How I get AI agents to write clearer reports
Borrowing from ASD-STE100, I built ste-zh, a skill that makes AI task reports in Chinese more accurate and easier to read.
Recent posts
Why I use a cheaper model for code review
Data from more than 2,000 tasks: why a top-tier model for execution and a cheaper model for review gives better value, and what that depends on.
Projects
Kander
Parallel development with multiple agents, driven by strict rules. Independent review and delivery gates are built in to protect the quality of automated delivery. From August to October 2026, it completed more than 2,800 tasks for me.
QuickTUI
Control the coding agents on your computer from your phone. Smooth terminal and chat modes let you monitor and take over the agents on a remote computer at any time. It supports more than 10 agents, including Claude Code, Codex and DeepSeek Harness, and connects through an end-to-end encrypted relay.
STE-zh (in Chinese)
An agent skill based on ASD-STE100 (Simplified Technical English), a strict writing standard used in aerospace and other industries. It makes agents report their work results in clear, precise and unambiguous Chinese.
An indie game
I’m building an indie game with the help of AI agents. I hope to share it with you soon.