nebius-datalab-pipeline

Pass

Audited by Gen Agent Trust Hub on Mar 9, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: Potential for indirect prompt injection via synthetic data processing. The skill ingests outputs generated by teacher models to construct fine-tuning datasets.
  • Ingestion points: batch_results processing within the curation step of SKILL.md.
  • Boundary markers: No delimiters or instructions to ignore embedded commands are present in the curation logic.
  • Capability inventory: The skill can launch fine-tuning jobs and deploy resulting model checkpoints.
  • Sanitization: The curation process only validates content length, which does not protect against malicious instructions embedded in synthetic outputs.
  • [COMMAND_EXECUTION]: Execution of pip install to manage the openai and requests Python libraries.
  • [EXTERNAL_DOWNLOADS]: The skill communicates with Nebius AI cloud endpoints (api.tokenfactory.nebius.com) and fetches official packages from standard registries.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 9, 2026, 02:02 AM