nvshmem-tune-performance
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill acts exclusively as a routing and coordination layer for NVIDIA NVSHMEM performance tuning. It explicitly forbids the agent from running benchmarks, editing source code, or executing system commands directly from within the router, delegating all technical operations to specialist skills that maintain their own safety boundaries.
- [INDIRECT_PROMPT_INJECTION]: The skill instructions allow for the ingestion and handoff of external data such as raw logs, performance reports, and source paths, which represents a potential attack surface for indirect prompt injection.
- Ingestion points: Performance reports, raw logs, source/kernel paths, and benchmark results supplied by the user (SKILL.md).
- Boundary markers: The instructions explicitly mandate preserving each specialist's execution boundary and safety rules (SKILL.md).
- Capability inventory: The skill is limited to routing logic and data handoff; it possesses no direct file-write, network, or subprocess capabilities as per its internal limitations (SKILL.md).
- Sanitization: The coordinator relies on the downstream specialist skills' internal guardrails rather than implementing local sanitization of the passed performance data.
Audit Metadata