Document what the AI can edit, what it must protect, and how to test a page before you paste or publish.
Learn how to make AI coding tools, agents, and assistants understand your project before they start changing things.
This course explains MCP for beginners, AGENTS.md, CLAUDE.md, and context engineering in plain English. You will learn how to write simple AI project instructions so tools like Codex, Claude Code, Cursor, and Replit make fewer mistakes.
Built for creators, founders, marketers, WordPress site owners, and non-technical builders.
MCP helps AI tools connect to other tools and context, like a universal adapter. AGENTS.md gives AI coding agents instructions for your project, like a project instruction manual. CLAUDE.md gives Claude-specific notes. Context engineering means preparing the right background, rules, examples, and limits before asking AI to do work.
The goal is simple: fewer mistakes, safer edits, and better outputs.
This AGENTS.md guide is for people who use AI tools but do not want to become full-time developers just to get safer, clearer results.
Document what the AI can edit, what it must protect, and how to test a page before you paste or publish.
Turn your content process, style rules, and examples into instructions an assistant can follow.
Give AI coding agents for beginners a clear picture of your product, audience, and risk areas.
Protect SEO-critical pages, analytics, tracking scripts, and brand voice while still moving faster.
Use plain language to explain your project before asking Codex, Claude Code, Cursor, or Replit to edit files.
Learn the difference between a one-off prompt and reusable AI project instructions.
By the end, you will understand context engineering for beginners and know how to create project instructions for AI tools.
Claude.md vs AGENTS.md can sound technical, but the practical difference is straightforward.
| Item | Plain-English meaning | Best use | Beginner example |
|---|---|---|---|
| MCP | A standard way for AI tools to connect to outside tools and context. | When an AI assistant needs access to systems like files, databases, docs, or apps. | A universal adapter that helps tools talk to each other. |
| AGENTS.md | A project instruction file for AI coding agents. | When you want Codex or other agents to understand your project rules. | A project manual that says what to edit, what to avoid, and how to test. |
| CLAUDE.md | A Claude-specific instruction file. | When you use Claude Code and want notes tailored to Claude. | Claude-specific notes next to your general project manual. |
| Context engineering | Preparing useful background before asking AI to work. | When you want more reliable results from prompts, agents, or workflows. | Onboarding a new assistant before giving them a task. |
Open each lesson to see a plain-English explanation, why it matters, a beginner example, a copyable prompt, and a simple action step.
Plain-English explanation: AI tools usually know only what is in the current chat, file, or task. If you do not tell them your project rules, they may guess.
Missing context can lead to broken layouts, changed tracking code, removed SEO text, or edits that do not match your brand.
You ask an AI to improve a WordPress page. It changes the design but removes a copy button because it did not know that button was important.
Before editing anything, ask me what project context, protected areas, and testing steps you need to know.
Mini action step: Write down three things your AI tool should always know about your project.
Plain-English explanation: Prompting is the request you type now. Context engineering is the setup you give the AI before the request, including background, examples, rules, and limits.
A great prompt can still fail if the AI does not understand the project. Context makes your prompt safer and more specific.
Instead of only saying “improve this page,” you also say who the page is for, what sections to keep, and what the AI must test.
Use the project instructions before you make changes. If the instructions do not cover this task, ask me before guessing.
Mini action step: Pick one recurring AI task and list the background information it needs every time.
Plain-English explanation: MCP means Model Context Protocol. Think of it as a shared connector that can help AI tools access approved context from other tools.
As AI tools become more useful, they often need access to files, docs, tickets, or other systems. MCP is one way to make those connections more consistent.
Instead of copying every detail into a chat, an AI tool may connect to the right source of context when it is allowed to.
Explain which tools or files you can access for this task, and tell me what context you still need from me.
Mini action step: Write one sentence describing what outside context would help your AI assistant most.
Plain-English explanation: AGENTS.md is a text file that tells AI coding agents how to work in your project. It can include goals, style rules, protected areas, and testing steps.
It turns repeated instructions into a reusable project manual, so you do not have to explain the same rules every time.
Your AGENTS.md can say: keep CSS scoped, do not edit payment code, test mobile layout, and ask before publishing.
Create a beginner-friendly AGENTS.md file for this project. Include project overview, allowed changes, protected areas, testing steps, and approval rules.
