wechat-local-vault

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). This code deliberately hooks the WeChat process with Frida to capture PBKDF2 outputs (passwords/derived keys), saves those keys to local key files and then decrypts SQLCipher databases into plaintext, which constitutes intentional credential theft and local data exfiltration capability (no remote network exfiltration is present in the code, but the captured keys and plaintext DBs enable abuse).

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

• Third-party content exposure detected (high risk: 0.85). Runtime LLM context can be fed by outsider-authored free text because scripts/vault_cli.py (and export_chat.py) reads decrypted SQLite fields like message_content / contentDesc / summary from the user’s local WeChat DBs, which contain other people’s chat text, and then returns it as JSON/text that an agent would place into the LLM context.