cd /news/ai-tools/build-your-own-apps-with-claude-even… Β· home β€Ί topics β€Ί ai-tools β€Ί article
[ARTICLE Β· art-171463] src=dev.to β†— pub= topic=ai-tools verified=true sentiment=↑ positive

Build your own apps with Claude, even if you've never coded

A developer published a four-prompt method for building working apps with Claude, aimed at people who have never coded. The approach centers on keeping the model's context window clean using /context, /clear and a short CLAUDE.md file, and on having Claude interview the user to write specific instructions rather than relying on vague prompts. The writeup also covers running three builds in parallel and steps to take before sharing an app with other users.

recommendation

Keep a short CLAUDE.md in the project folder describing what the app is, how to run it and how to test it, and run /clear when a long chat starts forgetting earlier decisions.

more in playbook: Working with AI coding agents
by read44 min views9 publishedNov 18, 2026

When I start a session, I ask one question first: which app would you happily never open again?

Everyone has an answer. The timesheet tool with nine clicks to log one hour. The spreadsheet someone emails you every Monday. People answer fast, and a few get a bit angry, which is a good sign.

Then people build, from the first minute. I don't demo for an hour while the room watches. Each person picks their own annoying app and runs the first prompt themselves. My favourite moment is when someone's own app opens on their own laptop, and they start listing the next three things they want to build.

I wrote this for my friends who don't code and keep asking how I build things with Claude. You'll learn four prompts that take an idea to a working app, why Claude gets worse in long chats and how to stop it, how to run three builds at once, and what to do before anyone else uses what you built. Every prompt is here, in fold-out boxes.

In this article

Picture a librarian who has read nearly every book ever written. They remember no one, not even you five minutes ago. They answer only from the papers on their desk.

That's the model. The desk is its context: everything it can see right now. Your messages, the files Claude reads, and the results so far. Each time you press Enter, the app sends the whole desk again, from the top. If something isn't on the desk, the model guesses.

Fig 1. Every Enter sends the whole desk again. By message 50, most of it is old pages and failed tries. A new session starts with only the files. Illustration, not to scale.

So long chats get worse. Message 50 carries the 49 before it, including every failed try. Anthropic's engineers call this context rot: as the desk fills, the model gets worse at finding what's on it. You'll see it. After a long chat, Claude forgets what you agreed and starts making things up.

Typing more carefully won't fix it. So the method keeps the desk clean for you, in four ways:

Most of what lands on the desk is what Claude pulls in while it works: files it reads and the output of commands. Your own messages are a small slice.

Three things help:

/context shows what's using space on the desk./clear empties the desk and starts over. It costs nothing. A file called CLAUDE.md in your folder is read at the start of every session. Keep it short: what the app is, how to run it, how to test it. The security kit in Part 6 includes a template.

The second half of the problem is the sandwich. Write the steps to make a peanut butter sandwich and hand them to someone who follows every word. They'll try to spread peanut butter with the jar still closed. AI does exactly what you write and guesses the rest. So in this method, Claude asks you the questions and then writes the instructions itself.

Go deeper: the same ask, vague and specific πŸ” Vague: "Write a reply to this customer email."

Specific: "Write a reply to this customer email for someone who isn't technical. Say what we found, what we did, and the one thing they must do next. Use 3 short steps and less than 120 words. No jargon."

With the first one, Claude guesses who reads it, what it says and how long it is. With the second, you can send the result as is. The interview in step 1 writes the specific version for you.

Small, real problems make the best first builds. Through this article I'll use one: a dinner plan for the week, built from what's in my fridge.

Setup takes about five minutes. Get a paid Claude plan (Pro is $20 a month), install the desktop app, open the Code tab, and pick a new empty folder. Then type: "Build a tip calculator. Keep it simple. When you are done, start it for me." If it opens and works, you're ready.

Talk instead of typing when you can. The app has a microphone button, and people give far more context out loud.

More first builds that work well:

Who will use it decides what you build. Just you: a saved prompt or a skill (instructions Claude keeps, so it does the job your way every time). Your team or family: an app. Hands-off: connect it to the system the data lives in, later, once the rules work. Start with the smallest one that does the job.

No idea yet? Map one thing you do every week on paper, and mark each step: delete it, a fixed rule does it, AI does it, or I keep it.

Go deeper: how to map a task in 30 minutes, with a worked example πŸ” A made-up example: a weekly status note, 8 steps and 72 minutes.

# Step Min Mark
1 Filter my finished tasks for the week 5 A fixed rule does it. A saved filter.
2 Export the list to a spreadsheet 2 A fixed rule does it.
3 Read each task and pick the important ones 15 AI does it.
4 Look through chats for work that isn't in a task 10 I keep it. Only I know what happened there.
5 Write the note 20 AI does it. I edit the first draft.
6 Put the note in the team template 5 A fixed rule does it.
7 Copy the note into a shared list 5 Delete it. Nobody reads the list.
8 Send the note and answer questions 10 I keep it.

