# Review Op Test Coverage

> Review a CV-CUDA operator's test coverage, including C++ correctness, required cross-layout parity, correctness rigor, and the Python API surface. Use when asked whether an operator is adequately tested or to find and fill test gaps.

- Skill: `cvcuda/review-op-test-coverage` (Agent Skill)
- Install (CLI): `npx skillmds@latest add cvcuda/review-op-test-coverage`
- Raw SKILL.md: https://api.skillmd.com/api/skills/cvcuda/review-op-test-coverage/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: cvcuda (https://skillmd.com/u/cvcuda)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/cvcuda/review-op-test-coverage

---


[//]: # "SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved."
[//]: # "SPDX-License-Identifier: Apache-2.0"

# Review Op — test coverage

Thin entry point. Run `python3 tools/review_op.py <Operator> --domain test` and interpret
per the **test** section (TST-*) of `.agents/guidance/REVIEW_OP_GUIDELINES.md`: independent gold,
required cross-layout parity, axis-coverage mirror, negative tests, tolerance discipline (bit-exact by
default), and the Python API surface. Resolve `MANUAL` items at the cited locations. When
fixing (`--fix`), author the missing positive, negative, or layout-parity cases - **never introduce
or widen a tolerance silently** (emit rationale + needs-human) — and re-run to confirm.
Findings-first.