Show HN: Rules that stop AI coding agents from breaking working code

gaiinmaster1 pts0 comments

GitHub - avenna01-ceo/claude-code-survival-kr: Rules and prompts that stop Claude Code / Cursor from breaking working code. Copy-paste. (EN/KR) · GitHub

/" data-turbo-transient="true" />

Skip to content

Search or jump to...

Search code, repositories, users, issues, pull requests...

-->

Search

Clear

Search syntax tips

Provide feedback

--><br>We read every piece of feedback, and take your input very seriously.

Include my email address so I can be contacted

Cancel

Submit feedback

Saved searches

Use saved searches to filter your results more quickly

-->

Name

Query

To see all available qualifiers, see our documentation.

Cancel

Create saved search

Sign in

/;ref_cta:Sign up;ref_loc:header logged out"}"<br>Sign up

Appearance settings

Resetting focus

You signed in with another tab or window. Reload to refresh your session.<br>You signed out in another tab or window. Reload to refresh your session.<br>You switched accounts on another tab or window. Reload to refresh your session.

Dismiss alert

{{ message }}

avenna01-ceo

claude-code-survival-kr

Public

Notifications<br>You must be signed in to change notification settings

Fork

Star

main

BranchesTags

Go to file

CodeOpen more actions menu

Folders and files<br>NameNameLast commit message<br>Last commit date<br>Latest commit

History<br>3 Commits<br>3 Commits

README.ko.md

README.ko.md

README.md

README.md

View all files

Repository files navigation

Stop Your AI From Breaking Working Code

Rules and prompts that keep Claude Code, Cursor, and Codex from wrecking things that already worked.<br>Everything here is copy-paste. Explanations kept to a minimum.

한국어 →

The first two weeks feel like magic. Then the codebase grows past what the model<br>can hold at once, and it starts fixing A while breaking B.

That isn't a skill problem. It's a structure problem — which means prompting<br>harder won't fix it. Constraints will.

The one thing to do tonight

Create CLAUDE.md in your project root and paste this.<br>Claude Code reads it automatically at the start of every conversation.

(Cursor → .cursorrules, Codex/Copilot → AGENTS.md. Same content.)

# Project

## What this is<br>[one sentence]

## Current state<br>[what works, what doesn't yet]

## Do not touch<br>- [file/folder] — [why]

## Rules<br>- Do not modify files I did not ask you to modify<br>- Do not refactor working code without being asked<br>- If unsure, say "I don't know". Never present a guess as fact<br>- Never report something as done without verifying it

Why "do not" instead of "please do"

Please write clean code does nothing. There's no threshold for clean,<br>so no behavior changes.

Prohibitions have a clear edge. They land immediately.

❌ No effect<br>⭕ Works

Write clean code<br>Do not modify files I didn't ask about

Be careful<br>Do not merge without tests

Keep it organized<br>Do not touch legacy/

5 prompts that prevent breakage

1. Before letting it change anything

Don't write code yet. First tell me which files you'd touch<br>and where the risky parts are.

Starting with "fix it" means the model edits things it can't itself account for.<br>Getting the plan first kills most collateral damage.

2. When you're scared it'll break something

Before changing this, tell me what else is affected by it.

Works better than scolding. It ends up confessing:<br>"ah, this is coupled to the character class."

3. When you get "Done!"

Don't tell me it works. Write a test and show me it passes.

Models are bad at catching their own errors. They review with the same<br>assumptions they built with — so a wrong assumption survives both passes.<br>"Looks good" is an opinion. A passing test is evidence.

4. Before starting a fresh chat

Summarize: what's done / what's next / what we tried and abandoned.<br>Write it so a fresh session can continue from that alone.

"What we abandoned" is the important part. Without it, the new session<br>confidently proposes the exact approach that just failed.

5. The line for your rules file

Do not modify files I did not ask you to modify.

Single highest-value line in the file.

Context rots

Long sessions get worse, and it isn't your imagination.

As the conversation grows, abandoned directions and corrected mistakes<br>all keep counting as current context. The signal gets buried in its own history.

Start a fresh session every 3–5 tasks. Run prompt #4 before you close.

Anger is the signal

If you're asking for the same thing a third time, stop rewording the request<br>and change the window. That session can no longer produce the answer.<br>Pushing harder only raises your blood pressure.

Make everything reversible

If git feels like a wall, you don't need the commands. Just say:

Before starting: save the current state<br>When it breaks: restore the state you saved

Nobody plays a game without saving. Same rule.

One rule for permissions

Can Ctrl+Z undo this?

Just let it run<br>Always confirm first

Reading, searching, analyzing<br>Deleting

Local edits (if you saved state)<br>Deploying

Drafting<br>Sending email / messages

Running tests<br>Payments,...

code session files claude rules from

Related Articles