Hello, I'm the captain.
It seems the last post received more interest than expected from many people, so I'm writing an updated follow-up post after a few days.
I still have the article pre-written on my personal blog.
https://ghost.redsub.net/labor-in-the-ai-era/
Different content to fit our bulletin board. - Some elements match, but the context of the blog post and the post on our bulletin board differ somewhat. - For our bulletin board, I'm adding tips on my acting. For the blog, I'm writing it in an 'essay' style.
There have been many changes to Claude Code over the past few weeks.
The default context size became 1M - wow, it's really convenient now. I almost never see auto-compact anymore - auto memory location became 'customizable', and features were enhanced. A feature called dream - to summarize in one line, it's a memory organization feature - was also added.
The official Claude Code plugins have grown to over 100, and it's changing day by day.
I wrote this much and many paragraphs, but I accidentally pressed Command+W and everything was deleted...........
Ah... From now on, I'll write it in notepad first and then paste it. At least the auto-save worked so the text above survived..........
Anyway. I wrote quite a lot, but the key points are these.
Because I couldn't trust context7 MCP, I created a skill called 'claude-code-guide' in Claude Code. (Created by parsing and summarizing the official Claude Code documentation)
And to update that skill, I set up a hook to notify me via system message whenever Claude Code gets a version update 'at the start of each session'.
I created a skill called upgrade-guide that checks the version and differences of the claude-code-guide skill, then checks the release-notes between versions to update claude-code-guide, and comprehensively reviews user-level (global) hooks, rules, and skills to refactor areas where updates can be applied, and also conducts a comprehensive review of hooks, rules, and skills at each sub-project level to refactor them.
And when 'new' official Claude Code plugins are added or 'existing plugins' are retired, I've created hooks and skills to automate the process of comparing them with the plugins I'm using and deciding whether to 'adopt' them or 'forget' them via memory.
I've set the user-level (global) Claude Code environment (for Mac users, everything inside the ~/.claude folder except 'local-only files and folders') to the main branch of a GitHub private repository, and set up a shell script as a 'hook' to automatically commit and push whenever changes are made. I'm also using this to always keep my development environments - Mac mini or MacBook Pro - in sync. I'm using zsh aliases so that when I run the claude command, it pulls from the branch first and then launches Claude.
In the end, since I work based on Claude Code, I've automated it so that every time Claude Code updates, all my environments are updated accordingly, just like developers receiving a pull from git every morning.
My global CLAUDE.md is only 38 lines. Claude ignores it anyway. However, I still check it regularly to maintain MECE, SSOT, and Evidence before claims. After all, it's a probability game.
Even when I was working as a planner, I did unit tests and integration tests anyway, and even with E2E automation, this life doesn't change. It seems like it might work if I write better scenarios, but the scope I need to cover alone as one person hasn't gotten smaller yet, so it's frustrating.
With a MAX of $100, if I'm 'not greedy', I could live comfortably working 10 hours a day. After organizing permissions at the global and project level and tidying up a few times, checking 2 projects while running them has become a habit, so MAX $200 gives me peace of mind. I don't feel burdened even working on weekends.
I'll run it for a few more weeks and if I feel changes in usage, I'll provide an additional update.
Since I lost the text and had to summarize more, the article above might feel slightly unfriendly, but if you have any questions, please feel free to ask anytime. Even though I may know less than you, I can share what I've 'confirmed' in case it might be helpful.
Then, hoping peace comes to Hormuz.