Mini action step: Create a rough list of what your AI tool can and cannot change.
Plain-English explanation: CLAUDE.md is a file used by Claude Code for Claude-specific project notes. It can sit alongside AGENTS.md when you use more than one AI tool.
Different AI tools may look for different instruction files. Keeping tool-specific notes clear helps avoid confusion.
AGENTS.md holds general rules. CLAUDE.md can hold instructions that only Claude Code needs.
Review my AGENTS.md and suggest any Claude-specific notes that should go in CLAUDE.md without duplicating everything.
Mini action step: If you use Claude Code, write one Claude-specific instruction you want it to remember.
Plain-English explanation: Good project instructions explain the project, audience, tools, important files, allowed changes, protected areas, testing steps, and approval rules.
The AI needs both permission and limits. Clear instructions reduce accidental changes and make review easier.
For a landing page, the AI can edit headline copy and CSS, but it cannot change analytics tags or legal disclaimers.
Turn my notes into clear AI project instructions with sections for overview, audience, allowed changes, protected areas, testing, and approvals.
Mini action step: Fill in the sentence: “The AI can change ___, but must not change ___.”
Plain-English explanation: Your first file does not need to be perfect. Start with a short, useful version and improve it as you learn what the AI gets wrong.
A simple instruction file is better than no instruction file. It gives the AI a starting point before it touches your work.
Create project-instructions.md with your website name, audience, goal, protected areas, and testing checklist.
Ask me simple questions, one at a time, then create my first project-instructions.md file from my answers.
Mini action step: Use the generator on this page to create your first draft.
Plain-English explanation: WordPress AI coding instructions should say where the code will be used, what theme or page is involved, and what must remain untouched.
WordPress pages often include theme styles, plugins, analytics, forms, and SEO elements. The AI needs to know what is safe.
If you paste HTML into a Custom HTML block, tell the AI to avoid global CSS like body, h1, or button unless scoped to your wrapper.
Create WordPress AI coding instructions for this page. Include the page URL, theme if known, Custom HTML block rules, SEO rules, protected areas, and testing checklist.
Mini action step: Write your website URL and the exact page or block the AI is allowed to edit.
Plain-English explanation: Testing instructions tell the AI what to check before it says the work is done. They should cover buttons, mobile layout, links, forms, and visual issues.
AI tools can create changes that look fine in code but fail in the browser. Testing instructions catch more of those problems.
After adding an accordion, the AI should click each item, test keyboard use, and check the mobile view.
Before you finish, test every button, link, form field, accordion, copy action, and mobile layout. Tell me what passed and what you could not test.
Mini action step: Add three testing steps to your instruction file.
Plain-English explanation: Dangerous changes are edits that could affect money, logins, customer data, legal pages, tracking, security, or live theme files.
AI can move quickly. Clear safety rules slow it down where a human review point is needed.
Your instruction file can say: do not edit checkout, login, database, analytics, or legal pages without permission.
Before making this change, identify whether it touches payment, login, database, customer data, legal, analytics, tracking, SEO-critical, or security areas. Ask before editing any risky area.
Mini action step: Add a “Protected Areas” section to your instruction file.
Use these simple definitions when technical terms start to blur together.
A connector standard that helps AI tools reach approved tools and context.
A project instruction manual for AI coding agents.
Claude-specific notes for Claude Code.
Giving the AI the right background before asking it to work.
Reusable rules that explain your project, goals, limits, and tests.
An AI tool that can inspect or change project files when you allow it.
A moment where the AI must stop and ask you before continuing.
A low-risk edit, such as improving text or styling inside an approved section.
A risky edit involving payments, logins, data, security, tracking, or legal content.
You do not need every file on day one. Choose the simplest option that matches your tool.
When you use Codex or another AI coding agent and want general project instructions.
When you use Claude Code and want Claude-specific project notes.
When you want a plain-language instruction file that works even if your tool does not look for a special filename.
These templates are practical starting points. Replace bracketed placeholders like [PROJECT NAME] and [WEBSITE URL] with your own details.
Use this for general AI coding agent instructions in any project.
