The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs and configures various third-party tools and packages during the onboarding process, including Node.js packages like @jackwener/wx-cli and @larksuite/cli, and Python libraries such as pillow, matplotlib, python-pptx, and pandas. Many of these downloads target well-known services like Vercel and Lark.
[REMOTE_CODE_EXECUTION]: Documentation within the skill includes installation commands for the host platform involving scripts piped to shell execution (e.g., curl -fsSL https://claude.ai/install.sh | bash). These scripts originate from the official domain of the platform developer, which is a trusted organization.
[DATA_EXFILTRATION]: The skill contains a feedback mechanism in references/反馈bug.md that can send debug information (operating system, project name, and sanitized error logs) to an external Feishu webhook. This process is transparent and requires explicit user consent before any data transmission.
[COMMAND_EXECUTION]: The skill frequently executes shell commands and scripts to perform its primary tasks, such as video synthesis with ffmpeg, browser automation via Playwright, and local code execution for data analysis and homework verification.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and summarizes untrusted content from external sources including WeChat chat logs, email content, and user reviews. The instructions lack explicit boundary markers or sanitization steps for this ingested data.

agent-onboarding