commit-as-prompt

g-claude

A Claude Code skills marketplace for Ascend NPU inference, model quantization, LLM evaluation, and structured Git commits.

Install as a marketplace

claude plugin marketplace add starmountain1997/g-claude

Then install individual skills:

claude plugin install ascend@g-claude
claude plugin install vllm-ascend@g-claude
claude plugin install msmodelslim@g-claude
claude plugin install aisbench@g-claude
claude plugin install commit-as-prompt@g-claude

Or install everything at once (also installs karpathy-skills and skill-creator):

curl -fsSL https://raw.githubusercontent.com/starmountain1997/g-claude/main/scripts/install-g-claude.py | python3 - --update

Pass any script arguments after python3 -, e.g. --update, --help, etc.

Install via OpenPackage

OpenPackage is a universal package manager for coding agent configs. Install all skills at once:

npm install -g opkg
opkg install gh@starmountain1997/g-claude

Or install individual skills:

opkg install gh@starmountain1997/g-claude --skills ascend
opkg install gh@starmountain1997/g-claude --skills vllm-ascend
opkg install gh@starmountain1997/g-claude --skills msmodelslim
opkg install gh@starmountain1997/g-claude --skills aisbench
opkg install gh@starmountain1997/g-claude --skills commit-as-prompt

Skills

Skill	Description
ascend	Ascend NPU hardware entry point — health check, environment setup, shell script template. Starting point for any Ascend workflow.
vllm-ascend	vLLM-Ascend serving toolchain — install, offline validation, scenario tuning, online serving, contribution guide.
msmodelslim	Model quantization on Ascend NPUs — W4A8/W8A8/W4A4, one-click and custom YAML, MoE mixed precision, VLM support, accuracy recovery.
aisbench	AISBench evaluation framework — accuracy benchmarks (GSM8K, MMLU, AIME) and performance benchmarks against vLLM services.
commit-as-prompt	Structured Git commits with WHAT/WHY/HOW format, optimized as AI context for future sessions.

Workflow

The Ascend skills form a pipeline:

ascend (NPU check)
  ├──► vllm-ascend (install → serve)
  ├──► msmodelslim (quantize → serve via vllm)
  └──► aisbench   (evaluate accuracy & performance)

License

MIT