Open-Source Liang Wenfeng AI Skill: How to Build and Replicate for Any Expert

I converted the PDF and nearly four‑hour audio of Liang Wenfeng’s investment conference into a functional AI Skill, and I am pleased with the result.

The Skill can answer questions that were not addressed in the original session, such as which domestic foundation‑model companies are likely to survive, the criteria that determine long‑term competitiveness, and what advice a newcomer should follow in today’s AI‑driven workplace.

The source material covers DeepSeek’s open‑source strategy, commercialization plans, AGI roadmap, agent architecture, continuous learning, domestic compute resources, organizational management, and the overall future landscape of China’s large‑model industry.

The implementation follows a Retrieval‑Augmented Generation (RAG) approach, making it more complex than a typical Skill. The repository (github.com/tjxj/z-skills/tree/main/z-liang-wenfeng-grounded-voice) provides the full code and data.

Key files and their purposes: SKILL.md: Core behavior rules that define trigger conditions, default mode, answer boundaries, and output format. references/voice-guide.md: Defines the answer rhythm – first give a judgment, then explain the mechanism, retain uncertainty, and finally refer to long‑term variables. references/transcript-by-page.md: Full transcript organized page by page. references/topic-index.md: Indexes the multi‑thousand‑word material by topic so the agent can locate relevant sections before retrieving the original text. examples/golden-answers.md: Sample target answers. tests/test-cases.md: Behaviour tests that specify how the model’s answer should look.

Installation command:

npx skills add tjxj/z-skills --skill z-liang-wenfeng-grounded-voice

The directory layout is straightforward, and each file solves a distinct problem in the Skill pipeline.

This design can be directly replicated: replace the files in references/ with any other person’s interview, speech, or public remarks, keep the three‑layer answer mechanism in SKILL.md and the reasoning path in voice-guide.md, and you obtain a "XXX Skill" for figures such as Elon Musk, Zhang Yiming, Duan Yongping, or any expert you wish to query.