What problem does it solve? When writing or debugging custom Ascend NPU operators with PyPTO, reviewers need a consistent checklist of structural rules and a fast way to enumerate every pypto.* call site for op-by-op debugging. This Skill provides both: the layout/structure requirements every custom/<operator>/ directory must satisfy, and a script that lists all PyPTO call sites by line number. ## Core Features & Use Cases - Structure checklist: Enforces rules such as the three-piece set (impl + golden + test_<op>.py), mandatory use of assert_allclose or detailed_tensor_compare, staged _module<k>_impl.py naming, no Python loops driving kernels from the host wrapper, pypto.view rank consistency, and set_cube_tile_shapes [L0, L1] validity with L1 % L0 == 0. - Call-site extraction: scripts/extract_pypto_calls.py parses a kernel file with the Python ast module and outputs every pypto. call (including aliased imports and from-imports) ordered by line number, with an optional --json mode. - Use Case: While debugging a failing custom operator kernel, run the extractor to get an ordered list of pypto calls, then step through them one by one against the API documentation and the structure checklist to locate the defect. ## Quick Start Ask the assistant to review the custom/<operator>/ directory against the PyPTO layout requirements and list all pypto call sites in the kernel file for op-by-op debugging.
