reinforcement-learning

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFE
Full Analysis
  • Reinforcement Learning Reference Material (SAFE): The skill provides structured guidance on RL algorithms (PPO, DQN, SAC) and RLHF pipelines. The content is technical and intended for helping users implement training loops correctly.
  • Static Code Validation (SAFE): The references/validations.md file defines regex patterns used to audit user code for stability issues (e.g., missing gradient clipping or entropy bonuses). These patterns are used for text-based analysis and do not involve dynamic code execution.
  • Data Privacy & Network (SAFE): No commands were found that access sensitive system files (e.g., SSH keys, environment variables) or attempt to initiate network connections to external domains.
  • Indirect Prompt Injection Surface (SAFE): While the skill is designed to process and review user-provided code (Category 8), it lacks high-risk capabilities such as code execution (eval/exec), filesystem writing, or network requests that could be abused via injected instructions in the analyzed data.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:03 PM