Skills
skills/ascend-ai-coding/awesome-ascend-skills/ai-for-science-deepfri-tf-npu/Gen Agent Trust Hub

ai-for-science-deepfri-tf-npu

Pass

Audited by Gen Agent Trust Hub on Mar 28, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill fetches the DeepFRI source code and pre-trained models from the Flatiron Institute's official research repositories. It also downloads the Bazel build tool from a Huawei Cloud mirror. These sources are recognized as well-known and appropriate for the technical scope of the skill.
  • [COMMAND_EXECUTION]: Provides an example verification script (verify_accuracy.py) that uses os.system to automate the execution of the prediction pipeline for output comparison. This is a standard pattern for local developer testing and verification workflows.
  • [SAFE]: The patches provided for model adaptation, specifically replacing CuDNNLSTM with standard LSTM layers, are transparent and technically sound solutions for running GPU-optimized TensorFlow models on Ascend NPUs. The skill correctly implements npu_device initialization according to the vendor's TF Community guidelines.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 28, 2026, 02:16 AM
Security Audit — agent-trust-hub — ai-for-science-deepfri-tf-npu