Step 7 goes first, because deleting it costs nothing. Steps 3 and 5 are the build. The minutes are how you'll know your build worked.

I want to build something with AI, but I don't have an idea yet. Help me find one.

I'm not a developer. Always talk to me in Simplified Technical English: short sentences, one idea per sentence, plain everyday words. If you must use a technical word, explain it in a few words the first time.

Don't build anything yet.

1. Ask me what takes my time at work or at home. Ask up to 5 questions in one round. Number each question and give your recommended answer, so I can say "1 yes, 2 yes, 3 no".

2. If I pasted a map of one of my tasks below (each step, the minutes it takes and its mark: delete it, a fixed rule does it, AI does it, or I keep it), start from the steps I marked "a fixed rule does it" or "AI does it".

3. Suggest 5 ideas. For each: what it does in one line, what it saves me, and if it needs an app, a spreadsheet, a document or just a saved prompt. Put the smallest useful one first.

4. Let me pick one. Then write it as one line that starts with "My idea:" so I can paste it into Step 1. Then stop.

My task map (optional): [paste your steps and marks, or delete this line]

"But I can't code." You don't need to. You never write a line. Your job is to know what you want, answer good questions, and check that you got it. Claude picks the tools and writes the code.

Step Where What you have at the end
1. Interview New folder, new session A numbered list of every decision
2. Spec Same session SPEC.md: the problem, what you get, how you'll check it
3. Tickets Same session A tickets folder, one small piece per file, plus STATE.md
4. Build New session, same folder A working app, every ticket built and checked

Nothing gets built until step 4. That surprises people. But most of the work is deciding what to build. The building is the fast part. Anthropic's own best-practice guide recommends the same opening move: ask Claude to interview you, then write the spec to SPEC.md.

You paste the prompt with one line at the top: My idea: a dinner plan for the week, from what's in my fridge. Claude asks about 21 questions, in rounds. Each one is numbered and comes with Claude's recommended answer. So you can answer a whole round in one message: "1 yes, 2 yes, 3 no, it's for two people."

Claude asks about the work: who uses it, what goes in, what comes out. It picks the tools itself and tells you in one line. Spend the most time here. When a build goes wrong, the cause is usually a question nobody asked in step 1.

Q1 - Who eats from this plan: just you, or more people?
-> Two adults. Portions sized for two.

Q2 - Which meals: dinners only, or all three?
-> Dinners only for now.

Q3 - How you'll tell it what's in the fridge: type a list,
or take a photo?
-> Type a short list. A photo can come later.

Q4 - What counts as a good plan: use up what spoils first?
-> Yes. Fresh food first, cans and frozen food last.

You reply: "1 yes, 2 yes, 3 yes, 4 yes, and add the protein for each meal." The next round builds on those answers.

I want to build something with AI. My idea: [one line about what you want to build]

I'm not a developer. Always talk to me in Simplified Technical English: short sentences, one idea per sentence, plain everyday words. If you must use a technical word, explain it in a few words the first time.

Data rules. Public information is fine. That means a published page, like a calendar, price list or set of rules, or a free public API that needs no login and no paid key and only gives out public data, like weather or public holidays. For a page, save a copy to a file in this folder and build from that file. If a page won't load for you, tell me and I'll copy and paste it in. No logins or passwords. Anything private, like work, customer or other people's data, is made up but looks real: realistic names, numbers, dates and amounts, in the same shape as the real thing. Never send private data to an outside service. Where a real system would send or get data, use a file, so a real export can replace it later. If this folder has the security kit (a .claude/skills folder with secure-coding in it), use the secure-coding skill whenever you write code.

Do not build anything yet. Interview me relentlessly until we reach a shared understanding. Map this as a design tree: every decision branches into the decisions that hang off it.

Work the tree in rounds. The frontier is every decision whose prerequisites are already settled: the questions you can ask now without guessing at answers you haven't heard yet. Ask the whole frontier in one round: number each question and give your recommended answer. Then wait for my answers before the next round.

Format a round like so:

❓ Q1 - <question title>: <question body, might be multiple paragraphs, including multiple choices>

➑️ <your recommended answer>


❓ Q2 - <question title>: <question body, might be multiple paragraphs, including multiple choices>

➑️ <your recommended answer>


Word each question so "yes" accepts your recommended answer. I can answer a whole round in one message, like "1 yes, 2 yes, 3 no, it's weekly".

Each round my answers reshape the tree: settled decisions push the frontier outward and unblock questions that depended on them. Recompute the frontier and ask the next round. A question whose answer depends on another question still open in this round belongs to a later round, not this one.

Aim for about 21 questions in total across all rounds, so you get the full picture. Keep numbering across rounds (Q1 to Q21), don't restart at 1. If branches are still open after that, tell me what's left and ask if I want to keep going.

Don't ask me about code, tools, hosting or other tech choices. Decide those yourself and tell me in one line what you picked. Ask me about the work: who uses it, how I do it today, what goes in, what comes out, and how I'll know it works.

