nvidia-skill-evaluator
Pass
Audited by Gen Agent Trust Hub on Nov 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides numerous CLI command examples for the skillevaluator tool. These commands are intended for environment setup, quality validation, and CI onboarding. All execution is gated by instructions requiring explicit user approval and environment preparation.- [EXTERNAL_DOWNLOADS]: The skill manages the installation of the NVIDIA/SkillEvaluator repository and its integrated security scanners (SkillSpector, Gitleaks). These resources are fetched from well-known/trusted sources, and the tool uses checksum verification to ensure integrity.- [DYNAMIC_EXECUTION]: The Tier 3 evaluation lane executes code from external skill packages. This is identified as a risky operation, which the skill mitigates by defaulting to Docker sandboxing and emphasizing that experimental host-based execution is less secure.- [INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect injection as it processes untrusted skill packages. Ingestion points: Evaluated files including SKILL.md and eval datasets. Boundary markers: Strong instructional boundaries requiring approval for tool actions and data egress. Capability inventory: Subprocess execution (skillevaluator), file system writes (reports, catalogs), and network egress (LLM providers). Sanitization: Use of integrated security scanners to validate ingested content before execution.
Audit Metadata