streaming
Pass
Audited by Gen Agent Trust Hub on Feb 20, 2026
Risk Level: SAFE
Full Analysis
- Prompt Injection (SAFE): No instructions were found that attempt to bypass AI constraints or override system prompts.
- Data Exposure & Exfiltration (SAFE): No hardcoded credentials or access to sensitive local files were detected.
- Obfuscation (SAFE): The code is clear and does not use encoding or hidden characters to mask its functionality.
- Unverifiable Dependencies & Remote Code Execution (SAFE): The skill does not download or execute external scripts or packages.
- Privilege Escalation (SAFE): No attempts to gain administrative or elevated system access were found.
- Persistence Mechanisms (SAFE): The skill does not attempt to establish long-term access or background processes.
- Metadata Poisoning (SAFE): Metadata is used correctly for descriptive purposes.
- Indirect Prompt Injection (SAFE): The skill ingests user text through chunking commands but lacks dangerous capabilities to act on injected instructions. Evidence: 1. Ingestion points: 'args' in index.ts; 2. Boundary markers: absent; 3. Capability inventory: none; 4. Sanitization: absent.
- Time-Delayed / Conditional Attacks (SAFE): No time-based or environment-gated malicious logic was found.
- Dynamic Execution (SAFE): Uses a static relative path for a dynamic import ('../../../streaming/index') which is a standard internal dependency pattern.
Audit Metadata