Author: Wang Jianshuo
Simply record my experience with Claude Code up to this point. This is purely personal exploration and may not be suitable for everyone.
1. Focus on mastering one tool intensely. I use Claude Code. I don't necessarily think it's better than Codex, but the ROI of comparing tools may not be high, even though being able to articulate the differences eloquently gives a false sense of accomplishment.
2. Remember the most important shortcuts. Control+G to open the editor, helpful for writing longer content; shortcuts like Control+A, Control+E, Control+U which are very practical for quickly moving the cursor in the command line. Although not new to the AI era, they are as important as Control+C and Control+V when in use.
3. Use voice input. HoldSpeak is very helpful.
4. For a project, start by writing PROJECT.md, using a structured method to jot down all thoughts at once.
5. Claude agents are the default way to start.
6. Claude Code, github.com, and cloudflare.com are a perfect match. Hand over the build process, release process, and all domain-related operations to the infrastructure.
7. Separate what is written by humans and by machines. Manually maintain the core CLAUDE.md; don't read the .md files or code written by Claude Code. Let machines handle machine things, humans handle human things. Understand AI-written content by asking the AI, don't look at the source code.
8. Drag and drop files into the Claude Code window—audio, video, documents, screenshots—if you can't explain it clearly, use Command+Shift+5 to take a screenshot and drag it over, it's the fastest.
9. Reconstruct the memory system. Center it around ~/.claude/CLAUDE.md, categorically referencing multiple memory files. Require not using the project's memory, and keep all memory files in git, synchronized to github (private). This way, your memory becomes permanent and cumulative, not scattered across each project.
10. Write Skills, and at the end of each work session, ask Claude to "precipitate what was learned into Skills"—it can do this automatically.
11. Whenever possible, use ultracode to trigger dynamic workflows for complex tasks. Although expensive and slow, the results are still guaranteed.
12. Accumulate skills and refactor skills along the way. Skills need to be kept in git.
13. Use git documents as the output of the previous task and the input for the next task. Let agents have clear handover documents, not relying on context for transitions.
14. Treat Claude Code as a horse (or a person), not as a car. A car turns under your command; a horse has its own ideas, we just need to set goals and boundaries. Its autonomous pathfinding feature is a characteristic, not a bug.
Does anyone have anything to add?
