perf-activation-recompute

Pass

Audited by Gen Agent Trust Hub on May 19, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill provides instructional content and configuration templates for deep learning performance optimization. No malicious patterns, such as prompt injection, data exfiltration, or obfuscation, were detected.
  • [COMMAND_EXECUTION]: The skill includes instructions to run a performance benchmarking script (scripts/performance/run_performance_workload.py) and unit tests (pytest). These are standard operational procedures for validating performance settings in a machine learning development environment.
  • [INDIRECT_PROMPT_INJECTION]: The skill documents the ingestion of configuration parameters (e.g., recompute_modules) through command-line arguments and configuration files. While this represents a surface for data ingestion, the parameters are validated by the underlying library, and no malicious exploitation paths were identified.
Audit Metadata
Risk Level
SAFE
Analyzed
May 19, 2026, 05:48 PM
Security Audit — agent-trust-hub — perf-activation-recompute