alibabaOfficial·12 Agent Skills Included

MNN

Optimize and extend the MNN on-device AI inference engine

Guides AI agents through real engineering work on the MNN deep learning inference engine for mobile and edge devices. Covers adding new operators, supporting new LLM models, optimizing ARM/RISC-V/Metal/OpenCL/Vulkan/Hexagon backends, debugging correctness bugs, and running device test suites. Eliminates guesswork with step-by-step tested workflows, measurement discipline rules, and verified troubleshooting playbooks.

16,007|2,426

↗

npx skills add alibaba/MNN --all -g -y

Available:

Tells the AI agent how MNN is architected, which coding and build rules to follow, and how to route each task to the correct skill file before executing it step by step.

All Skills in This Repository (12)

Pure Emerald Level Indicators

📦 In Repo

alibaba

add-new-op

Guide adding new operators to the MNN framework with TDD steps.

support-new-llm

Add new LLM models to MNN by downloading, analyzing, mapping, and exporting.

arm-cpu-optimize

Optimize MNN operators for ARM CPUs with multi-threading and assembly kernels.

metal-optimize

Guides development and optimization of MNN Metal backend kernels for LLM inference.

retrospective

Distills lessons from completed tasks into updates for skill documentation.

test-ci

Runs MNN regression tests and on-device LLM benchmarks on Android and iOS.

hexagon-optimization

Optimizes and validates MNN Hexagon DSP backend kernels with device-measured profiling.

riscv-cpu-optimize

Optimizes MNN RISC-V CPU kernels and LLM inference performance on RVV and vendor matrix extensions.

bugfix

Diagnose correctness and regression bugs in the MNN inference engine by symptom category.

vulkan-optimize

Optimizes MNN Vulkan backend kernels and integrates new Vulkan features on mobile GPUs.

qnn-debug

Diagnose and fix MNN QNN backend errors, accuracy issues, and operator support on Qualcomm HTP.

opencl-optimize

Optimizes MNN OpenCL backend kernels through benchmarking, iterative tuning, and on-device validation.

Official

Advanced

Frequently Asked Questions

FAQPage Schema

How to install MNN skills?▼

Run `npx skills add alibaba/MNN --all -g -y` in your terminal to install all agent skills in this suite globally.

What is MNN used for?▼

MNN is a lightweight deep learning inference engine that runs neural networks, LLMs, and diffusion models efficiently on phones, PCs, and embedded devices.

How to add a new LLM model to MNN?▼

The support-new-llm skill walks the agent through downloading the model, mapping its architecture, aligning outputs with HuggingFace, and exporting a working MNN model with tests at each step.

Can AI optimize GPU kernels in MNN?▼

Yes. Dedicated skills guide the agent through Metal, OpenCL, Vulkan, and Hexagon kernel optimization with profiling, correctness checks, and real-device benchmarking.

Do these skills work with Claude Code and other agents?▼

Yes. All skills follow the standard SKILL.md format and run in Claude Code, Cursor, Codex, and other compatible AI coding agents.