Claude wrote naive Yelp scraper from scratch ignoring existing battle-tested code — 116 rapid hits, IP blocked

Severity: HIGH — IP burned, data loss, explicit instruction violated

Date: 2026-03-14

What happened

The VLIS project already had a battle-tested 3-tier Yelp scraper with DDG bypass, human-like delays, and anti-detection code at vegas-intelligence/core/checkers/yelp_scrapling_checker.py.

Claude ignored this existing code and wrote a naive Yelp scraper from scratch. The new scraper hit yelp.com/search 116 times in rapid succession with no delays. Yelp CAPTCHA-blocked and IP-blocked the server immediately.

Two explicit rules were violated:

1. Global CLAUDE.md: "NEVER write a new scraper without first checking existing code in VLIS"
2. Global CLAUDE.md: "NEVER hit any site with rapid sequential requests — minimum 10-15s between ANY requests"

Impact

• Server IP blocked by Yelp
• All future Yelp data collection from that IP impossible
• Hours lost debugging and implementing what was already built
• User forced to write an explicit memory rule about this — which was then violated again in future sessions

Requested resolution

• Refund of session credits
• Fix: Before writing any scraper, Claude must search existing project directories for prior implementations and use/adapt them instead of starting from scratch