# AGENTS.md
## Project
- Name: [PROJECT NAME]
- Website or app: [WEBSITE URL]
- Audience: [TARGET AUDIENCE]
- Main goal: [PROJECT GOAL]
## How the AI should work
- Read these instructions before editing.
- Keep changes focused on the task.
- Explain risky changes before making them.
- Ask if instructions are unclear.
## Allowed changes
- [WHAT AI CAN CHANGE]
## Do not change without permission
- [DO NOT CHANGE]
## Testing steps
- [TESTING STEPS]
## Approval rules
- Show what changed.
- Mention anything that was not tested.
- Do not publish or deploy without human review.
Use this when Codex, Claude Code, Cursor, or another AI tool is helping with a WordPress-related page.
# WordPress Website AGENTS.md
## Project
- Project name: [PROJECT NAME]
- Website URL: [WEBSITE URL]
- WordPress theme, if known: [THEME NAME]
- Page URL: [PAGE URL]
- Goal: [GOAL]
- Audience: [TARGET AUDIENCE]
## Files or page sections involved
- [FILES OR PAGE SECTIONS]
## Design rules
- Match the existing brand and page style.
- Keep layout mobile-friendly.
- Do not add external fonts, scripts, or libraries unless approved.
## SEO rules
- Preserve SEO-critical headings, links, titles, and important page copy.
- Do not make unsupported claims.
- Do not keyword stuff.
## AI can change
- [WHAT AI CAN CHANGE]
## AI must not change
- Payment systems
- Login systems
- Databases
- Customer data
- Tracking scripts
- Analytics tags
- Legal pages
- Live theme files
- [DO NOT CHANGE]
## Testing checklist
- Test links and buttons.
- Check desktop and mobile layout.
- Check forms, accordions, copy buttons, and interactive elements.
- Confirm no obvious visual overlap.
- [TESTING STEPS]
## Rollback instructions
- Tell me which file or block changed.
- Keep a copy of the previous version if possible.
- If something breaks, restore the last working version.
Use this when building a single copy/paste HTML block inside WordPress.
# Custom HTML Block Instructions
## Goal
Create or edit a single self-contained HTML block for [PROJECT NAME].
## Requirements
- Must be one copy/paste block for WordPress Custom HTML.
- Use vanilla HTML, CSS, and JavaScript only.
- Do not use React.
- Do not use external libraries.
- Do not use external fonts.
- Do not use external scripts.
- Do not require a build step.
- Scope all CSS under [WRAPPER CLASS].
- Do not use global CSS that affects body, html, headings, buttons, inputs, or tables outside the wrapper.
- Keep JavaScript scoped to this block.
- Build mobile-first.
- Use accessible buttons and form labels.
- Make text readable on mobile.
## AI can change
- [WHAT AI CAN CHANGE]
## AI must not change
- [DO NOT CHANGE]
## Testing steps
- Paste into a test WordPress Custom HTML block.
- Preview desktop and mobile.
- Test all buttons, links, copy actions, forms, and accordions.
- Confirm no console errors if testing tools are available.
Use this for small tools like calculators, quizzes, checklists, or lead magnets.
# AI Calculator or App Instructions
## Project
- Tool name: [PROJECT NAME]
- Tool type: [CALCULATOR / QUIZ / CHECKLIST / LEAD MAGNET]
- Audience: [TARGET AUDIENCE]
- User goal: [WHAT USER WANTS TO DO]
## Inputs
- [INPUT FIELD 1]
- [INPUT FIELD 2]
- [INPUT FIELD 3]
## Outputs
- [RESULT OR RECOMMENDATION]
- Include plain-English explanations.
- Do not make medical, legal, financial, or technical claims unless I provide approved wording.
## Design rules
- Keep the tool easy to use on mobile.
- Use clear labels and buttons.
- Show helpful error messages.
## Testing steps
- Test empty fields.
- Test normal examples.
- Test unusually long answers.
- Test mobile layout.
- Test reset or copy buttons if included.
Use this for creating or improving SEO articles without hallucinating or making unsupported claims.
# Blog SEO Instructions
## Article
- Topic: [TOPIC]
- Target reader: [TARGET AUDIENCE]
- Primary keyword: [PRIMARY KEYWORD]
- Related phrases: [RELATED PHRASES]
- Search intent: [WHAT THE READER WANTS]
## Writing rules
- Write for humans first.
