Claude God Tip #3: How to Never Get Mad at Claude Again

Published: November 19, 2025 • 6 min read

One great thing about attempting to become a Claude God is that I am more attentive, or would a better word be curious, about ways to make my workflows faster, more efficient, and more effective.

We are now at a stage where we can almost (keyword: almost) ditch traditional methods of programming and bring our ideas to life through prompting. It's easy to get comfortable with taking a process that used to take 5 months down to 2 weeks.

But if you are like me, you should never be content with whatever level of speed or automation you have achieved. You should always be searching, always be curious for better, faster, more optimal ways to take full advantage of automation tools that are now at our disposal.

So I've got a small tip for you today to help you with just that. Therefore, introducing Claude God Tip #3!

Why I Sometimes Get Mad at Claude

Okay, by now you are probably wondering why I titled this post "How to Never Get Mad at Claude Again."

Well, here's the thing: I get mad at Claude sometimes.

I specify instructions within a session that I want it to follow strictly, and sometimes it just forgets them. For instance, when I ask Claude to commit changes, I have a very specific format that I want my commit messages written in, and I want Claude to always follow that format.

However, it tends to forget at times, and then I have to soft reset multiple commits or worse, rebase them if I pushed them to my remote branch already.

Well, there is a very simple way to get Claude to never forget specific instructions again. And if Claude never forgets, I'm not going to be mad at it.

The Solution: The Power of #

All you have to do is type the # symbol in the input area within a running Claude session. Then you type an instruction that you never want Claude to forget during that specific session.

# Always use single line messages when committing changes

That's it. Claude will remember this instruction for the entire session (and beyond, depending on the memory level you choose).

If you have used Claude Projects on the web, this is sort of like adding instructions to a Claude Project.

Useful Memory Instructions

Here are some instructions I think you might find useful to set up for your projects:

Commit-Related

# Always use single line messages when committing changes
# Do not automatically commit a change until you receive approval from me
# When committing, always write the message in past tense (e.g., 'Added' not 'Add', 'Fixed' not 'Fix')
# When committing, group changes by FEATURE, not by files changed

Build-Related

# Always replace ' with the HTML entity &apos; to prevent build errors
# Run the build command after making changes to verify nothing is broken
# Always check for TypeScript errors before committing
# Ensure all imports are used before finalizing changes

Code Style-Related

# Use const instead of let whenever possible
# Always add explicit return types to functions
# Use async/await instead of .then() chains
# Keep functions under 30 lines when possible

Testing-Related

# Run existing tests after making changes to ensure nothing breaks
# Suggest unit tests for new functions you create
# Never mock external dependencies without asking first
# Always test edge cases when writing new tests

Communication-Related

# Explain your reasoning before making significant changes
# Ask for clarification if requirements are ambiguous
# Summarize what you changed after completing a task
# Warn me if a change might have side effects in other files

Memory Levels: Project vs User

When you add a message to memory using #, you get 2 options for what memory level to save that message to:

Project Memory

Stored in the CLAUDE.md file that is automatically created in your current project's directory.

/your-project/CLAUDE.md

Whatever you add to this file will only be applied to that specific project. Use this for project-specific preferences like coding conventions, commit message formats, or build configurations unique to that codebase.

User Memory

Stored in ~/.claude/CLAUDE.md.

~/.claude/CLAUDE.md

Think of this as the memory file for all the projects you ever create using Claude in your terminal. What you add here will be applied to all projects on your computer.

If you've used Claude Projects online, think of User Memory as the Memory feature in Claude web that you can enable or disable. Then think of Project Memory as the instructions you specify for a specific project that you create.

Priority: Project Over User

You should note that whatever is specified in the Project Memory will always be prioritized over User Memory. So if you happen to set conflicting rules in both Project and User Memory, the instruction in the Project Memory will take precedence.

For example:

User Memory: # Always use conventional commit format
Project Memory: # Use simple one-line commit messages

In this case, Claude will use simple one-line commit messages for that specific project.

Editing Your Memory

Now you are probably wondering: what if I change my mind? How do I edit the content of the memory?

Awesome question! There are 2 ways to do this.

Method 1: Direct File Editing

You can open the CLAUDE.md file that was created when you added an item to memory and manually edit it by adding or removing items as you wish.

For Project Memory:

nano CLAUDE.md
# or
code CLAUDE.md

For User Memory:

nano ~/.claude/CLAUDE.md
# or
code ~/.claude/CLAUDE.md

Method 2: Using the /memory Command

You can type the /memory command in the Claude input area, then select which memory you want to edit (Project Memory or User Memory). This will open the memory file in your default text editor where you can make your changes.

/memory

Then choose:

Project memory to edit the current project's CLAUDE.md
User memory to edit your global ~/.claude/CLAUDE.md

What's Next: Adding an Extra Safety Layer

And yeah, that's all for the small tip of today. By doing this, you will never have to get mad at Claude again for forgetting instructions you specified.

Now, for the commit-related messages I specified above, there is a way to add an extra layer of protection. In case Claude happens to forget the instructions you specified, you can have another layer catching it before the commit goes through.

It's by combining the prepare-commit-msg Git hook with Claude. But more on that in the next tip.

As always, thanks for reading!