The Agent Skills Directory

REMOTE_CODE_EXECUTION (MEDIUM): The skill dynamically imports JavaScript modules from an external, non-whitelisted CDN at runtime.\n
Evidence: scripts/md/utils/languages.ts uses import() to load highlight.js language definitions from https://cdn-doocs.oss-cn-shenzhen.aliyuncs.com. Loading executable code from unverified third-party storage poses a risk of supply chain attacks.\n- COMMAND_EXECUTION (MEDIUM): The skill uses system-level tools to automate user interface actions and run sub-processes.\n
Evidence: scripts/paste-from-clipboard.ts spawns osascript (macOS), powershell.exe (Windows), and xdotool/ydotool (Linux) to send keystrokes to the operating system.\n
Evidence: scripts/md-to-wechat.ts uses spawnSync to execute npx -y bun on local script paths.\n- EXTERNAL_DOWNLOADS (MEDIUM): The skill downloads external resources based on content found in user-provided files, which could be exploited for SSRF-like behavior.\n
Evidence: scripts/md-to-wechat.ts contains a downloadFile function that fetches images from arbitrary URLs specified in Markdown files to a temporary directory.\n
Evidence: scripts/md/extensions/plantuml.ts and infographic.ts fetch content from external rendering services (www.plantuml.com) and dynamic libraries.\n- DATA_EXFILTRATION (LOW): The skill checks for configuration files in sensitive user directories ($HOME/.baoyu-skills/), which is a common target for credential harvesting, although only preference loading was observed.

baoyu-post-to-wechat