reinforcement-learning-engineer
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-defined tasks for RL environment design and policy training, creating a surface for indirect prompt injection. While the skill includes rigorous safety and reward-hacking checks, it lacks explicit input sanitization or boundary markers when handling external task data within a writable workspace.
- Ingestion points: User-provided problem definitions and MDP parameters.
- Boundary markers: Absent.
- Capability inventory: File system write access via the workspace-write sandbox mode.
- Sanitization: Not explicitly specified for input processing.
Audit Metadata