How I work with Claude Code
The setup behind most of what you read on this site: Claude Code on my Mac, one folder per project, and a few habits that keep it useful.
The setup
Most of the projects on this site were built with Claude Code, Anthropic’s coding assistant, running in the desktop app on my Mac. Every software project has its own folder, and a session is a conversation anchored to that folder: Claude reads the files, runs commands, edits code and documents, and I direct it and review what it does. It is closer to working with a colleague who types very fast than to asking a chatbot for snippets.
A typical day
The requests are short and repeat all the time, and that is the point:
- “update the X page”: Claude reads the notes it kept about project X and the local git history, then updates the page of that project on this very site, in Italian and English.
- “commit and push”: stage the right files, write the commit message, publish. Only when I ask.
- “update the website memory”: log what changed, so the next session does not start from zero.
- For a website, Claude also opens the page in a built-in browser and checks that it really works before telling me it does.
What I learned
- It is fast, and it assumes. Most of my corrections are not about code but about facts only I know: which device is wired, which share exists, what depends on what. Hence the habits in The rules I set myself.
- Continuity is the real feature. A tool that forgets everything each morning is a tool I have to re-brief each morning. See Memory.
- Small, repeatable rituals beat clever prompts. Skills and hooks exist to make exactly those rituals reliable.
In this section
Memory, Skills and plugins, The advisor, Hooks and The rules I set myself: how each piece is set up on my machine, including what I have not done yet.