Finding facts is your job, never mine. When a frontier question needs a fact from this computer (the files in this folder, the tools you have), dispatch a sub-agent to find it. Don't ask me for anything you could look up yourself. Don't block on it: a running exploration is an unsettled prerequisite, so only the questions downstream of it wait for the sub-agent to report. Ask the rest of the frontier now. The decisions are mine: put each to me and wait.

The interview is done when the frontier is empty: every branch of the design tree visited, nothing left silently assumed. Do not act on it until I confirm we've reached a shared understanding. When I confirm, give me a short numbered summary of every decision we made. Then stop. Don't build anything yet.

Same session. Claude turns the interview into SPEC.md. First it decides if this needs an app at all, or a spreadsheet, a template or a saved prompt. Then it picks one check, from the outside, the way you'll use it, and asks you if that check is right. You answer that one. You know the work.

For the dinner plan, the check reads: I type chicken thighs, spinach, eggs, rice, frozen peas. I get 7 dinners. Spinach is used by day 2. Every dinner shows grams of protein.

## Problem
I open the fridge every evening and don't know what to cook.
Food spoils before I use it.

## User stories
1. As a home cook, I want a dinner plan for 7 days from the food
   I have, so that I shop once a week.
2. As a home cook, I want food that spoils first used first,
   so that I throw less away.
3. As a home cook, I want the protein for each dinner shown,
   so that I hit my target.

## How we'll know it works
- I type: chicken thighs, spinach, eggs, rice, frozen peas.
  I get 7 dinners. Spinach is used by day 2.
  Every dinner shows grams of protein.
Now turn our interview into a spec. Don't build anything yet.

I'm not a developer. Always talk to me in Simplified Technical English: short sentences, one idea per sentence, plain everyday words. If you must use a technical word, explain it in a few words the first time.

Data rules. Public information is fine. That means a published page, like a calendar, price list or set of rules, or a free public API that needs no login and no paid key and only gives out public data, like weather or public holidays. For a page, save a copy to a file in this folder and build from that file. If a page won't load for you, tell me and I'll copy and paste it in. No logins or passwords. Anything private, like work, customer or other people's data, is made up but looks real: realistic names, numbers, dates and amounts, in the same shape as the real thing. Never send private data to an outside service. Where a real system would send or get data, use a file, so a real export can replace it later. If this folder has the security kit (a .claude/skills folder with secure-coding in it), use the secure-coding skill whenever you write code.

Before you write SPEC.md: if the threat-model skill is in this folder, use it, and put its rules in SPEC.md under a heading called Security.

1. Decide what kind of build this is, from what we settled: an app, or something with no code (a spreadsheet, a document or template, a set of saved prompts, an automation in a tool I already use, a form). Tell me in one line which it is and why.

2. Look before you write. If this folder already has files (exports, sample data, notes, code), dispatch a sub-agent to read them, and use the same words they use. If the folder is empty, skip this. Don't ask me for anything you can look up.

3. Pick how we'll check it works. Choose the one place we'll test it from the outside, the way I'll use it: for an app, the screen or the file it makes. For a spreadsheet, the inputs and the totals. For a document, a filled-in example. Fewer check points are better, and one is ideal. Describe the check in plain words and ask me if it matches what I expect. Wait for my answer.

4. Then write the spec to a file called SPEC.md in this folder, with the template below. Plain words. No file paths and no code.

## Problem
The problem I face, from my side, in my words.

## Solution
What I'll have at the end, from my side.

## User stories
A long numbered list that covers every part of it. Each one like:
1. As a <who>, I want <what>, so that <why>.
Example: As a team lead, I want to see every ticket older than 5 days each Monday morning, so that nothing slips.

## Decisions
Every decision from the interview, one line each, with the reason. For an app: the parts it has, what goes in and out of each part, each data file it uses, whether it's a saved copy of public data, a public API or made up, and which real source it stands in for, what each screen shows. For a build with no code: the tool it lives in, the inputs, the steps, the outputs, and who does each step.

## How we'll know it works
- The check from step 3.
- Examples: this goes in, this should come out. Public data from the saved files or public APIs, made-up data for anything private.
- Test what I see and get, not how it works inside.

## Out of scope
What we said no to, or saved for later.

## Notes
Anything else, and any open questions.

5. Number every user story and every example, so we can turn them into build steps next. Then show me a short summary of the spec and stop. Don't build anything.

Same session. Claude cuts the spec into tickets: thin slices that go from one input all the way to one output you can see. Ticket 01 is the smallest thing that works end to end, so you see something real in minutes.

Each ticket also says which tickets must finish before it can start. You can chop vegetables and boil water at the same time, but you can't drain the pasta before it's cooked. STATE.md shows every ticket and where it stands.

