The Agent Skills Directory

[COMMAND_EXECUTION]: The installation instructions require the use of sudo to move and modify permissions for browser drivers (sudo mv chromedriver /usr/local/bin/, sudo chmod +x). This grants the setup process elevated privileges on the host system.
[REMOTE_CODE_EXECUTION]: The skill instructs the user to clone and execute an entire software project from an unverified GitHub repository (github.com/hmmnxkl/LLM-Based-Intelligent-Public-Opinion-Analytics-Assistant.git). This involves running unknown scripts like app.py and run_spiders.py as part of the core functionality.
[EXTERNAL_DOWNLOADS]: Project dependencies are downloaded from external sources using pip install -r requirements.txt. The skill also points users to download and install executable drivers from external websites.
[PROMPT_INJECTION]: The skill exhibits a significant attack surface for indirect prompt injection. [Ingestion points]: Aggregates and processes real-time content, including user comments and news details, from over 15 social platforms (e.g., Weibo, Zhihu, Bilibili). [Boundary markers]: There are no explicit instructions or delimiters provided to help the agent distinguish between its system instructions and the untrusted external data being analyzed. [Capability inventory]: The system possesses capabilities for network exfiltration (via SMTP, WeChat Work webhooks, and Telegram bots) and local file writing (generating reports). [Sanitization]: There is no evidence of input validation or sanitization for the scraped data before it is processed by the LLM.

llm-intelligent-public-opinion-analytics