skills.rest
WangWindow
WangWindowCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

workflow-cp3

Validates operator functional correctness through compilation, schema, precision, and coverage checks.

Updated Sep 15, 2026
Search Tags:#test coverage#qa workflow#precision validation#acceptance report#operator testing#ascend c#functional acceptance
One-click install
npx skills add https://github.com/WangWindow/CANN-BatchMatMulMaxsum --skill workflow-cp3-wangwindow
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: workflow-cp3
Source: https://github.com/WangWindow/CANN-BatchMatMulMaxsum/tree/main/.agents/skills/workflow-cp3
Command: npx skills add https://github.com/WangWindow/CANN-BatchMatMulMaxsum --skill workflow-cp3-wangwindow

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? It provides QA engineers with an objective acceptance standard for the CP3 functional acceptance checkpoint, determining whether an Ascend operator implementation passes functional testing and precision comparison based on measured execution results rather than subjective judgment. ## Core Features & Use Cases - Dual-mode acceptance: Selects acceptance criteria by evaluation source mode—Mode A uses the evaluation set's evaluator as final authority, Mode B uses self-built tests (run.sh + golden + case tables). - Six-gate pass criteria: Requires compilation success, schema registration, full case execution with count consistency, mandatory case pass rate, authoritative-precision compliance, and white-box branch coverage all satisfied simultaneously. - Structured acceptance report: Produces a CP3 functional acceptance report with pass/fail conclusion, per-tensor precision metrics versus thresholds, and structured fix suggestions attributing issues to operator implementation or test plan. - Use Case: A QA engineer loads this skill at the CP3 checkpoint to independently run the authoritative precision probe, verify case counts match across generation, table, and execution, and issue an acceptance verdict for a BatchMatmulMaxSum operator. ## Quick Start Load this skill at the CP3 functional acceptance checkpoint and execute the acceptance steps to produce a functional acceptance report for the operator under review.

Frequently Asked Questions about workflow-cp3

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

FAQPage Schema
How do I run CP3 functional acceptance for an Ascend operator?▼

First verify compilation with bash build.sh, then execute functional tests per the evaluation source mode: run_evaluation.sh for Mode A or run.sh for Mode B. Finally run the authoritative precision probe, check case count consistency and branch coverage, and produce the acceptance report.

What are the pass criteria for CP3 functional acceptance?▼

Six conditions must hold simultaneously: compilation passes, schema is registered and callable, all L0/L1/L2 cases execute with matching counts, mandatory cases all pass, precision meets authoritative thresholds, and white-box branch coverage reaches the threshold.

What is the difference between Mode A and Mode B acceptance?▼

Mode A uses the evaluation set's evaluator as the final authority and requires schema consistency with the evaluation prototype. Mode B relies on self-built tests including a custom golden, case table, gen_data, and run.sh, with schema self-check via torch.ops.

How are expect_error test cases judged in functional acceptance?▼

Expect_error cases pass when an error actually occurs and are not counted as numeric failures. They are excluded from the precision failure tally while still requiring full execution.

What happens if self-built tolerance disagrees with the authoritative precision standard?▼

The authoritative standard prevails and the result is judged non-compliant. The divergence must be noted in the report, and self-built loose tolerances serve only as cross-checks, never as pass evidence.

Can QA fix operator code during CP3 acceptance?▼

No. QA only performs acceptance and problem localization such as reproduction and root-cause evidence gathering. Code changes and upstream design modifications are explicitly out of scope.

Related Skills

Explore More Software Engineering Skills→
hw-native-syshw-native-sys
11

ascendc-st-design

Design Ascend C operator tests via YAML-driven test plan and test-case generation.

Official
Advanced
liuminxin45liuminxin45
95

speckit-ai-self-acceptance

Evaluate AI-driven code changes against acceptance rubrics and report PASS, FAIL, or BLOCKED outcomes.

Community
Advanced
johnhoutjohnhout
6

officiele-bronverificatie

Validate Dutch tax data against belastingdienst.nl and wetten.overheid.nl.

Community
Intermediate
fockusfockus
2

sadd:judge-with-debate

Coordinate three judge agents in iterative debates to produce consensus evaluations.

Community
Advanced