Hey everyone!
I've been diving deep into using Claude Code with my Claude Max subscription, and while it's a fantastic tool, I've started noticing some recurring issues that really drain my time and energy. After working on a few different codebases, I created a `CLAUDE.md` file to streamline my prompts and tackle these problems. However, I've found that this file started diverging across multiple projects, which got frustrating. So, I decided to put together a central repository for my prompt that I thought might help others as well.
Here are some problems I'm hoping to address with my prompt:
- Claude Code sometimes generates overly long files, which is generally bad practice and makes it harder for AI tools to function effectively. If you've experienced this, you know how things can slow down.
- I've noticed that even when I specify not to, Claude can create 'dummy' implementations. My prompt works to prevent this.
- Another issue is that Claude often uses incorrect syntax and instead of fixing it, it suggests switching libraries or dummy implementations, which my prompt aims to discourage.
- For larger tasks, the possibilities for misunderstanding increase. My prompt helps Claude to zoom in on specific tasks instead of taking on too broad ones.
- Lastly, I find that Claude often jumps into coding without fully understanding the context first. My prompt instructs it to analyze the codebase before writing any code.
The good news is that this prompt should work well with other AI tools too!
I'm curious if anyone else has similar prompts or ideas to improve upon this one!
7 Answers
Great initiative! One thing I've learned is to double-check that any libraries you use are up-to-date since Claude can sometimes rely on outdated syntax. If you're not sure, using resources like Perplexity or a quick web search can really help clarify newer approaches. Also, have you considered integrating Context7’s MCP with Claude Code? It could enhance your prompts even more!
I've had similar issues where Claude forgets instructions when the context gets too large. It would be super convenient if it could be programmed to revisit your `claude.md` file after a few prompts. What do you think?
That's a good point! A workaround could be to use a custom command that references your instructions. It might not be perfect, but it's worth exploring.
Thanks for sharing your prompts! I'm excited to test them and see if it really enhances my coding experience. Already looking forward to the improvements!
I think your idea is solid! You could extend it by implementing a custom command with $ARGUMENTS. For instance, using a command like /project:d could streamline the process by automatically linking to your tasks. Have you tried that yet?
I'm curious how that works! Does it really help reduce repeated prompts?
Thanks for this! Your approach seems really smart and could save a lot of headaches in the long run!
Awesome prompts! By making Claude read entire files, I can see how you might burn through tokens, but I guess it saves time later when avoiding code duplication. Definitely a trade-off worth considering!
Can’t wait to try your setup! Let me know how it works out for others, I'm really interested in seeing those improvements.
I'll keep you posted! Fingers crossed for better results.
That sounds interesting! I'll look into it, thanks for the suggestion.