Problem
AI Agent Skills are often iterated by trying them manually. A model change, an agent change, or a prompt edit can shift behavior, but teams rarely have a stable way to evaluate, compare, and regress those changes.
Skill Up addresses that gap: it makes Skills testable as engineering artifacts.
Project Ownership
Skill Up is publicly released by Alibaba Group, carries a Copyright 2026 Alibaba Group notice, and is open-sourced under the Apache License 2.0.
My Contribution
While employed at Alibaba, I was responsible for the project's overall design and the development of its core evaluation capabilities, and I contributed to its maintenance.
System
- Multi-engine support for environments such as Claude Code, Codex, and Qodercli.
- Reusable evaluation cases that can be replayed across agents.
- Layered assertions through expect + judge checks.
- CI-friendly workflows for automated regression.
- Compatibility with Anthropic skill-creator format.
Why It Matters
AI engineering is not only about connecting a model to an application. It is about building workflows that can continuously judge quality. Skill Up is public work I helped build while employed at Alibaba, and it reflects my practice in engineering Agent Skill evaluation.
Link
Notice
This page describes my publicly disclosed contributions during my employment. It does not represent the views of, or imply endorsement by, Alibaba Group. Rights in the project name, code, and related marks remain with their respective owners.