SiYuan MCP's asset.upload tool allows reading arbitrary absolute file paths without workspace boundary checks, enabling an attacker who can influence the AI Agent to upload sensitive local files into the workspace. This vulnerability affects versions up to 3.8.0 and was patched in version 3.8.1. Exploitation requires user approval of a generic confirmation prompt, making the attack vector reliant on prompt injection and user interaction. The vulnerability leads to confidentiality breaches by exposing sensitive files such as SSH keys and cloud credentials via the workspace file API.