The Milvus Project

Five skills are available: research (vendor repos as submodules and write architecture memos), spec (produce PRDs, component designs, and phased implementation plans), impl (build one plan phase end-to-end with independent code review), agent-eval (build LLM/agent evaluations with golden sets and layered graders), and human-writing (Chinese long-form articles, stories, and speeches).

Software engineers and technical leads doing spec-driven development benefit from research, spec, and impl. ML engineers shipping LLM or agent features use agent-eval for quality gates and regression testing. Chinese-language content creators writing Zhihu answers, blog posts, fiction, or speeches use human-writing.

Research vendors prior-art repos under ./vendors and writes memos to ./docs/research. Spec turns requirements into a dependency-ordered spec set under ./specs cross-referenced with those memos. Impl then builds one phase of the plan end-to-end and runs an independent code review against the polished specs before declaring done.

Yes. Milvus is an open-source vector database designed for AI applications, and the organization publishes its repositories publicly on GitHub under the milvus-io account. The skills are distributed as part of this open-source project at no cost.

The research skill expects git submodule support and writable ./vendors and ./docs/research directories. Spec and impl assume a project workspace with ./specs and an implementation plan or roadmap. Agent-eval presumes an existing LLM or agent system whose prompts, models, or configurations you want to measure and compare.