I run marketing. I build the software I need to run it, by talking to agents.
Not an engineer. Everything below started as a problem in my own week — messages piling up, work scattered across tools, a version launch that needed shipping — and ended as something small enough to give away.
Playbook
-
2026-08-29
The Agent Playbook
One task system per project, machine-verifiable acceptance, and rules agents cannot rewrite for themselves — distilled from four real projects.
Open source9 repositories
-
2026-08-24
wps365-cli-tips
Field notes on the WPS 365 CLI: device-code login, reading smart docs straight from share links.
-
2026-08-14
mailroom
An AI mailroom for your work messages: fetch them, file each one into the right project or task, and reply under a three-tier permission gate. Node built-ins only, no dependencies.
-
2026-08-13
notion-workspace-blueprint
An open architecture blueprint for enterprise workspaces, autonomous agents, developer tools, and personal productivity in Notion.
-
2026-08-10
hap-skills
Agent skills for Mingdao HAP — the enterprise no-code platform my team builds.
-
2026-08-04
ghostty-theme-picker
A live theme picker for Ghostty. Arrow through themes on your real screen, apply to every open window at once, sorted by background brightness.
-
2026-07-17
workmd
Run all of your work on plain markdown and git. One folder per project, one folder per task. Humans read it, agents write it. No database, no API, no lock-in.
-
2026-05-24
agy-tips
Tips and automation scripts for the Google Antigravity CLI (agy).
-
2026-04-02
claude-code-tips
A collection of practical Claude Code tips.
-
2026-03-28
articles
Experiences, perspectives, and frameworks on AI, technology, and business.
What I do
I lead marketing at Mingdao, an enterprise no-code platform, and at Nocoly, its international product. Positioning, content, campaigns, events, and the systems underneath them.
The tools here exist because that job kept hitting the same wall: the work is spread across a dozen surfaces, and none of them can be read by a machine. So I moved my own work onto plain files, and taught agents to operate it. The results are small, opinionated, and built for one person — mine. They might fit yours.