optimize-op-verify

Verify a CV-CUDA optimization campaign's deterministic definition-of-done and concise versioned MR summary per .agents/guidance/OPTIMIZATION_GUIDELINES.md. Use to gate whether a perf campaign or performance MR is ready, including reference-SKU statistics, hard checklist evidence, baseline validation, lead exhaustion, memory checks, and API/ABI stability.

Jump to install Review files

Repository: CVCUDA/CV-CUDA
Last source activity: August 11, 2026 at 21:59
Detected SKILL.md language: English
Stars: 2,727
Forks: 264

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Install with an AI assistant

Copy this prompt into the AI assistant you're using.

Follow https://skillsmp.com/skill-install/prompt.md to install only the Agent Skill at https://github.com/CVCUDA/CV-CUDA/tree/main/.agents/skills/optimize-op-verify.

A direct command skips the review prompt. Inspect the source before running it.

npx skills add https://github.com/CVCUDA/CV-CUDA --skill optimize-op-verify

The command stays on one line. Scroll horizontally to inspect it before copying.

Prefer a local copy? Download the files currently available to SkillsMP.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

SKILL.md

Source instructions · Read-only preview

name: optimize-op-verify
description: Verify a CV-CUDA optimization campaign's deterministic definition-of-done and concise versioned MR summary per .agents/guidance/OPTIMIZATION_GUIDELINES.md. Use to gate whether a perf campaign or performance MR is ready, including reference-SKU statistics, hard checklist evidence, baseline validation, lead exhaustion, memory checks, and API/ABI stability.

[//]: # "SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved." [//]: # "SPDX-License-Identifier: Apache-2.0" # Optimize Op — verify Read `.agents/guidance/OPTIMIZATION_GUIDELINES.md`, then refresh the v1 summary and run the done-gate: ```bash python3 tools/optimize_op.py <Op> --phase summary \ --results <description> --state final --out <description> python3 tools/optimize_op.py <Op> --phase evidence \ --base <ref> --results <description> ``` Report findings first. Require the exact full-union key set expanded from every declared `basic` and `advanced` entry, before/after C++ timing for every key on every configured reference SKU, exact optimized/full counts, verified min/median/max statistics, verified layout ratios or verified n/a rows, and all six checklist items checked with hard evidence. Require the canonical companion `## Memory footprint` section outside the bounded v1 block, and treat ODO-9 as the memory-footprint authority; its `MANUAL` result requires explicit human review before submission. Declare the campaign done only after a rerun shows zero `GAP` and zero unresolved `MANUAL`. Never weaken tests, fabricate baselines, check unsupported evidence, weaken the ODO-9 memory policy, or change public API/ABI.

View on GitHub

Back to search Back to installation

make-op

Add a new CV-CUDA operator end-to-end per .agents/guidance/MAKE_OP_GUIDELINES.md, with a deterministically-enforced definition-of-done. Use when asked to create/add a new operator, scaffold one, or verify that a new operator is complete (approved spec, wired scaffold, bit-exact regression tests across the declared support matrix, required layout parity, complement negatives, docs + relnote, benched dtypes, tests that run and pass, optimization-ready).

2026-09-16

refactor-op

Find and safely apply per-operator refactoring / redundancy-reduction opportunities in a CV-CUDA operator (near-duplicate Tensor/VarShape kernels, reinvented shared utilities, dead code). Use when asked to reduce code duplication, de-duplicate or unify an operator's kernels/bindings, remove dead code, or verify a refactor changed nothing observable. Produces a deterministic findings-first report; the apply path is gated by a strict bit-exact / feature-set / test-coverage parity check.

2026-09-16

make-op-scaffold

Scaffold a new CV-CUDA operator — a complete, wired, building skeleton — and delegate the implementation to a human or another AI. Use when asked to set up / stub out a new operator without implementing it, or to bootstrap one for someone else to finish. Supports --bare (skip the spec when the definition itself is delegated).

2026-08-11

View all from this repository

optimize-op-verify

Source facts

Install options

Install with an AI assistant

Review the source files

More from this repository

Install with an AI assistant

More from this repository