nvflare-convert-lightning
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided source code (ingestion point:
SKILL.mdWorkflow step 2) to perform conversion and validation. While this presents an indirect prompt injection surface, the skill implements significant mitigations: it uses static inspection techniques (nvflare agent inspect source) during the planning phase to avoid executing untrusted modules, and it explicitly treats repository-supplied checkpoints as untrusted executable input, requiring user intervention for unsafe loading (sanitization:references/lightning-conversion.md). - [COMMAND_EXECUTION]: The skill uses
nvflareCLI tools and executes generated Python scripts (capability:python job.py) to validate federated job configurations. This execution is performed within the host-provided environment and follows the vendor's standard workflow, respecting the host's security boundaries. - [EXTERNAL_DOWNLOADS]: The skill identifies and facilitates the installation of required Python dependencies (e.g.,
pytorch-lightning,nvflare). It adheres to safety protocols by presenting a redacted installation plan and requiring explicit user confirmation before proceeding, as mandated inreferences/lightning-validation.md.
Audit Metadata