coldiq-search-enrich

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The skill documentation describes features for large-scale harvesting of personal contact data (personal emails, phone numbers, 500M-profile search, scrapers) and includes an explicit webhook callback (notification_url) and BYOK/proxying patterns that enable automated exfiltration of harvested data to arbitrary external endpoints — high abuse potential though no hidden backdoor or remote-code execution is present in the docs.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.65). The required workflow includes a “Web search” endpoint (/v1/limadata/search/web, which corresponds to runtime-fetched arbitrary public web content) that can return scraped page text which the agent would then place into the LLM context, enabling indirect prompt injection from outsider-authored web content.