ai-post-training
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFENO_CODE
Full Analysis
- [SAFE]: The skill consists entirely of Markdown documentation and a JSON metadata file containing references to academic papers, technical books, and open-source documentation. No executable code (scripts, binaries, or configuration files that trigger code execution) is present in the skill.
- [EXTERNAL_DOWNLOADS]: The skill references several external resources, including ArXiv research papers, technical books from Manning and O'Reilly, and official GitHub repositories for well-known projects like Hugging Face TRL, verl (Volcano Engine), and SymPy. These are well-known services and organizations, and the references are documented neutrally for educational purposes.
- [REMOTE_CODE_EXECUTION]: There are no patterns of remote code execution. The skill discusses training frameworks (TRL, verl) and mathematical libraries (SymPy) as tools for a user to implement, but does not attempt to install or execute them itself.
- [DATA_EXFILTRATION]: No network operations or sensitive file access patterns were detected. The skill focuses on the conceptual and diagnostic aspects of LLM alignment and post-training.
- [OBFUSCATION]: No obfuscation techniques such as Base64 encoding, zero-width characters, or homoglyph attacks were found in the provided files. All text is clear and readable technical English.
Audit Metadata