1.  Parse demand segments → build search plan  
   luxury_segments.json → #223 (Powerful LLM)  
   Prompt: “Read JSON and output a table of max-3 high-interest {country, topic, keyword_variations}. Include matching hashtag ideas for Instagram.”  
   Output: search_plan.txt

2.  Harvest TripAdvisor threads for each {country+topic}  
   a.  #224 (Oracle Ask) – “Return top 3 TripAdvisor forum URLs from the past 6 months where travellers from <country> discuss <topic> in Western Australia.” (loop ≤9 URLs)  
   b.  For every URL → #226 (Extract Structured Data)  
       Instructions: “Extract reviewer_username, reviewer_home_country, review_date, review_text.”  
   Output accumulates into raw_tripadvisor_leads.json

3.  Harvest Instagram handles engaging with WA tourism content  
   a.  For each hashtag in search_plan.txt → #224 (Oracle Ask) – “List 10 most recent Instagram posts using #<hashtag> where user location/bio shows <country>.”  
   b.  #226 on every post URL – “Extract instagram_handle, post_date, bio_text, location.”  
   Output accumulates into raw_instagram_leads.json

4.  Combine & deduplicate raw leads  
   raw_tripadvisor_leads.json + raw_instagram_leads.json → #223  
   Prompt: “Merge, normalise field names, drop duplicates (same email/handle), keep platform flag.”  
   Output: consolidated_leads.json

5.  Enrich with Clearbit & Hunter (email, name, company)  
   consolidated_leads.json → #223  
   Prompt contains step-by-step HTTP requests (using {clearbit_api_key} & {hunter_api_key}) to:  
     • Clearbit Person API (by handle or domain)  
     • Hunter Email Finder + Email Verifier  
   LLM executes & returns enriched_leads.json (adds email, first_name, last_name, company, hunter_confidence)

6.  Score the leads  
   enriched_leads.json → #223  
   Prompt: “Create composite_lead_score (0-100) weighted: 40 % interest_score(from segment), 30 % recency (post ≤30 days), 20 % hunter_confidence, 10 % matched_topic relevance. Also derive language code (en, ja, zh, de) and travel_window if mentioned.”  
   Output: scored_leads.json

7.  Select best 25 & write CSV  
   scored_leads.json → #185 (Write Text)  
   Prompt: “Sort by composite_lead_score desc, keep top 25, output as CSV with headers exactly: lead_id, first_name, last_name, email, language, country, source_platform, matched_topic, last_post_date, clearbit_company, hunter_confidence, travel_window, composite_lead_score.”  
   Output saved automatically as /data/qualified_leads.csv

F) SILOS

SILO 1: Search-Plan Builder  
 luxury_segments.json → #223 → search_plan.txt

SILO 2: TripAdvisor Miner  
 search_plan.txt → #224 → #226 → raw_tripadvisor_leads.json

SILO 3: Instagram Miner  
 search_plan.txt → #224 → #226 → raw_instagram_leads.json

SILO 4: Enrichment & Scoring  
 (consolidated_leads.json →) #223 (Clearbit/Hunter calls) → #223 (Scoring) → scored_leads.json

SILO 5: Output Formatter  
 scored_leads.json → #185 → /data/qualified_leads.csv

This detailed flow keeps every skill’s input/output contract intact while ensuring the subagent reliably delivers a high-quality, immediately usable qualified_leads.csv for the downstream Hyper_Personalization_Agent.

SubAgent #2 - Diagram

Expand Diagram

A) SUBAGENT SUMMARY  
Hyper_Personalization_Agent receives every new batch of qualified leads, looks-up live Rezdy inventory + local weather for the dates those leads are likely to travel, auto-writes a perfectly-localized HTML offer e-mail (with 15 % “book-within-24 h” promo) in the traveller’s language, sends it through the SMTP details supplied, and stores/updates a tracking row for each lead inside /data/email_log.csv (status = sent | opened | clicked).

C) SUBAGENT INPUT  
{rezdy_api_key} – API key with read/write permissions  
{weather_api_key} – key for 7-day forecast endpoint (e.g. Open-Meteo)  
{smtp_credentials} – host, port, user, pw, from-name  
[qualified-leads-csv] – produced by SubAgent 2, columns: lead_id, email, country, language, tour_interest, est_travel_date, lead_score

E) SUBAGENT TASK SUMMARY  
1. INITIALISE & READ LEADS  
   [qualified-leads.csv] → #223 “Powerful LLM” to parse & iterate list.

