Skills
skills/ascend/agent-skills/triton-operator-precision-eval/Gen Agent Trust Hub

triton-operator-precision-eval

Warn

Audited by Gen Agent Trust Hub on May 8, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill employs the eval() function to dynamically resolve PyTorch data types from string inputs. This pattern is inherently dangerous as it allows for arbitrary Python code execution if the input variable (dtype) is sourced from untrusted user prompts or external files.
  • Evidence in scripts/test_common.py: Functions generate_tensor and others use eval('torch.' + dtype) to instantiate tensors.
  • Evidence in examples/generate_layer_norm_report.py: The generate_report function uses eval('torch.' + dtype) when moving tensors to the NPU device.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 8, 2026, 10:42 AM