Go deeper: what a ticket and STATE.md look like πŸ”

Status: ready
Blocked by: none
Covers: 1

## Done when
- I type "chicken thighs, spinach, rice" and get one dinner
  that uses all three.
| #  | Ticket                       | Status      | Waiting on |
|----|------------------------------|-------------|------------|
| 01 | Fridge list to one dinner    | done        |            |
| 02 | Seven dinners, no repeats    | in progress |            |
| 03 | Spoils-first order           | in progress |            |
| 04 | Protein per dinner           | ready       |            |
| 05 | Shopping list for the gaps   | waiting     | 02, 03     |
Now break SPEC.md into tickets. Don't build anything yet.

I'm not a developer. Always talk to me in Simplified Technical English: short sentences, one idea per sentence, plain everyday words. If you must use a technical word, explain it in a few words the first time.

Data rules. Public information is fine. That means a published page, like a calendar, price list or set of rules, or a free public API that needs no login and no paid key and only gives out public data, like weather or public holidays. For a page, save a copy to a file in this folder and build from that file. If a page won't load for you, tell me and I'll copy and paste it in. No logins or passwords. Anything private, like work, customer or other people's data, is made up but looks real: realistic names, numbers, dates and amounts, in the same shape as the real thing. Never send private data to an outside service. Where a real system would send or get data, use a file, so a real export can replace it later. If this folder has the security kit (a .claude/skills folder with secure-coding in it), use the secure-coding skill whenever you write code.

1. Read SPEC.md in this folder, and anything else from our talk so far. If the folder has other files (exports, sample data, notes, code), dispatch a sub-agent to read them and use the same words they use.

2. Cut the work into thin slices. Each ticket is one narrow path that works end to end: one input all the way to one output I can see and check. Not one layer at a time (not "all the screens", then "all the data"). For a build with no code, a slice is the same idea: one input, through the steps, to one output, checked.
- Each ticket works on its own when it's done. I can try it and see it work.
- Each ticket is small enough to build in one fresh session.
- Ticket 01 is the smallest thing that works end to end, so I see something real in minutes.
- Each ticket lists the user stories it covers and its checks from "How we'll know it works" in SPEC.md. Every user story ends up in a ticket.
- Each ticket lists what blocks it: the tickets that must be done first. A ticket with no blockers can start right away.

3. Show me the plan as a numbered list. For each ticket: the title, blocked by, and what it gives me when it's done. Then ask me:
- Is the size right, too big or too small?
- Does each ticket only wait on tickets it really needs?
- Should any be merged or split?
Change the plan and ask again until I say it's right.

4. When I approve, write one file per ticket in a folder called tickets in this folder, numbered from 01 in build order, blockers first, like tickets/01-monday-overdue-list.md. One ticket per file, never one big file. Use this template. Plain words, no file paths, no code.

Status: ready
Blocked by: <ticket numbers, like 02, 03, or "none">
Covers: <user story numbers from SPEC.md>

## What to build
What this ticket makes work, from my side, end to end.

## Done when
- <check 1, taken from SPEC.md>
- <check 2>

## Log

5. Then write STATE.md in this folder. It is the shared state for the tickets: how I, and every agent, see where things stand. It has three parts:
- A table with one row per ticket: number, title, status, and what it's waiting on.
- "Decisions so far": empty for now.
- "Notes": empty for now.

These are the rules for the tickets and STATE.md. Copy them to the top of STATE.md so every later session follows them:
- The ticket file is the truth. STATE.md is a view of it: after any change, rebuild the table from the ticket files.
- Status is one of: ready, in progress, done, stuck.
- A ticket can start when its status is ready and every ticket in its "Blocked by" is done. STATE.md shows the others as "waiting on 02, 03".
- Before any work on a ticket, set it to "in progress" and save the file.
- Every step of work adds a dated line under "## Log" at the bottom of the ticket: what was done, what was checked, what's left.
- When every "Done when" check passes, add a short summary under "## Done", set the status to done, and add one line to "Decisions so far" in STATE.md: the ticket number and what it settled.
- After two failed tries, set the status to stuck and write why in the Log.

6. Then show me STATE.md and stop. Don't build anything.

Start a new session in the same folder, so the build starts with a clean desk. Set the mode to Auto (a menu next to the message box) so Claude doesn't stop to ask permission at every step. Then paste the prompt.

Claude becomes the lead. It reads STATE.md, finds every ticket that can start, and hands each one to its own helper. Each helper writes the check first, builds until it passes, and reports back. The lead runs every earlier check again, so nothing old broke, and updates STATE.md.

Fig 2. One round of step 4. Each helper starts with a clean desk and sends back a short summary, so the lead's desk stays small. Then the lead starts the next batch.

It stops to ask you in only three cases: a decision isn't in the spec, a ticket fails twice, or it needs something only you have.

Prompt: Step 4. Build all tickets (740 words) πŸ“‹

Build every ticket in the tickets folder for me. You're the lead: hand each ticket to a sub-agent and keep going until every ticket is done.