- Use beginner-friendly language.
- Explain technical terms immediately.
- Do not keyword stuff.
- Do not invent facts, stats, quotes, or sources.
- Flag anything that needs verification.
## SEO rules
- Use clear H2 and H3 sections.
- Answer the main question early.
- Include examples and practical steps.
- Preserve important internal links.
## Do not change
- [DO NOT CHANGE]
## Review
- Check claims.
- Check headings.
- Check links.
- Check readability on mobile.
Use this to tell the AI how to test buttons, mobile layout, links, forms, accordions, copy buttons, and visual layout.
# Testing Instructions
Before saying the work is done, test:
## Interaction
- Buttons
- Links
- Forms
- Accordions
- Copy buttons
- Navigation
## Layout
- Desktop view
- Tablet view if possible
- Mobile view
- Long text
- Narrow screens
- Tables and code boxes
## Accessibility
- Buttons are real buttons.
- Form fields have labels.
- Keyboard focus is visible.
- Text contrast is readable.
## Report back
- What passed: [TESTING STEPS]
- What failed:
- What could not be tested:
- Any risky areas that need human review:
Use this when you want the AI to stop before making risky changes.
# Safety Rules
The AI must ask before making risky changes.
## Do not edit without permission
- Payment systems
- Checkout flows
- Login systems
- User accounts
- Databases
- Customer data
- Tracking scripts
- Analytics tags
- Legal pages
- Privacy policy
- Terms pages
- Live WordPress theme files
- Security settings
- SEO-critical pages
- [DO NOT CHANGE]
## Publishing rule
- Do not publish without review.
- Do not deploy without approval.
- Do not remove analytics or tracking.
- Ask before making any change that could affect money, access, data, legal content, security, or search visibility.
Fill in the fields and generate a simple AGENTS.md or project-instructions.md draft. This tool runs in your browser. It does not send your answers anywhere.
These areas can affect money, access, customer trust, legal risk, search visibility, or site security.
Here are simple ways to use AI project instructions in real work.
Tell the AI the page URL, Custom HTML block rules, protected plugins, SEO rules, and mobile testing steps.
Document inputs, outputs, assumptions, disclaimers, and edge cases before the AI builds the tool.
Give audience, search intent, approved claims, internal links, and facts that need verification.
Define the offer, CTA, audience, brand voice, conversion goal, and sections that must stay intact.
Describe screens, user flow, allowed features, data rules, and testing steps.
List title rules, script style, thumbnail notes, publishing steps, and review points.
Create your first project instruction file for your website, app, or content workflow.
Edit these placeholder links after the related guides are live.
MCP is a standard way for AI tools to connect to approved tools and context. A simple analogy is a universal adapter.
No. You only need the basic idea: MCP helps AI tools connect to useful context instead of relying only on what you type.
AGENTS.md is a project instruction file for AI coding agents. It tells the AI how to work, what it can change, what it must protect, and how to test.
No. Non-technical builders can use AGENTS.md to explain goals, rules, protected areas, and review steps in plain language.
CLAUDE.md is a Claude-specific instruction file often used with Claude Code. It can hold notes that are meant for Claude in particular.
AGENTS.md is usually a general instruction file for AI agents. CLAUDE.md is specifically for Claude Code. Some projects use both.
Context engineering means giving AI the right background, rules, examples, and limits before asking it to do work.
Yes. You can document WordPress pages, Custom HTML blocks, theme limits, SEO rules, and testing steps before asking AI to edit anything.
Yes. Codex can use project instructions to understand what you want changed, what should stay protected, and how to verify the work.
Yes. The same ideas work across Claude Code, Cursor, Replit, and other AI coding tools, though each tool may handle instruction files differently.
Yes, especially if they edit risky areas without review. That is why protected areas, testing steps, and approval rules matter.
Tell the AI the goal, allowed changes, protected areas, testing steps, and review rules. If possible, keep a backup or previous version.
Start with one simple instruction file. Improve it whenever an AI tool misunderstands your project.
Kingy AI helps AI companies turn complex products into clear, useful YouTube videos that drive awareness, product understanding, demos, clicks, and search visibility.
© 2026 Kingy AI
© 2026 Kingy AI