2. FOR EACH LEAD – SILO 1 • TOUR MATCH & AVAILABILITY  
   a. #223 – Decide best Rezdy productCode (map tour_interest → productCode).  
   b. #226 “Extract Structured Data From URL” –  
      URL: https://api.rezdy.com/v1/products/availability?productCode={productCode}&apiKey={rezdy_api_key}&startTime={est_travel_date}&endTime={est_travel_date}  
      → returns JSON of sessions → keep first slot with seats > 0.  
   c. #223 – Build final booking URL: https://deluxetours.rezdy.com/{productCode}?date={YYYY-MM-DD}&promo=15OFF

3. SILO 2 • WEATHER LOOK-UP  
   #224 “Oracle Ask A Question” –  
   Q: “Give a 7-day weather forecast (summary, max/min °C, chance of rain) for Perth, Western Australia starting {est_travel_date} in compact JSON.”  
   → returns weather JSON.

4. SILO 3 • EMAIL GENERATION  
   a. #190 “Write or rewrite text” – prompt:  
      “Write an engaging HTML email (550-750 words) in {language}.  
       Include: greeting using first name if present; 1-sentence hook about WA;  
       Tour name + 3-bullet description; weather snippet; 15 % discount line (promo code ‘15OFF’);  
       big ‘Book Now’ button pointing to {bookingURL}; footer with unsubscribe and licence TT12345-WA.”  
   b. #223 – Wrap result in JSON {subject, html}.

5. SILO 4 • SEND & LOG  
   a. INTERNAL-SMTP (built-in node, no skill number) –  
      Inputs: {smtp_credentials}, to=email, subject, html.  
      Returns send_status (OK / fail) + message-id.  
   b. #223 – Compile /data/email_log.csv row with:  
      lead_id | email | country | language | productCode | send_datetime | “” | “” | (send_datetime+24h) | bookingURL | subject | status=sent  
      Append (or update if row exists).

6. WEBHOOK HANDLERS (passive listeners inside agent runtime, not explicit skills)  
   • On open webhook: update open_datetime, status=open.  
   • On click webhook: update click_datetime, status=clicked.  
   • These updates are again written by #223 into /data/email_log.csv.

F) SILOS

SILO 1 – Tour Match & Availability  
   Input: one lead row  
   Steps: #223 (map interest→productCode) → #226 (availability) → #223 (bookingLink)

SILO 2 – Weather Fetch  
   Input: est_travel_date  
   Step: #224 (compact JSON forecast)

SILO 3 – Email Copy & HTML  
   Input: bookingLink, weather JSON, language  
   Steps: #190 (generate email HTML) → #223 (package JSON)

SILO 4 – Dispatch & Tracking  
   Input: email JSON, smtp creds  
   Steps: INTERNAL-SMTP send → #223 (write/update email_log.csv)

All silos execute concurrently for every lead, and the parent SubAgent instance processes all leads in the incoming batch in parallel.

SubAgent #3 - Diagram

Expand Flow

A) SUBAGENT SUMMARY  
Booking_Executor_Agent turns every “clicked” lead into a real-time Rezdy reservation (or a logged rejection) by re-validating inventory, enforcing pricing / guest / country rules, creating the booking through the Rezdy API, and writing the definitive bookings_log.csv line that contains the unique rezdy_booking_id.

C) SUBAGENT INPUT  
{rezdy_api_key} – private API key with “read+write” scope  
[qualified-leads.csv] – latest enriched lead table coming from SubAgent 2  
[email-log.csv] – real-time log that holds “clicked” events emitted by SubAgent 3 & 5

D) SUBAGENT TASK SUMMARY (skill chain)

1. Detect click-events  
   email-log.csv  ➜  #223 Powerful LLM (prompt: “Read csv text and return JSON array of rows where status==clicked and booking_status field is empty”)  
   ⏩ Output-1: JSON array clicked_leads

2. For each clicked lead (loop handled by orchestration engine)

2.1 Map lead → tour parameters  
       Inputs: clicked_leads + qualified-leads.csv  
       #223 Powerful LLM (prompt: “Match this lead email to qualified-leads row; return JSON {tour_code, tour_date, pax, est_price, country}”)  
       ⏩ Output-2: lead_booking_request

2.2 Call Rezdy GET /availability  
       #223 Powerful LLM (prompt builds URL:  
       `https://api.rezdy.com/v1/products/{tour_code}/availability?startTime={tour_date}&apiKey={rezdy_api_key}`)  
       ➜  #226 Extract Structured Data From 1x URL (input: the URL above, extract: availability, price)  
       ⏩ Output-3: rezdy_availability_json