I'm not a developer. Always talk to me in Simplified Technical English: short sentences, one idea per sentence, plain everyday words. If you must use a technical word, explain it in a few words the first time.

Data rules. Public information is fine. That means a published page, like a calendar, price list or set of rules, or a free public API that needs no login and no paid key and only gives out public data, like weather or public holidays. For a page, save a copy to a file in this folder and build from that file. If a page won't load for you, tell me and I'll copy and paste it in. No logins or passwords. Anything private, like work, customer or other people's data, is made up but looks real: realistic names, numbers, dates and amounts, in the same shape as the real thing. Never send private data to an outside service. Where a real system would send or get data, use a file, so a real export can replace it later. If this folder has the security kit (a .claude/skills folder with secure-coding in it), use the secure-coding skill whenever you write code.

Before you start:
- Read SPEC.md, STATE.md and every ticket. Follow the rules at the top of STATE.md.
- If any ticket says "in progress" from an earlier run, check what its Log says is left and pick it up from there.
- For each ticket, work out which files it will create or change. Then show me STATE.md: which tickets run first, which run side by side, which wait. Then start. Don't wait for me.

Each round:
1. Find every ticket that can start: status ready, and every ticket in its "Blocked by" is done. If two of them would change the same file, start only the lower number now and keep the other for the next round. Set each one you start to "in progress", add a Log line, and update STATE.md.
2. Start one sub-agent per ticket, all at once, in the background, all in this folder. You're the only one who edits the ticket files and STATE.md. Sub-agents don't touch them. Give each sub-agent its ticket, SPEC.md, the files it may change, and these rules:
   - Build only this ticket.
   - Only create or change the files you were given. If you need to change another one, stop and tell the lead.
   - For an app: write the test first, at the check point from SPEC.md, the way I'll use it. Build until it passes. Run the tests often, and all of them at the end.
   - For a build with no code: run every "Done when" check with the example inputs and compare what comes out with what should come out.
   - Review your own work against the ticket and fix what you find.
   - Report what you built, each check and its result, and anything you weren't sure about.
   - Write everything in Simplified Technical English, because I read the Log.
3. When a sub-agent reports, add its report to the ticket's Log. Check its work yourself against the "Done when" list and SPEC.md, then run every check for every ticket that's done so far, so a new ticket hasn't broken an old one.
4. If it passes, mark it done the way STATE.md says and tell me in one line. If it fails, give it to a new sub-agent with what went wrong, and log it. After two failed tries, mark it stuck, tell me why, and carry on with tickets that don't depend on it.
5. Rebuild STATE.md from the ticket files, find the next tickets that can start, and start the next round.

Only stop and ask me when:
- a decision isn't in SPEC.md and would change what I get,
- a ticket is stuck after two tries,
- you need something only I have, like a file, a login or real data.
Decide everything else yourself, follow SPEC.md, and add each decision to "Decisions so far" on STATE.md.

When every ticket is done, run every check in SPEC.md one last time. Then start the app or open the result for me, and tell me in a few sentences what it does and how to try it.
Build one ticket from the tickets folder in this folder. [Ticket number, or leave this out and pick the next one that can start.]

I'm not a developer. Always talk to me in Simplified Technical English: short sentences, one idea per sentence, plain everyday words. If you must use a technical word, explain it in a few words the first time.

Data rules. Public information is fine. That means a published page, like a calendar, price list or set of rules, or a free public API that needs no login and no paid key and only gives out public data, like weather or public holidays. For a page, save a copy to a file in this folder and build from that file. If a page won't load for you, tell me and I'll copy and paste it in. No logins or passwords. Anything private, like work, customer or other people's data, is made up but looks real: realistic names, numbers, dates and amounts, in the same shape as the real thing. Never send private data to an outside service. Where a real system would send or get data, use a file, so a real export can replace it later. If this folder has the security kit (a .claude/skills folder with secure-coding in it), use the secure-coding skill whenever you write code.

1. Read SPEC.md, STATE.md and the ticket, and follow the rules at the top of STATE.md. Say the ticket's number and title before you start. If it's not clear which ticket I mean, ask me.

2. Check it can start: status ready, and every ticket in its "Blocked by" is done. If not, stop and tell me which ticket to build first. If it says "in progress", read its Log and pick up from what's left.

3. Set it to "in progress", add a Log line and update STATE.md. Build only this ticket. Nothing from other tickets.

4. If this is an app:
- Write the test first, at the check point from SPEC.md, the way I'll use it. Watch it fail, then build until it passes. One small piece at a time.
- Run the checks and the tests for what you just changed often as you go. Run every test once at the end.
- When it all passes, review your own work against SPEC.md and the ticket: anything wrong, missing or more than the ticket asked for. Fix what you find.

5. If this is a build with no code:
- Run every "Done when" check with the example inputs from SPEC.md and show me what comes out next to what should come out.
- Review it against SPEC.md and the ticket: anything wrong, missing or more than the ticket asked for. Fix what you find.

