The Agent Skills Directory

Prompt Injection (SAFE): The content is purely educational and technical. No instructions were found that attempt to override AI behavior, bypass safety constraints, or extract system prompts.
Data Exposure & Exfiltration (SAFE): No sensitive file paths, hardcoded credentials (API keys, tokens), or suspicious network operations were identified. The network calls are limited to standard model downloads from trusted sources like HuggingFace.
Obfuscation (SAFE): The skill contains no Base64-encoded commands, zero-width characters, or homoglyph-based evasion techniques. All code and documentation are in plain, human-readable text.
Unverifiable Dependencies & Remote Code Execution (SAFE): Dependencies such as mamba-ssm, torch, transformers, and causal-conv1d are well-known, industry-standard libraries for deep learning. There are no piped remote script executions (e.g., curl|bash).
Privilege Escalation & Persistence (SAFE): No commands requesting root/admin privileges (sudo) or attempts to establish persistence via shell profiles or cron jobs were found.

mamba-architecture