Prompts and reasoning checks
One recent project is Codex Agents, for refining prompts and examining reasoning. Colleagues and friends use it repeatedly too. Its uses and limitations are documented in the case study.
About mantou
I turn recurring problems in my own life into tools. I am also learning to understand other people’s needs and improve my projects through real use.
I make serious plans. I also write things like “Saturday. Tired.” Both belong here.
Read the Agent usage and validation notes →Edited English version based on the Chinese original. Read the Chinese original
A few habits can grow into more possibilities.
I imagine the good habits I keep practising as nearby starting points. They extend and intertwine, gradually developing into new abilities and possibilities.
Explore my projects →Writing and projects
My diary keeps thoughts as they happened. Project pages follow what I made and learned. Some problems are solved; others are putting up a good fight.
Four principles I currently work by
- Start with a real problemA project should address something I or an actual user has encountered. I do not invent projects just to fill a portfolio.
- Define success before startingWrite down what I want to establish, how I will check it, and which results would count as only partial success.
- Keep failures and contrary evidenceRecord failures, fixes, and unresolved issues alongside the successes.
- Finish, review, then iterateComplete the checks and review for each stage, then use the results of real use to decide whether to continue.
Things we can talk about
These are areas I am currently exploring. They are starting points for a conversation.
Personal experiments
Turn questions about attention, health, learning, and long-term skills into experiments I can record, test, and improve.
Small tools and PWAs
Turn repetitive personal routines into mobile tools that work and can be tested. A published PWA project is available to try.
AI-assisted workflows
Explore practical collaboration steps, acceptance checks, and low-friction ways to use AI for specific tasks.
Writing and project documentation
Turn scattered practice into public case studies that explain the problem, evidence, limitations, and lessons.
My public projects show my experience so far.
I distinguish observed results from ideas and future plans. I do not publish account returns, balances, or positions. Individual investments are discussed only at a review point or after I have fully exited.
Have a specific problem? Let’s talk.
If you run into problems using these tools, or have a specific task you would like to discuss, get in touch. You do not need a complete plan: tell me what you want to do, where you are stuck, and what you have tried.