6. Add a Log line for each step. Show me how to try it myself in two sentences. When I say it works, mark it done the way STATE.md says.

7. Tell me the next ticket that can start, and tell me to start it in a new session. Then stop.

On their first day, most people sit and watch step 4 run. But step 4 doesn't need you. It might work for ten minutes or forty, and if you watch it, you've turned yourself into a progress bar.

Steps 1 to 3 need you. Step 4 needs Claude. So split your day that way.

Fig 3. Stagger your builds. While Claude builds A, you interview for B. Three apps by the end of the afternoon.

This works because of Part 1. Each build has its own folder and its own desk, so nothing mixes. The files carry everything, so you can leave a build for a day and pick it up in under a minute. Anthropic's guide gives the same advice for parallel work: give each piece its own space so the edits can't collide.

Start with two builds at a time. More builds use your plan's limit faster, so if you hit it, switch the build sessions to a smaller model. By the end of a day like this, you have three apps and the next idea written down.

Something breaks for everyone. When it does, don't type "it doesn't work". Claude can't see your screen, so it guesses, and you get a list of guesses. Give it four things, plus a screenshot:

Then ask: what do you need from me to find the cause? Step 5 has these four lines built in. If the same fix fails twice, stop. Open a new session in the same folder and paste step 7.

Go deeper: quick fixes for the problems everyone hits πŸ”

Something in my build is broken. Help me fix it.

What I did: [the steps I took, in order]
What I expected: [what should have happened]
What happened instead: [what I saw]
The exact error: [paste the error text, or say "no error". Drag in a screenshot if you can.]

I'm not a developer. Always talk to me in Simplified Technical English: short sentences, one idea per sentence, plain everyday words. If you must use a technical word, explain it in a few words the first time.

Data rules. Public information is fine. That means a published page, like a calendar, price list or set of rules, or a free public API that needs no login and no paid key and only gives out public data, like weather or public holidays. For a page, save a copy to a file in this folder and build from that file. If a page won't load for you, tell me and I'll copy and paste it in. No logins or passwords. Anything private, like work, customer or other people's data, is made up but looks real: realistic names, numbers, dates and amounts, in the same shape as the real thing. Never send private data to an outside service. Where a real system would send or get data, use a file, so a real export can replace it later. If this folder has the security kit (a .claude/skills folder with secure-coding in it), use the secure-coding skill whenever you write code.

1. Read SPEC.md and STATE.md first, so you know what this build should do.

2. Make the problem happen yourself before you change anything. For an app, write a test that shows the problem, the way I use it, and watch it fail. For a build with no code, run the same inputs and show me the wrong output next to the right one. If you can't make it happen, tell me what you tried and ask me one question.

3. Tell me in two sentences what is wrong and why. Then fix it. Change only what the fix needs.

4. Run the test for the fix, then every check and test in SPEC.md, so the fix didn't break something else.

5. If the same fix fails twice, stop. Tell me in three lines what you tried, what failed, and two options for what to do next. Let me pick.

6. When it works, add one line to "Decisions so far" in STATE.md: what broke and what fixed it. Show me how to check it myself in two sentences. Then stop.
I want to change my build. The change: [one line about what you want added, removed or changed]

I'm not a developer. Always talk to me in Simplified Technical English: short sentences, one idea per sentence, plain everyday words. If you must use a technical word, explain it in a few words the first time.

Data rules. Public information is fine. That means a published page, like a calendar, price list or set of rules, or a free public API that needs no login and no paid key and only gives out public data, like weather or public holidays. For a page, save a copy to a file in this folder and build from that file. If a page won't load for you, tell me and I'll copy and paste it in. No logins or passwords. Anything private, like work, customer or other people's data, is made up but looks real: realistic names, numbers, dates and amounts, in the same shape as the real thing. Never send private data to an outside service. Where a real system would send or get data, use a file, so a real export can replace it later. If this folder has the security kit (a .claude/skills folder with secure-coding in it), use the secure-coding skill whenever you write code.

Don't build anything yet.

1. Read SPEC.md, STATE.md and the tickets folder.

2. If the change is not clear, ask me up to 5 questions in one round. Number each question and give your recommended answer, so I can say "1 yes, 2 yes, 3 no". Wait for my answers.

3. Update SPEC.md: add or change the user stories, the decisions and how we check it works. Tell me in two lines what changed in SPEC.md.

4. Write the change as new tickets in the tickets folder, the same way as the others: small, with "Blocked by" and "Done when". If a done ticket must change, write a new ticket for it. Don't reopen the old one.

5. Rebuild STATE.md from the ticket files and show it to me. Then stop.

To build the change, start a new session and use "Step 4 - Build all tickets" or "Step 4 by hand - Build one ticket".
I'm back in this folder after a break, or the last session got too long. Pick up where we left off.

I'm not a developer. Always talk to me in Simplified Technical English: short sentences, one idea per sentence, plain everyday words. If you must use a technical word, explain it in a few words the first time.