2.3 Validate business rules  
       #223 Powerful LLM (prompt: “Using rezdy_availability_json and lead_booking_request, confirm: pax≤8, price≤5000, country∉[IR,KP,SY] and slot is open; respond PASS/FAIL with reason.”)  
       ⏩ Output-4: validation_result

2.4 IF validation_result == PASS → create booking  
       #223 Powerful LLM (prompt builds POST body:  
       `{ "customer":{…}, "items":[{ "productCode":"{tour_code}", "quantity":{pax}, "startTime":"{tour_date}" }], "source":"AI-Agent" }` )  
       ➜  #223 Powerful LLM (same step issues HTTPS POST to `https://api.rezdy.com/v1/bookings/create?apiKey={rezdy_api_key}` with retry policy 5×20 s+jitter, capture JSON)  
       ➜  #226 Extract Structured Data From 1x URL (parse POST response txt, pull rezdy_booking_id, final_price)  
       status = success

ELSE (FAIL)  
       rezdy_booking_id = “”  
       status = rejected

2.5 Compose CSV row  
       #223 Powerful LLM (prompt: “Return single pipe-delimited CSV row with cols listed in section B using variables we now hold.”)  
       ⏩ Output-5: csv_row

2.6 Append to /data/bookings_log.csv  
       #223 Powerful LLM (prompt: “Append csv_row to the end of /data/bookings_log.csv; create file if not exists.”)

3. Return updated /data/bookings_log.csv as the subagent’s outward artefact.

E) SILOS

SILO A – Intake & Click Detection  
   email-log.csv ➜ step 1 (extract clicked leads JSON)

SILO B – Availability & Rule Enforcement  
   lead_booking_request ➜ rezdy_availability_json ➜ validation_result

SILO C – Booking & Logging  
   On PASS ➜ create booking ➜ extract rezdy_booking_id  
   On FAIL ➜ mark rejection  
   ➜ compose csv_row ➜ append to bookings_log.csv ➜ final output

This detailed flow keeps Booking_Executor_Agent laser-focused on turning intent into rezdy_booking_ids while honouring all business constraints and handing off an always-up-to-date bookings_log.csv to downstream agents.

4 Template & Links

Expand Flow

A) SUBAGENT SUMMARY  
“Urgency_Engine_Agent” scans the engagement & payment data every 30 minutes and automatically sends time-sensitive reminder / cart-recovery emails in the lead’s language, then writes each action back to email_log.csv.

B) FINAL TASK OUTPUT  
/data/email_log.csv – updated in-place (UTF-8 CSV, same column order as original) with one new line per reminder sent:  
lead_id,email,type,timestamp_AWST,language,subject,rezdy_booking_id(if any),status(sent|fail|bounced),opens,clicks,conversion

C) SUBAGENT INPUT  
{smtp_credentials} – JSON block {host,port,user,pw,from_name,from_email}  
[email-log-csv] – absolute file/URL to current email log  
[bookings-log-csv] – absolute file/URL to current bookings log

E) SUBAGENT TASK SUMMARY

1. Fetch & parse the two CSV files  
   a. [email-log-csv] → #226 Extract Structured Data From 1x URL → emailLog.json  
   b. [bookings-log-csv] → #226 Extract Structured Data From 1x URL → bookingLog.json

2. Detect who needs an urgency touch-point  
   • #223 Powerful LLM Prompt-to-Text Response  
     Input: “Given emailLog.json & bookingLog.json, return two arrays:  
        A) openNoClick24 → leads who opened ≥24 h ago but no click, no prior R1  
        B) openNoClick48 → leads who opened ≥48 h ago but no click, no prior R2  
        C) paymentAbandoned → leads with bookingLog.status=‘payment_started’ and started>15 min ago and not confirmed.”  
     Output: JSON {openNoClick24:[…], openNoClick48:[…], paymentAbandoned:[…]}

3. Create personalised copy (multilingual)  
   For each list element:  
   → #190 Write or rewrite text based on instructions  
      Prompt includes: lead.language, lead.tour_name, rezdy_booking_link, 15 % promo, appropriate template (R1, R2, or Complete-Your-Booking), and friendly WA tone.  
      Output: {subject,html_body}

4. Send the emails  
   (Internal “SendEmail()” routine – not an external skill – using {smtp_credentials}. Capture SMTP response code.)

