MindSpore Skills

MindSpore development skills for AI coding agents. Build CPU/GPU/NPU operators and migrate models with guided workflows.

Compatible with Claude Code, OpenCode, Gemini CLI, and Codex.

Installation

Claude Code

Register the marketplace and install:

/plugin marketplace add vigo999/mindspore-skills
/plugin install mscode@mindspore-skills

Then use slash command:

/mscode:cpu-plugin-builder
/mscode:cpu-native-builder
/mscode:gpu-builder
/mscode:hf-diffusers-migrate
/mscode:hf-migrate
/mscode:hf-transformers-migrate
/mscode:hf-transformers-migrate-test
/mscode:migrate
/mscode:model-migrate
/mscode:npu-builder
/mscode:mindspore-aclnn-operator-devflow
/mscode:api-helper

OpenCode

Clone to your global config:

git clone https://github.com/vigo999/mindspore-skills.git ~/.config/opencode/mindspore-skills
ln -s ~/.config/opencode/mindspore-skills/skills ~/.config/opencode/skills
ln -s ~/.config/opencode/mindspore-skills/commands ~/.config/opencode/commands

Or for a specific project:

git clone https://github.com/vigo999/mindspore-skills.git .opencode

Then in OpenCode:

/cpu-plugin-builder

See OpenCode Skills docs for more details.

Gemini CLI

Install as an extension:

gemini extensions install https://github.com/vigo999/mindspore-skills.git --consent

Or from local clone:

git clone https://github.com/vigo999/mindspore-skills.git
gemini extensions install ./mindspore-skills --consent

See Gemini CLI extensions docs for more details.

Codex

Clone to your project root:

git clone https://github.com/vigo999/mindspore-skills.git .mindspore-skills

Codex reads AGENTS.md automatically. Verify with:

codex "Summarize the MindSpore skills available."

See Codex AGENTS guide for more details.

Available Skills

Operator Development

Skill	Description
api-helper	Find API call chains and operator wiring in MindSpore codebase
cpu-plugin-builder	Build CPU operators via ATen/libtorch adaptation
cpu-native-builder	Build native CPU kernels with Eigen/SLEEF
gpu-builder	Build GPU operators with CUDA
npu-builder	Build NPU operators for Huawei Ascend
mindspore-aclnn-operator-devflow	End-to-end ACLNN operator adaptation workflow for MindSpore Ascend

Model Migration

Skill	Description
hf-diffusers-migrate	Migrate HF diffusers models to mindone.diffusers
hf-transformers-migrate	Migrate Hugging Face transformers models to mindone.transformers
hf-transformers-migrate-test	Generate minimal MindOne transformer tests for migrated models
model-migrate	Migrate PyTorch repos to MindSpore

Available Commands

Operator Development

Command	Description
/api-helper	API chain discovery workflow
/cpu-plugin-builder	ATen adaptation workflow
/cpu-native-builder	Native kernel workflow
/gpu-builder	CUDA kernel workflow
/npu-builder	Ascend NPU workflow
/mindspore-aclnn-operator-devflow	ACLNN deep development workflow

Model Migration

Command	Description
/migrate	Migration router (HF/third-party), routing only
/hf-migrate	HF library router (diffusers/transformers), routing only
/hf-diffusers-migrate	HF diffusers migration workflow
/hf-transformers-migrate	HF transformers migration workflow
/hf-transformers-migrate-test	HF transformers test generation for migrated models
/model-migrate	PyTorch repo migration workflow

Usage Examples

Build a CPU operator

/cpu-plugin-builder

> Help me implement the linspace operator for MindSpore CPU

Run minimal public example

bash examples/cpu/plugin_add/run.sh

See expected result contract in examples/cpu/plugin_add/expected.md.

Contract and Tests

• Contract docs:
  • docs/concepts/skills-contract.md
  • docs/concepts/artifacts-and-reporting.md
• Cross-skill contract tests: tests/contract/
• Skill-specific tests: skills/<skill>/tests/

Repository Structure

mindspore-skills/
├── .claude-plugin/          # Claude Code plugin config
├── commands/                # Slash commands
│   ├── api-helper.md        # API chain discovery
│   ├── migrate.md           # Migration router
│   ├── hf-migrate.md        # HF library router
│   └── ...
├── skills/                  # Skill definitions
│   ├── cpu-plugin-builder/  # ATen/libtorch operators
│   ├── cpu-native-builder/  # Native CPU kernels
│   ├── gpu-builder/         # CUDA operators
│   ├── npu-builder/         # Ascend NPU operators
│   ├── mindspore-aclnn-operator-devflow/ # ACLNN deep development workflow
│   ├── hf-diffusers-migrate/   # HF diffusers migration
│   ├── hf-transformers-migrate/ # HF transformers migration
│   ├── hf-transformers-migrate-test/ # HF transformers test generation after migration
│   └── model-migrate/       # PyTorch repo migration
├── AGENTS.md                # Codex instructions
└── gemini-extension.json    # Gemini CLI config

Contributing

Contributions are welcome. Please submit a pull request.

When adding a new skill:

1. Add skills/<skill-name>/SKILL.md with matching frontmatter and directory name
2. Add a slash command in commands/<command-name>.md if needed
3. Update AGENTS.md (skill table + activation triggers)
4. Update README.md (skill list and commands)
5. Update gemini-extension.json with name/path/description
6. Update .claude-plugin/marketplace.json if skills or metadata changed

When modifying an existing skill:

1. Update skills/<skill-name>/SKILL.md and any referenced files
2. Refresh AGENTS.md triggers if scope/keywords changed
3. Update README.md if descriptions or commands changed
4. Update gemini-extension.json if name/path/description changed

tools:

• Run python tools/check_consistency.py before submit
• (Optional) Install git hooks with python tools/install_git_hooks.py
• Tip: set up hooks with make hooks (see Makefile).

License

Apache 2.0