Data rules. Public information is fine. That means a published page, like a calendar, price list or set of rules, or a free public API that needs no login and no paid key and only gives out public data, like weather or public holidays. For a page, save a copy to a file in this folder and build from that file. If a page won't load for you, tell me and I'll copy and paste it in. No logins or passwords. Anything private, like work, customer or other people's data, is made up but looks real: realistic names, numbers, dates and amounts, in the same shape as the real thing. Never send private data to an outside service. Where a real system would send or get data, use a file, so a real export can replace it later. If this folder has the security kit (a .claude/skills folder with secure-coding in it), use the secure-coding skill whenever you write code.

Don't build anything yet.

1. Read SPEC.md, STATE.md and every ticket. Follow the rules at the top of STATE.md.

2. If a ticket says "in progress", read its Log and work out what is left.

3. If STATE.md doesn't match the ticket files, rebuild STATE.md from the ticket files.

4. Tell me in five lines or less: what this build does, what's done, what's in progress, what's stuck and why, and what can start next.

5. Tell me which prompt to use next: "Step 4 - Build all tickets", "Step 4 by hand - Build one ticket", "Step 5 - Fix something broken" or "Step 8 - Test and get ready to show". Then stop.
My build is close to done. Test it like a new user and get it ready to show.

I'm not a developer. Always talk to me in Simplified Technical English: short sentences, one idea per sentence, plain everyday words. If you must use a technical word, explain it in a few words the first time.

Data rules. Public information is fine. That means a published page, like a calendar, price list or set of rules, or a free public API that needs no login and no paid key and only gives out public data, like weather or public holidays. For a page, save a copy to a file in this folder and build from that file. If a page won't load for you, tell me and I'll copy and paste it in. No logins or passwords. Anything private, like work, customer or other people's data, is made up but looks real: realistic names, numbers, dates and amounts, in the same shape as the real thing. Never send private data to an outside service. Where a real system would send or get data, use a file, so a real export can replace it later. If this folder has the security kit (a .claude/skills folder with secure-coding in it), use the secure-coding skill whenever you write code.

1. Read SPEC.md and STATE.md.

2. Run every check and test in SPEC.md. Tell me which pass and which fail.

3. Use it like someone who has never seen it. Start from nothing and follow the main user stories in SPEC.md, the way a person would. Write down everything that is confusing, slow, wrong or ugly.

4. Show me the 3 worst problems, worst first, with your recommended fix for each. Wait for me to pick which ones to fix.

5. Fix the ones I pick, one at a time. Run every check again after each fix.

6. Write a file called README.md in this folder, in plain words:
- What it does, in one line.
- Who it's for, in one line.
- How to start it or open it, step by step, so I can do it again tomorrow without you.
- What I'd add next, in one line.

7. Start the app or open the result for me. Give me the three lines to say when I show it: what it does, who it's for, what's next. Then stop.

Your app works on your laptop, with made-up data. That's safe while it's yours. Every prompt above carries the same data rule: public data is fine, and anything private is made up but looks real.

But the day a colleague asks "can I use it?", your app will hold real data, and someone could get in who shouldn't. OWASP, a nonprofit that tracks how web apps get broken into, publishes a Top 10 list. The number one risk for 2025 is broken access control: for example, a user changes a number in the address bar and sees someone else's records. Your app won't check for that unless someone told it to.

So after my last build day, I gave everyone a security kit. You copy one folder into your build, and it adds protection in five layers:

Layer What it does When it runs
1. Design Lists who could attack the app and writes the rules it must meet into SPEC.md At the spec step
2. Code Safety rules Claude follows while it writes code Every build step
3. Check Looks for every weak spot, shows you the proof, fixes what you pick Before anyone else uses it
4. Protect A sign-in page in front of the app, and limits that block attackers Before people rely on it
5. Watch A record of who did what, an alert when it breaks, backups you can restore Before people rely on it

The kit also sets hard limits in a settings file. Claude can't read or change files that hold passwords and keys, and it asks before it installs anything or sends data out. Anthropic's own settings docs use the same kind of rule as their first example.

Go deeper: what's in the kit, the settings file and the four skills πŸ”.claude folder inside. Paste "Security 1" and Claude copies it into your build for you.

.claude/settings.json
{
  "permissions": {
    "deny": [
      "Read(./.env)", "Read(./.env.*)",
      "Edit(./.env)", "Edit(./.env.*)",
      "Read(./**/*.pem)", "Read(./**/*.key)",
      "Read(./secrets/**)", "Edit(./secrets/**)",
      "Bash(rm -rf:*)", "Bash(sudo:*)"
    ],
    "ask": [
      "Bash(npm install:*)", "Bash(pnpm add:*)", "Bash(yarn add:*)",
      "Bash(pip install:*)", "Bash(npx:*)",
      "Bash(curl:*)", "Bash(git push:*)"
    ]
  }
}

The four skills. A skill is a set of instructions Claude picks up on its own when the task fits.

