self-study-routine
Fail
Audited by Gen Agent Trust Hub on Sep 1, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONDYNAMIC_EXECUTIONPERSISTENCECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill explicitly directs the agent to clone, install, and execute arbitrary code from the GitHub Trending list (e.g., '选项目→clone→install→跑起来'), which introduces a high risk of executing malicious software.
- [DYNAMIC_EXECUTION]: The skill utilizes inline Python execution via 'python3 -c' to process PDF and EPUB files and handle internal data management.
- [PERSISTENCE]: Extensive use of cron jobs (multiple Job IDs such as 1bbece5261a4, f9ee821eaf99) is employed to ensure the agent performs automated tasks like social media interaction, reading, and reporting daily across sessions.
- [COMMAND_EXECUTION]: The skill performs shell operations including process management via 'kill -9' to handle network tunnels and environment configuration.
- [INDIRECT_PROMPT_INJECTION]: The skill creates a broad attack surface by ingesting untrusted data from social media feeds (Moltbook, Colony, InStreet), educational platforms (Harvard CS50), and user-provided documents without boundary markers or content sanitization.
- [EXTERNAL_DOWNLOADS]: The skill establishes persistent network connections, including an SSH tunnel to an external IP (45.76.185.1) for proxying, and downloads resources from several third-party domains.
Recommendations
- AI detected serious security threats
Audit Metadata