AI builder · live commerce · history at play.
我把真实问题拆成 skill、agent 和小工具 — I turn the problems in front of me into Claude Code skills, agent pipelines and small tools.
| 🧭 What I work on | 🛠️ How I build |
|---|---|
|
|
- Shipping think-like-fandeng, a 樊登-style book-talk skill built from ~30 hours of corpus
- Iterating on think-like-wangziru, edition 2 of the 王自如 skill
- Running a Claude × GLM agent team that just audited and re-documented 48 of my repos
Everything I keep public, grouped by theme.
🗣️ Expression-Analysis Agent Skills · 5 repos
Corpus-based Claude Code skills that decompose how specific Chinese speakers structure an argument — each ships a zero-dependency "AI-flavor" detector.
| Repo | What it does | Stack |
|---|---|---|
| think-like-fandeng | 樊登-style book-talk skill distilled from ~30 h of corpus (786 coded units) — analysis, narration, dialogue and revision modes | Markdown skill · Python · CI |
| think-like-wangziru | Edition 2 of the 王自如 skill, built on 32 transcripts (~860k-char corpus base); one line gets you analysis, drafting, dialogue or revision — answer first, reasons clear, survives follow-up questions | Markdown skill · Python · CI |
| wangziru-analysis-skill-v1 | Edition 1 (v1.0–v1.3) — the original, more general-purpose recipe, kept as a stable archive | Markdown skill · Python |
| wangziru-analysis-skill | The original 王自如 repo, frozen at v1.4; development moved to the two repos above | Markdown skill · Python |
| luoyonghao-analysis-skill | Unofficial 罗永浩 expression analysis — logic-first decomposition and writing for speeches, benchmarked on 310k characters of corpus | Markdown skill · Python |
📚 Content & Reading Tools · 2 repos
| Repo | What it does | Stack |
|---|---|---|
| script-to-video | Turns any document into a 3:4 vertical explainer short, rendered entirely by code — deterministic timeline, reproducible frame by frame | Claude skill · Python · Playwright · ffmpeg |
| novel-insight | Local, bring-your-own-LLM analyzer for Chinese web novels — chapter-level mining of character conduct and dialogue tactics into a card knowledge base | FastAPI · Vue 3 · SQLite |
🏯 Historical Simulation · 1 repo
| Repo | What it does | Stack |
|---|---|---|
| courtsim | Court-politics simulation across seven dynasties, Qin through Qing — LLM-driven debates on real agenda items, grounded in a historical-source RAG corpus; fully local, BYO-LLM | Next.js · TypeScript · Prisma |
🎮 Desktop Widgets & Games · 3 repos
| Repo | What it does | Stack |
|---|---|---|
| tetris-widget | Frameless, always-on-top pixel Tetris — full SRS handling, heuristic-AI battle royale, and a boss key that disguises it as VS Code | Electron · JavaScript |
| PixelPal | Pixel-art desktop pet with FSM + behavior-tree AI — 11 species, 90+ job dispatches, a collection book and weather awareness | Electron · TypeScript |
| forced-reminder | A periodic reminder that refuses to be dismissed — modal popups, absolute-timestamp catch-up after reboots, Excel export; Windows 7–11 | Python · PyQt5 |
🌐 Web Tools · 3 repos
| Repo | What it does | Stack |
|---|---|---|
| cs2-lineups | CS2 lineup sandbox — 3D grenade paths over 11 map overviews, offline PWA, plus portal/rental pages and a small self-hosted sync service | JavaScript · three.js · Python |
| cs2-portal | Single-file, zero-dependency CS2 news page fed by the public Steam API, with built-in fallback data | HTML · JavaScript |
| engineer-buddy | Conversational requirements clarification — role-based staged interviews that turn vague ideas into OpenSpec proposal packages; ships a key-hiding proxy with rate limits and spend fuses | React · TypeScript · Vite |
📡 Live-Streaming E-Commerce · 1 repo
| Repo | What it does | Stack |
|---|---|---|
| livestream-coach-oss | Sales-talk analysis agent for livestreams — unattended capture → local CPU transcription → quantified talk structure → gap-finding against category benchmarks; rule layer runs at zero model cost | Python · Playwright · sherpa-onnx |
🏯 History, one simulation at a time · 🤖 AI, one pipeline at a time
