workflow-cp2-2

Reviews Ascend C operator development plans against acceptance checklists and user confirmation.

Updated Sep 15, 2026
One-click install
npx skills add https://github.com/WangWindow/CANN-BatchMatMulMaxsum --skill workflow-cp2-2-wangwindow
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: workflow-cp2-2
Source: https://github.com/WangWindow/CANN-BatchMatMulMaxsum/tree/main/.agents/skills/workflow-cp2-2
Command: npx skills add https://github.com/WangWindow/CANN-BatchMatMulMaxsum --skill workflow-cp2-2-wangwindow

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? It provides QA reviewers with a structured acceptance standard for judging whether an Ascend C operator development plan is sound and feasible before direct-launch kernel coding begins, preventing unverified designs from reaching implementation. ## Core Features & Use Cases - Checklist-Based Review: Verifies architecture alignment, schema consistency with proto.yaml, Buffer planning, Tiling strategy, multi-core partitioning, and Ascend C API validation against authoritative sources. - Two-Stage Acceptance: Runs QA objective checks first, then generates a structured questionnaire of key decision points for item-by-item user confirmation. - Semantic Attribution: Classifies every defect or objection as belonging to the development plan or the requirements document, routing rework to the correct upstream stage. - Use Case: A QA engineer loads this skill at the CP2.2 checkpoint to review a BatchMatmulMaxSum operator design document, produces structured revision feedback for failed items, and archives questionnaire results under .cannbot/<operator>/questionnaires/. ## Quick Start Load this skill to review the development plan document for the current operator and run the CP2.2 acceptance checks.

Frequently Asked Questions about workflow-cp2-2

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I review an Ascend C operator development plan before coding?

Load this skill at the CP2.2 checkpoint and check the plan against the acceptance items: architecture alignment, schema consistency with proto.yaml, Buffer planning, Tiling strategy, multi-core partitioning, and verified Ascend C API usage. Failed items produce structured revision feedback.

What does the CP2.2 acceptance check verify in a design document?

It verifies architecture inheritance from the requirements document, schema alignment with the evaluation proto.yaml, completeness of Buffer/Tiling/multi-core designs, interface feasibility on the target chip, resource constraint consistency, bottleneck predictions with quantified reasoning, and optimization items with landing requirements.

How are Ascend C API choices validated during plan review?

Validation depends on the implementation path: Cube/Blaze routes must be verified against ascendc-blaze-best-practice, RegBase/MemBase/SIMT routes against ascendc-docs-search official documentation, and MC2 fusion designs against ascendc-mc2-best-practice. Unverified APIs cause the review to fail.

When does the user confirmation questionnaire get sent?

The questionnaire is sent only after all QA checklist items pass. It extracts key decision points from the development plan template sections, each with decision content, source basis, semantic attribution label, and confirm/object options, then archives results under .cannbot/<operator>/questionnaires/.

What happens when a review item fails or the user objects?

Failed items produce structured revision feedback without entering user confirmation. Each issue is attributed semantically to either the development plan or the requirements document, and requirement-attributed issues are returned upstream for re-decision rather than patched in the plan.