search

name	search
description	Use when orchestrating a full apartment search run — reads config, fans out parallel scraper agents per platform, collects results, deduplicates against cache, and sends Telegram notifications for new listings

Search Orchestrator

Coordinates the full apartment search pipeline: scrape all platforms, dedup, notify via Telegram.

Input

Either:

• From config: Read ~/.claude/plugins/find-apartments/data/config.yaml for search profiles
• From CLI args: city and min_rooms provided directly (one-off search)

If CLI args are provided, use them as a single search profile. Otherwise, use all profiles from config.

Pipeline

Phase 1: Load Config

Read ~/.claude/plugins/find-apartments/data/config.yaml.

If the file doesn't exist, stop and tell the user:

"No config found. Create ~/.claude/plugins/find-apartments/data/config.yaml with your search profiles and Telegram credentials. See the plugin README for the format."

Extract:

• searches — array of search profiles
• telegram.bot_token and telegram.chat_id
• facebook_groups — optional

Phase 2: Load Cache

Invoke the cache skill to read the current cache from ~/.claude/plugins/find-apartments/data/cache.json. Collect all known URLs.

Also check for un-notified listings (notified: false) — these will be retried alongside new findings.

Phase 3: Fan Out Scrapers

For each search profile, launch scrapers:

Parallel group (web-based, launch all at once using the Agent tool):

• Invoke search-yad2 skill — pass city, min_rooms, max_price
• Invoke search-winwin skill — pass city, min_rooms, max_price

Sequential group (Chrome-based, one at a time):

• Invoke search-madlan skill — pass city, min_rooms, max_price
• Invoke search-facebook skill — pass city, min_rooms, facebook_groups from config

Launch the parallel agents using the Agent tool with run_in_background: true for Yad2 and WinWin. Run Madlan and Facebook sequentially in the foreground.

Wait for all agents to complete and collect their listing arrays.

Phase 4: Dedup Against Cache

Combine all listings from all scrapers into one array. For each listing:

• If listing.url exists in the cache AND notified is true → skip (fully processed)
• If listing.url exists in the cache but notified is false → include for retry
• If listing.url is NOT in the cache → new listing, include

Phase 5: Update Cache

Invoke the cache skill to add all new listings to cache with notified: false.

Phase 6: Send Telegram Notifications

Invoke the notify skill with the new + un-notified listings, bot_token, and chat_id from config.

If successful, invoke the cache skill to update notified: true for the notified URLs.

Phase 7: Summary

Report to the user:

## Search Complete

- **Profiles searched:** N cities
- **Platforms scraped:** Yad2, WinWin, Madlan, Facebook (list which succeeded)
- **Total listings found:** X
- **New listings:** Y (not seen before)
- **Telegram notifications sent:** W messages

### Skipped Platforms
- [List any platforms that were unavailable with reason]

### Also Check Manually
- WhatsApp groups for {city} real estate
- [Any Facebook groups that couldn't be accessed]

Error Recovery

• If one scraper fails, continue with the others. Report the failure in the summary.
• If Telegram fails, listings remain in cache with notified: false for next run.
• Never let a single platform failure abort the entire pipeline.