AI Coding Cheat Sheet

For the non-developer founder. Print this. Keep it on your phone. Follow it every time.

BEFORE CODING (Pre-Flight)

• Started a new chat for this task
• AI has read AI-RULES.md
• Goal written as one clear sentence

Copy-paste to start every session:

Follow AI-RULES.md strictly. Before writing any code: 1) Explain in plain English how you will do this. 2) List exactly which files you will change. 3) Tell me what could break. 4) Ask me questions if anything is unclear. Do not write code until I say go.

Then ask:

• "What is the simplest way to do this?"
• "What files will you touch?"
• "What could break?"

DURING CODING

• Working on one file at a time
• Testing every change before moving to the next
• AI is explaining what it does, not just dumping code

Magic phrases:

• "One file at a time. Show me what changed in plain English."
• "Is this the simplest way? No new libraries."
• "Do not write any code yet." (when you need it to stop and think)
• "Explain this like I am 5 years old."
• "What edge cases are we forgetting?"

Ask after each change:

• "How do I test if this worked? Tell me exactly what to click."
• "Does this affect login, payments, or the database?"

AFTER CODING (Post-Flight)

• AI explained every change in plain English
• Ran the 5-minute test (below)
• Got rollback instructions

Copy-paste when AI says it is done:

Show me every file you changed and explain each change in plain English. Then tell me: what should I test? What else could this have broken? How do I undo this if something goes wrong?

The final question:

Would you bet $10,000 that this works perfectly in production?

5-MINUTE TEST (do this every single time)

• Does the app start without errors?
• Can I still log in?
• Does the main page load?
• Does the new feature work as expected?
• Click 3 old features — do they still work?
• Is it fast or suddenly slow?
• Any error messages on screen?

If anything fails — STOP. Do not ask the AI to "just fix it." Go to the emergency section.

WHEN THINGS BREAK (Emergency)

Do NOT say: "It broke, fix it" Do say:

STOP. Something is broken. Do not guess. Here is the exact error: [paste error]. Tell me the top 2 most likely causes and what info you need from me to know for sure.

If the AI has failed 3 times:

We are stuck in a loop. Forget our last 5 messages. Revert all changes to the last working state. Let us zoom out — what is the actual problem and what is a completely different approach?

Nuclear option:

Revert everything. Go back to the last commit that worked. We are starting over with a simpler approach.

RED FLAGS — STOP if the AI says:

• "Let me refactor this while I am here..." — NO. Fix only what was asked.
• "I have rewritten the entire file..." — REVERT. It will break things.
• "This is a quick hack/workaround..." — NO. Do it right or do not do it.
• "We should probably upgrade/install..." — NO. Use what is already there.
• "It should work" / "Probably fine" — NOT GOOD ENOUGH. Prove it.
• "Just ignore that warning..." — NO. Warnings become errors.

Your response to any red flag:

Undo that. We are taking the safer approach. Explain the safer alternative.

MAGIC PHRASES (keep these in your back pocket)

GOLDEN RULE

If you cannot explain what just changed in one simple sentence, revert and start over.

Source: EdgeClaw Council System — 5 models, 2026-04-09