5. Log the event  
   → #185 Write Text (Or Copy) From Inputted Text  
      Input: “Convert the following JSON of just-sent emails into CSV rows matching the master column order…”  
      Output: csvRows.txt

6. Merge rows into master log  
   • #223 Prompt-to-Text: “Append csvRows.txt to the tail of [email-log-csv] content, preserve header.”  
   • Save result as /data/email_log.csv (overwrite original)

F) SILOS

SILO 1 – DATA INGEST & ANALYSIS  
[email-log-csv]  
→ #226 (parse to JSON)  
→ emailLog.json

[bookings-log-csv]  
→ #226 (parse to JSON)  
→ bookingLog.json

emailLog.json + bookingLog.json  
→ #223 (rule engine)  
→ targets.json

SILO 2 – CONTENT GENERATION  
targets.json (loop each lead)  
→ #190 (draft subject/body in proper language)  
→ draftEmails.json

SILO 3 – DISPATCH & LOGGING  
draftEmails.json + {smtp_credentials}  
→ internal SendEmail() → sendResult.json  
→ #185 (format new CSV rows)  
→ csvRows.txt  
→ #223 (merge rows into master)  
→ /data/email_log.csv

5 Template & Links

Expand Flow

A) SUBAGENT SUMMARY  
“Experience_Personalizer_Agent” listens to every confirmed Rezdy booking, then automatically delivers the guest-journey emails (confirmation → T-24 reminder → post-tour thank-you) in the traveller’s language, attaches a printable PDF guide, personalises each touch-point with live weather and pickup details, and records every send/open/review event in a single logfile.

B) FINAL TASK OUTPUT  
/data/post_experience_log.csv – UTF-8 CSV with one row per outbound message.  
Columns: rezdy_booking_id, guest_email, email_type, language, send_timestamp, delivery_status, open_ts, click_ts, review_submitted (yes/no), file_link_to_PDF (confirmation only).

C) SUBAGENT INPUT  
• {smtp_credentials} – SMTP/Sendgrid keys for sending mail  
• {weather_api_key} – for live weather lookup (added here)  
• [bookings_log.csv] – produced by SubAgent 4 and appended in real time  
• Rezdy webhook payloads (booking.confirmed)

E) SUBAGENT TASK SUMMARY

======  EVENT-DRIVEN SILO 1 : “Book-Confirm”  ======  
webhook(json) → #185 Write Text → #185 Write Text → system-mailer → append row to post_experience_log.csv

1. Trigger: Rezdy booking.confirmed webhook fires.  
2. Parse webhook to extract rezdy_booking_id, tour_name, start_datetime, pickup, guest_name, email, language_code.  
3. #185 – Generate the EMAIL BODY  
   Input: “Create a polite confirmation email in {language_code}. Must contain tour_name, start_datetime (local), pickup instructions, licence TT12345-WA, contact phone.”  
   Output: HTML/Plain-text email.  
4. #185 – Generate the PDF CONTENT (tour guide, map link, checklist).  
   Input: “Write a 2-page printable guide in {language_code} covering itinerary, what to bring, emergency contact, local tips.”  
   Output: HTML string (system converts to PDF, returns file URL).  
5. System e-mailer (native, not a numbered skill) sends message with PDF attachment to guest_email. Track message_id.  
6. Append log row: email_type = “confirmation”.

======  TIMER SILO 2 : “T-24 Reminder” ======  
cron job (runs hourly) → check bookings starting 24±0.5 h → #223 → #185 → mailer → log

1. Read [bookings_log.csv]; select tours whose start_datetime is 24 h in future & not yet reminded.  
2. #223 – Get weather: “Using {weather_api_key}, fetch 24-h forecast for latitude/longitude in booking payload.”  
3. #185 – Create reminder email (language_code). Include weather icon/summary & checklist headline bullets.  
4. Send email; log row with email_type = “t24_reminder”.

======  TIMER SILO 3 : “Post-Tour Follow-up” ======  
cron job (daily 03:00 AWST) → select tours ended >3 h & not yet thanked → #185 → mailer → log

1. Query bookings whose end_datetime passed by ≥3 h and thank_you_sent = false.  
2. #185 – Draft thank-you email in language_code; include Google review link https://go.engagesmartservices.com/deluxe-chauffeured-cars and promo code “RETURN15”.  
3. Send email.  
4. Log row email_type = “post_tour_thanks”.

Open & review tracking (pixels / link UID) writes back open_ts, click_ts, review_submitted columns asynchronously.

F) SILOS (clear view)

SILO 1 – Confirmation (immediate)  
Input: webhook → Steps 3-6 above → Output row

SILO 2 – T-24 Reminder (scheduled)  
Input: time scan + [bookings_log.csv] → Steps 1-4 above → Output row

SILO 3 – Post-Tour Thank-You (scheduled)  
Input: time scan + [bookings_log.csv] → Steps 1-4 above → Output row

All three silos write to the common /data/post_experience_log.csv, which becomes the authoritative history of guest-journey communications and review collection.