nebius-batch-synthetic
Pass
Audited by Gen Agent Trust Hub on Mar 8, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [SAFE]: Analysis of the provided scripts and instructions confirms that the skill performs legitimate API operations for synthetic data generation without malicious patterns.
- [EXTERNAL_DOWNLOADS]: The skill interacts with 'https://api.tokenfactory.nebius.com/v1/', which is the official endpoint for Nebius, a well-known AI infrastructure provider.
- [COMMAND_EXECUTION]: Includes instructions for standard package installation ('pip install openai') and environment variable configuration ('export NEBIUS_API_KEY').
- [PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it ingests untrusted LLM-generated content from the Nebius API and writes it to local files ('training.jsonl') for fine-tuning. Evidence: (1) Ingestion point: 'client.files.content' reads remote model output in SKILL.md. (2) Boundary markers: None present. (3) Capability inventory: File write operations in SKILL.md. (4) Sanitization: Only basic length filtering is implemented. This is a known architectural risk for synthetic data workflows and does not indicate a malicious skill.
Audit Metadata