The Agent Skills Directory

[SAFE]: The skill implements a structured workflow for static code analysis using benign system utilities such as grep, ripgrep (rg), and find. These operations are limited to searching the local filesystem for specific architectural patterns.
[SAFE]: Technical recommendations provided by the skill (e.g., torch_npu, mindiesd, yunchang) are standard libraries and frameworks within the Huawei Ascend NPU ecosystem.
[SAFE]: Although the skill processes untrusted user code (Indirect Prompt Injection surface), the instructions are constrained to scanning and reporting findings. The agent is not directed to execute the files it reads, and the structured output template further limits the potential impact of adversarial content within scanned files.

ascend-migration-analysis