Author: Wang Jianshuo
Simply record the experiences with Claude Code up to now, purely personal exploration, and may not be applicable to everyone.
1. Focus on one tool and use it intensively. I use Claude Code. I don't think it's better than Codex, but the ROI of comparing tools may not be high, although it can create a false sense of achievement by articulating differences.
2. Remember the most important shortcuts. Control+G to open the editor, which helps in writing longer content; Control+A, Control+E, Control+U are very practical shortcuts for quickly moving the cursor in the command line. Although they are not new in 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, first write PROJECT.md, using a structured approach to write everything down at once.
5. Claude agents are the default opening method.
6. Claude Code and github.com and cloudflare.com are a perfect match, delegating all operations related to the building process, publishing process, and domain name to the infrastructure.
7. Separate human-written content from machine-written content. Manually maintain the core CLAUDE.md and avoid reading .md or code written by Claude Code. Machines are for machines, humans are for humans. Understand AI-written content by asking AI, not by looking at the source code.
8. Drag files into the Claude Code window—audio, video, documents, screenshots—if it's unclear, use Command+Shift+5 to take a screenshot and then drag it over, it's the fastest.
9. Restructure the memory system. Center around ~/.claude/CLAUDE.md, categorize and reference multiple memory files, requiring not to use the project's memory, and place all memory files in git, syncing to GitHub (private), so that your memories are permanent and accumulative, rather than scattered across every project.
10. Write Skills, and after each work session, require Claude to "consolidate what has been learned into Skills"—it can be done automatically.
11. When possible, use ultracode to trigger dynamic workflows for complex tasks. Although it's expensive and slow, the results are still guaranteed.
12. Accumulate skills as you go, and continuously restructure skills. Skills need to be stored in git.
13. Use the documentation of git as the output of the previous task and also the input for the next task. This creates clear handover documents between agents, without relying on context for transitions.
14. Treat Claude Code as a horse (or a person), not as a vehicle. The vehicle turns under your command, but the horse has its own thoughts; we just need to set goals and boundaries. Its autonomous pathfinding feature is a characteristic, not a bug.
Does anyone have any additions?
Disclaimer: This article represents only the personal views of the author and does not represent the position and views of this platform. This article is for information sharing only and does not constitute any investment advice to anyone. Any disputes between users and authors are unrelated to this platform. If the articles or images on the webpage involve infringement, please provide relevant proof of rights and identity documents and send an email to support@aicoin.com. The relevant staff of this platform will conduct an investigation.