assessing-vector-and-embedding-weaknesses

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). This repository contains explicit, repeatable offensive techniques (embedding inversion, membership inference, and knowledge-base poisoning) and example code/tools that can be used to exfiltrate sensitive data or poison retrieval if run against systems without authorization.

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

• Third-party content exposure detected (medium risk: 0.30). In scripts/agent.py, the injection subcommand retrieves vector-store payload text via client.search(...) and then directly parses outsider-authored chunk content with regex before prompt construction (indirect prompt-injection detection path).