What I build我造过什么

My AI skills我的 AI 产品

Open-source AI products I designed and shipped. I don't just analyze AI — I build it, which is how I read it from the inside, not just the outside.我设计并发布的开源 AI 产品。我不只分析 AI——我亲手造它;正因为造过,我能从内部读懂它,而不只从外部观察。

/analyst-research
Hypothesis → publication-grade report假设 → 出版级报告

Three scope modes over one rigorous 8-step workflow. Source provenance, three-state labeling and human checkpoints — encoded as procedures an LLM runs end to end, and proven on a real deliverable before release.三档 scope,一套严谨的 8 步工作流。来源可追溯、三态标注、人工节点——编码成 LLM 能端到端执行的流程,并在发布前用真实交付物验证过。

Open the workflow查看工作流
/verifying
Every claim back to a primary source每个论断回到一手来源

Fact-checks any statement against whitelisted primary sources only — catching out-of-context quotes, running one-level decomposition (Z = P × Q), and laying conflicting sources side by side.只拿白名单原始来源核对任何说法——抓断章取义、做一层拆解(Z = P × Q)、把多源冲突并排呈现。

Open the workflow查看工作流
/topic-brief
Public signal → a shareable briefing公开信号 → 可分享的简报

Turns public news into a focused topic briefing — a self-contained HTML page, styled for the browser and one-click paste into WeChat.把公开新闻做成聚焦的主题简报——独立 HTML 页面,适合浏览器看,也能一键贴进公众号。

Open the workflow查看工作流
/local-vault
Your files → a queryable Markdown vault你的文件 → 可检索的 Markdown 库

Two functions in one skill: convert PDFs, Office docs, images and code into clean Markdown with retrieval-friendly frontmatter (local-first, cloud OCR only as fallback) — then answer questions over the vault with coverage self-checks and Maps-of-Content.一个 skill 两件事:把 PDF、Office 文档、图片和代码转成带检索友好 frontmatter 的干净 Markdown(本地优先,云端 OCR 仅兜底)——再以覆盖自检和内容地图(MOC)在资料库上作答。

Open the workflow查看工作流

Building these is how I read AI from the inside — not just the outside.正因为造过这些,我能从内部读懂 AI,而不只从外部观察。

All open-source on GitHub · MIT全部开源在 GitHub · MIT