Dataset Curator
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [Prompt Injection] (SAFE): No direct instructions to bypass safety filters or override system prompts were detected in the skill's instructional text.
- [Data Exposure & Exfiltration] (SAFE): No hardcoded credentials, sensitive file paths, or unauthorized network calls were found. The code snippets focus on local data processing.
- [Obfuscation] (SAFE): The content is entirely human-readable with no encoded strings or hidden Unicode characters.
- [Unverifiable Dependencies] (SAFE): The skill references 'cleanlab', a well-known open-source Python library for data quality. There are no commands to install untrusted third-party packages.
- [Indirect Prompt Injection] (LOW): This skill represents a surface for indirect prompt injection.
- Ingestion points: Processes external machine learning datasets (Workflow 1, 2, 3).
- Boundary markers: None explicitly defined in the suggested workflows to separate data from instructions.
- Capability inventory: The skill encourages the agent to execute Python code for profiling, cleaning, and augmentation (SKILL.md).
- Sanitization: No specific sanitization or validation of the dataset content is mentioned before processing.
Audit Metadata