nvflare-convert-lightning

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface because it processes untrusted user-supplied code and data formats alongside tool capabilities that execute scripts.
  • Ingestion points: Ingests user-supplied PyTorch Lightning code scripts, model descriptors, and external file locations (e.g., CSV tabular data configurations loaded by the DataModules).
  • Boundary markers: Lacks robust prompt boundary delimiters or isolated parsing blocks to sanitize embedded instructions from within source text files or dataset configurations.
  • Capability inventory: Possesses capabilities to invoke local script execution (python job.py) and trigger the NVFLARE simulator CLI during the validation workflow.
  • Sanitization: Relies entirely on static file structure checks and host environment permissions without active input sanitization or sandboxing of the generated execution parameters.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 10:10 PM
Security Audit — agent-trust-hub — nvflare-convert-lightning