Regional economic analysis for 10 Chinese cities. Four scenarios: compare, site selection, sales forecast, government policy. 70+ REST endpoints, ECharts 5.6, FastAPI + zero-build SPA. Built to answer: where to build the next factory?

Hand-curated index of 900+ agent skills (repos, cookbooks, MCP servers, case studies) across 49 categories. One YAML file, bilingual, no library, no tutorial site — just a list you can grep and vendor.

Small library of CS + economics papers I've read or want to read. 503 entries, DOIs verified, VitePress site so I can find them again. No PDFs bundled, no analytics, no comments.

AI configuration hub for dev workflows — the master planner for my other AI tooling repos. Not a project to develop; a tool to point your AI at. Shell + YAML, that's the whole stack.

Folder-as-Agent for thesis writing. The folder itself is the agent — feed it to Doubao, Claude, GPT, or Gemini, and it runs the workflow. Compatible with any LLM that reads files.

Windows desktop toolkit for power users and sysadmins. Tauri 2 + Vue 3.5, wt-service runs as SYSTEM for the high-privilege ops (services, registry, hosts, scheduled tasks). 35-command palette to skip the menus.

Multi-agent data assistant built on CrewAI. Two endpoints: /api/analyze (researcher → analyst → writer) and /api/knowledge/query (single agent). FastAPI backend + a CLI that hits the same code.

MIRU INDEX — a curated index of manga, visual novel, and ACG resources on GitHub. Read it, click through, use the upstream. Nothing hosted here, nothing redistributed.

Statistical + ML decoder for multiple-choice option distribution patterns. CET-4 / CET-6 first, more exams later. Single-page SPA, real data, no hand-waving.

学术资源聚合与检索 — 我的论文/教材/课程笔记目录。还在搭骨架, 但骨架已经比上礼拜多了。Next.js + SQLite FTS5, 先本地后服务。

My fork of Nous Research's hermes-agent. Pulled in to study the TUI / supply-chain / MCP patterns. I don't ship fixes upstream from this fork — issues belong to the original repo.

CET-6 option research, decoded. Real hit rate 27–30% vs random 25% with the full code, the data, and the audit. Yes, the folklore patterns check out.

My GitHub profile README. Auto-rebuilt every day by a Python script and a GitHub Actions cron. Astro is too heavy for this; stdlib is enough.