Claude Skill
SJTU-IPADS/SkVM
SkVM is a language virtual machine for executing agent skills, built by SJTU-IPADS. It provides a secure, efficient runtime for skill-based agent programs using TypeScript.
Overview
Repository
Install this Skill
git clone https://github.com/SJTU-IPADS/SkVM.gitRegistry
Summary
SkVM is a language virtual machine designed specifically for executing agent skills. Developed by SJTU-IPADS, it provides a runtime environment that enables agents to run skill-based programs efficiently and securely.
面向代理技能的语言虚拟机
Key features
- Dedicated virtual machine for agent skill execution
- TypeScript-based implementation for modern development
- Lightweight runtime optimized for agent workflows
- Secure isolation for skill programs
Use cases
- Running agent skills in a controlled environment
- Building modular agent systems with skill isolation
- Prototyping new agent skill languages
- Educational projects on virtual machine design
README excerpt
<div align="center"> # SkVM **Compile and run LLM agent skills across heterogeneous models and harnesses** [English](./README.md) | [中文](./README.zh-CN.md) [Website](https://skillvm.ai) · [GitHub](https://github.com/SJTU-IPADS/SkVM) · [Paper](https://arxiv.org/abs/2604.03088) [](https://www.npmjs.com/package/@ipads-skvm/skvm) [](./LICENSE) [](https://github.com/SJTU-IPADS/SkVM/releases) [](https://github.com/SJTU-IPADS/SkVM/commits) </div> SkVM is a compilation and runtime system that makes LLM agent skills portable across heterogeneous models and harnesses. It has four major parts: - **Profiling** — measure a model+harness against pre-defined primitive capabilities - **AOT-Compilation** — compile a skill with multiple passes in AOT compiler - **JIT-Optimization** — improve runtime speed (JIT-boost) and skill content (JIT-optimize) - **Benchmark** — evaluate original, compiled, and optimized skills across tasks, conditions, and models <p align="center"> <img src="./docs/skvm_arch.png" alt="SkVM architecture" width="66%" /> </p> Reference: **SkVM: Revisiting Language VM for Skills across Heterogenous LLMs and Harnesses** — https://arxiv.org/abs/2604.03088 ## News - **2026-05-18 — Headless pi driver, on by default.** `jit-optimize` and `jit-boost` now run their internal coding agent via [`@mariozechner/pi-coding-agent`](https://www.npmjs.com/package/@mariozechner/pi-coding-agent) in-process by default. Opencode remains a first-class peer — set `headlessAgent.driver: "opencode"` in `s
Topics
No topics yet.