What problem does it solve? It gives QA reviewers a consistent, criteria-driven way to judge whether all changed files in an Ascend C operator project meet submission quality, catching anti-cheat violations, spec drift, and documentation-code mismatches before evaluation. ## Core Features & Use Cases - Six-dimension inspection: Checks submission anti-cheat red lines (A-class rules A1-A7), coding standards (B-class rules), design consistency with proto.yaml, documentation-code synchronization, potential risks, and dead code cleanup. - Zero-tolerance anti-cheat gating: Flags blocking violations such as PyTorch/torch_npu proxy computation, CPU fallback, built-in operator routing, output caching, profiler tampering, and FakeTensor returns, which zero out the operator score in evaluation submissions. - Structured review report: Produces a pass/fail code review report using the CP5 template, with per-issue severity, file:line location, and documentation-vs-code discrepancy annotations. - Use Case: Before submitting a BatchMatmulMaxSum operator implementation for evaluation, load this skill at the CP5 checkpoint to verify no A-class red lines are violated and all delivered documents match the final code. ## Quick Start Load the workflow-cp5 skill and ask it to review all changed files in the repository against the repo-coding-rules and proto.yaml, then produce the CP5 code review report.