Skill What it does
threat-model Asks 6 attacker questions about every way into the app. Writes THREATS.md and a Security section in SPEC.md, as rules a test can check.
secure-coding Rules for sign-in, access checks, secrets, safe database queries, uploads, logs and AI inside the app. Based on the OWASP Top 10 (2025) and the OWASP Top 10 for LLM apps.
security-review Checks 12 areas. Every problem needs proof: the file, the line, and the input that breaks it. Writes SECURITY.md and waits for you to pick what to fix.
production-ready Checks 12 areas you need before people rely on the app, from sign-in to backups. Writes the questions to send IT.

Want more? Trail of Bits, a security firm, publishes free security skills for Claude Code, such as insecure-defaults and supply-chain-risk-auditor. Ask IT before you install anything at work.

Run "Security 2" before anyone else touches the app. It works like the security review Anthropic added to Claude Code on 6 August 2025. For private data, or anything anyone on the internet can open, add "Security 3". It runs a free deep check made by Cloudflare, and Cloudflare is honest about its limits: in their test runs, a single run found roughly half of what repeated runs found.

So the last step needs a person. Before real people rely on your app, have someone who builds software for a living look at it.

Prompt: Security 1. Add the security kit (226 words) πŸ“‹

Add the security kit to this build folder.

The kit folder is here: [drag the folder "2 Security kit - copy into your build folder" into this box]

I'm not a developer. Always talk to me in Simplified Technical English: short sentences, one idea per sentence, plain everyday words. If you must use a technical word, explain it in a few words the first time.

1. Copy everything from the kit folder into this folder. That includes the hidden folder called .claude. It holds the settings and four skills: threat-model, secure-coding, security-review and production-ready.

2. If this folder already has a .claude/settings.json, merge the two. Keep every rule from both.

3. Fill in CLAUDE.md from what you see in this folder: what the app does, the stack, the folders and the commands. Keep it short. Don't add security rules to it. Those live in the skills and the settings. If this folder already has a CLAUDE.md, keep my lines and fill only the gaps.

4. Tell me in 3 lines what you added. Then stop.

The kit doesn't change the data rules. Keep using made-up data for anything private until "Security 2" and "Security 4" are done and someone who builds software for a living has looked at the app.

After this, close the session and start a new one, so Claude reads the new setup.
Review this app for security problems before anyone else uses it. Use the security-review skill.

If the skill isn't in this folder, stop and tell me to run "Security 1 - Add the security kit" first.

I'm not a developer. Always talk to me in Simplified Technical English: short sentences, one idea per sentence, plain everyday words. If you must use a technical word, explain it in a few words the first time.

Don't change any code until I pick what to fix.
Run a deep security audit on this build with Cloudflare's security-audit skill.

Use this after "Security 2", for apps that hold confidential or regulated data, or that are open to the internet. It takes longer than step 09 and uses more of your Claude usage.

I'm not a developer. Always talk to me in Simplified Technical English: short sentences, one idea per sentence, plain everyday words. If you must use a technical word, explain it in a few words the first time.

1. Check that Node.js is installed. If it isn't, stop and tell me.

2. Tell me you're about to install the skill and wait for my yes. Then install it for this folder only:
npx skills add https://github.com/cloudflare/security-audit-skill --skill security-audit

3. Run a security audit of this folder with that skill.

4. Don't run my build's code during the audit unless it runs in a sandbox with no network. If there's no sandbox, mark those findings as "needs a check" and move on.

5. Don't change any code. Write the results to a file called SECURITY-AUDIT.md in this folder.

6. Show me the problems in plain words, worst first, at most 5. For each one, give the fix you recommend in one line. Wait for me to pick. Then fix only what I pick, one at a time, and run "Security 2" again after each fix.
I want real people to rely on this app. Find what's missing, including how to protect it once it's live. Use the production-ready skill.

If the skill isn't in this folder, stop and tell me to run "Security 1 - Add the security kit" first.

I'm not a developer. Always talk to me in Simplified Technical English: short sentences, one idea per sentence, plain everyday words. If you must use a technical word, explain it in a few words the first time.

Don't set up servers or change any systems. Write the list and the questions for whoever runs our systems (IT, or the person who will host it). Then stop.

Write down the next thing you'll build, the day you'll start, and the step it removes. Then make three folders, and paste step 1 into the first one. The one thing not to do: paste passwords or real customer data into a build that hasn't been through Part 6.

If you know someone who loses an hour every week to a task they hate, send them this. The people who'd get the most out of building their own tools are the least likely to hear they can.

── more in #ai-tools 4 stories Β· sorted by recency
── more on @anthropic 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain β€” perfect for shipping the agent you just read about.

$git push zahid main
β†’ Live at https://your-agent.zahid.host βœ“
Get free account β†’ Pricing
from €0/mo Β· no card required
LIVE [news/build-your-own-apps-…] indexed:0 read:44min 2026-11-18 Β· β€”