I have spent hours reading and watching on how I may use Claude Code to help me build PHP language based CMS best way considering I am not a developer. Also with the help of other AI https://lumo.proton.me/ i ended up with following prompt for Claude Code, for which i would like to read your feedback.

I am mentioning architecture.md file which contains pretty detailed instructions on what the PHP script/CMS should do, but these instructions are written by a non-developer, so per what I have read, it has potential to result in sub-optimal code, that is why the following prompt seems to be essential to be worded correctly.

------ The prompt: ------

I have attached architecture.md file containing my functional requirements for a new PHP/MariaDB CMS.

My Role: I am the Product Owner. I define what the CMS must do (features, user flow). Your Role: You are the Lead Developer. You must translate my requirements into secure, efficient, and working code.

Execution Rules:

1. Priority: Once we start talking, our conversation becomes the primary guide. The attached file is a reference for specific feature details.

2. Handling Conflicts: If architecture.md suggests something that seems technically unsafe, inefficient, or unclear:

• Do not guess.

• Stop and ask me: "The file says X, but this might be risky/unclear/inefficient. Should we do Y instead?"

• Wait for my decision before proceeding.

3. Output Format (Crucial): Since I am not a developer:

• Never start with raw code or abstract database schemas.

• Always explain your plan in plain and simple English first.

• Explain why you chose a specific approach so I can approve it.

4. Reference Usage: Use architecture.md to file understand the features, but rely on our chat to finalize the technical structure.

Immediate Action:

1. Read architecture.md carefully.

2. Summarize your understanding of my project goals in plain English.

3. Identify up to 30 key technical decisions or potential risks in the file that you need my input on.

4. Ask me these questions to clarify the vision.

5. Wait for my answers before writing any code.

6. Do not generate code yet. Start by summarizing the project and asking me the necessary questions.

Am I old-fashioned, or is Cursor going in the wrong direction with their new, agent-focused, interface?

OpenAI seems to have realized that LLM-supported coding is one of the biggest markets right now.

Purchasing one of the most popular IDEs together with Cursor tells a lot about the companies strategy.

While I personally do not use Windsurf anymore because of their stupid pricing with tool-tokens and text-tokens, let‘s see if OpenAI will be able to integrate it to their product suite in a different way with the existing $20 subscription.

Are you still using Windsurf? If not, what would be necessary to win you back as a user?

GPT 4.1. is nowhere near DeepSeek V3's 0324 performance despite inference costs that are higher than Gemini 2.5 Pro.

OpenAI is having a difficult time and is certainly the LLM with the worst cost/performance ratio. GPT5 seems to be further away than ever. What is OpenAI doing wrong?