llm-public-opinion-analytics-assistant

Pass

Audited by Gen Agent Trust Hub on May 20, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface due to its core function of ingesting and analyzing untrusted content from social media platforms.\n
  • Ingestion points: Data is crawled from 15 platforms (e.g., Weibo, Bilibili, Douyin) and news URLs via the DetailCrawler and DatabaseConnector components.\n
  • Boundary markers: There are no explicit instructions or delimiters shown in the logic to prevent the LLM from executing commands embedded in the crawled social media posts or news transcripts.\n
  • Capability inventory: The system is connected to a MySQL database and multiple push notification services (Email, WeChat, Telegram), creating a path for untrusted data to influence external communications or data storage.\n
  • Sanitization: No specific sanitization or filtering logic is documented for the external data before it is passed to the OpinionAnalyzer for sentiment and theme extraction.\n- [EXTERNAL_DOWNLOADS]: The skill facilitates the download of browser drivers (ChromeDriver, EdgeDriver) from official project domains and an LLM model from the Gitcode platform to support crawler and analysis functionality.\n- [COMMAND_EXECUTION]: Operation of the system involves running the main application server (app.py), database initialization scripts (init.py), and automated web crawlers using Selenium to simulate user interactions.
Audit Metadata
Risk Level
SAFE
Analyzed
May 20, 2026, 05:16 PM
Security Audit — agent-trust-hub — llm-public-opinion-analytics-assistant