perf-optimization

Pass

Audited by Gen Agent Trust Hub on May 8, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill follows a least-privilege coordination model, explicitly delegating implementation, file editing, and benchmarking to domain-specific specialists while prohibiting the coordinator from direct code generation or file-system browsing.
  • [COMMAND_EXECUTION]: The coordination workflow includes remote execution on SLURM clusters via an SSH/srun wrapper. This is an expected and localized capability for performing performance profiling on targeted high-performance computing hardware.
  • [REMOTE_CODE_EXECUTION]: The agent manages the generation and execution of GPU kernels (such as Triton and CuTe DSL) on remote systems. This risk is managed by a structured workflow that mandates backups before modification and performance-driven validation with automatic rollback procedures.
  • [PROMPT_INJECTION]: The skill processes bottleneck analysis and recommendations from an external orchestrator. This represents an indirect prompt injection surface; however, the impact is limited by the agent's restricted action space and the requirement for verification of all outcomes.
Audit Metadata
Risk Level
SAFE
Analyzed
May 8, 2026, 02:11 PM
Security Audit — agent-trust-hub — perf-optimization