gpu-kernel-episode-loop

Pass

Audited by Gen Agent Trust Hub on Sep 5, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple external sources which could be manipulated to influence agent behavior.
  • Ingestion points: Reads definition.json, workload.jsonl, memory/v*.json, and structured responses from gpu-wiki tools.
  • Boundary markers: The instructions do not specify the use of delimiters or boundary markers when interpolating this data into the context.
  • Capability inventory: The skill utilizes tools/sandbox.py to execute shell scripts and profile drivers, and allows modification of the kernel.py source file.
  • Sanitization: There is no mention of sanitizing or validating the content of the workload files or wiki responses before processing.
  • [DYNAMIC_EXECUTION]: The skill instructs the agent to dynamically generate and execute Python scripts in specific scenarios.
  • Evidence: Instructions describe building a local fallback driver at <PROFILE_DIR>/harness/profile_driver.py when the seeded driver is insufficient. This generated script is then executed via python tools/sandbox.py to profile GPU kernels.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 5, 2026, 06:49 PM
Security Audit — agent-trust-hub — gpu-kernel-episode-loop