CJ Zafir Profile picture
Dec 31, 2024 11 tweets 5 min read Read on X
Bolt .new Prompting Handbook:
🧵 Image
1. Use "Enhance prompt" button

Your prompt quality makes a huge difference in AI output.

Provide as much details as you can in the input field, then click on "enhance prompt" button.

A detailed and will structured prompt can help you save tokens and time with less errors. Image
2. Use this structure for starting a NEW project:

“I need a [type] application with:

Tech Stack:
- Frontend: Nextjs 14 with TypeScript using App router
- Styling: Tailwind CSS
- Auth: Supabase
- Database: Supabase

Core Features:
- User authentication
- [Main feature]
1.
2.
3.
4.
- [Secondary features]

Start with the main page containing: [Detailed page requirements]”
3. Turn ON diffs

Whenever you ask bolt to make a specific change on any file. Bolt used to rewrite the code for entire file.

This burns alot of tokens.

To solve this issue: Go to Settings → Preview features → Turn ON the toggle button for Diffs

After this bolt will make changes to only few lines of code.

Results: Fast Loading & Less Token BurnImage
4. Use "Lock File" and "Target File" Feature

If you want to make changes to only one file than use "Target file" feature by right clicking on the directory.

You can also "lock" files that are done and you don't want bolt to make any unwanted changes.

This will help you minimize token burn and AI hallucinations.
5. Build a Context Boundary

Provide detailed context about your project: What it is all about? User flow? Tech stack? Scope of work? etc

I call it the "Context Boundary."

Buildin a strong boundary around AI so it works for you and just follow the plan.

For this you need to attach your projects docs, like:
- Project Requirements Doc (PRD)
- App Flow Doc
- Frontend Guidelines
- Backend Structure

After you start the project with initial prompt. attach PRD in .md format and ask:

"before you code anything, go through the PRD and tell mw what do you understand about my project?"Image
6. Attach Docs to File Directory

You can also create an"Instructions" folder and create docs files in .md format under that folder.

Then paste the content in each file and save.

Then right click on Instructions folder and hit "Target all."

Then ask bolt, what do you understand about my project.

And when working on specific requests, you can tag files by using "copy path" of that file and pasting in input fileld.

Then ask: "Refer to (eg: file path) doc first and then help me build x,yx,z"Image
7. Write Detailed Instructions

While building you need to instruct bolt with detailed requests.

Instead of "move the button to right" say "in top header move the sign up button to left side of the page"

Use can use "enhance prompt" button throughout the building process.

The better the instructions, the less the errors.
8. Build Step-by-Step.

Don't tell bolt to do 5 tasks at same time. This will force AI to mess up things.

Here's how I do it:

→ Start with Frontent design, page by page, section by section.

→ Then attach backend using Supabase as bolt has supabase integration built in.

→ and then I refine the UI/UX if needed.

This step by step flow help AI to focus on 1 task at a time.
9. Enjoy 15% OFF on all Bolt plans.

I am happy to announce that we made an exclusive partnership with bolt @stackblitz to offer @CodeGuidedev community a 15% Discount on all of its plans. 🎉

If you haven't joined CodeGuide yet. I am offering 30% discount on Yearly membership for next 36 hours.

More deals and perks are coming for you.Image
CodeGuide is becoming the Home of AI coders.

Dump your project idea and CodeGuide will help you write your Project Development Documentation for AI models.

Join us here: codeguide.dev

+ CodeGuide 1.2 is coming. ✨
→ CodeGuide in-house community (No Discord)
→ Video tutorials from Jan 2nd
→ Exclusive reports on "What to build."
→ Prompt templates, Boilerplates, Design templates.
and few surprise features.

Hang tight. We are just 8 days old startup. 🙌

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with CJ Zafir

CJ Zafir Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @cjzafir

Aug 16, 2025
Introducing Codespace

A virtual computer that runs custom Claude Code on server.

It is using 2 models: Kimi K2 + Claude Sonnet 4 in a sync.

I achieved same code quality with 2x better reasoning and 53% less model cost.

Access it today @CodeGuidedev
(500 credits for everyone)
We fixed some core issues:
1. It works with new and existing codebases and it won’t touch your existing code.

It can scan large codebases, 10M tokens (250,000 lines of code) and remember all core details.

Just attach your GitHub Account and select the repo you want to work on. Image
Image
2. It works autonomously (you don’t need to hit accept all) everytime.

It generates 7 docs to build a solid knowledge base to refer and create detailed tasks/sub-tasks list on autopilot.

Resulting in 88% less hallucinations and better code structure. Image
Read 8 tweets
Aug 14, 2025
Every AI startup is focused on "Build me an ........... app" (New projects)

But what about millions of existing codebases on GitHub with billions of lines of code?

To focus on existing codebases, I built a coding environment: Codespace.

(Explained below)
Simple Steps:
1. Connect your GitHub Account
2. Open Codespace and select the repo you want to work on. (AI agent will make an isolated environment for every request)
3. Let AI agent Analyze, generate docs and tasks list
4. analyze the kanban board to track tasks
5. Go have coffee and wait for the agent to email you when done.
I built a Claude Code wrapper that:
- runs on cloud (access it with phone, or laptop)
- can analyze upto 10M token size codebases
- has better visual UI & flow
- generate docs (knowledge base)
- tasks/sub-task list (2x better than claude)
- notifies me when its done via email
Read 5 tweets
Jun 27, 2025
I built "Clade Tasks" to make Claude Code 10x better and 2x cheaper.

I integrated 3 engines in 1 flow:

1. Claude Code (best agentic coding interface)
2. CodeGuide (to provide deep context to AI)
3. Taskmaster (to create detailed implementation plan)

Here's how it works: ↓
Claude Code is the most powerful AI coding tool right now.

Yes better than Cursor/windsurf.

But Claude 4 (the underlying model) can spit out unnecessary files/code and burn tokens.

Also it's plan mode is good but I found a better alternative: ↓
1. Plan before code

"Build me x, y, z." type prompts don't work because of no technical context.

So turn your idea into a set of well structured docs:
- PRD
- Tech stack doc.
- App flow doc
- UI guidelines
- backend structure
etc

@CodeGuidedev write these docs in only 3 steps Image
Read 7 tweets
Jun 15, 2025
The most important document in AI coding: PRD
(Project Requirements Document)

It serves like a blueprint & explains:
- project's overview
- tech stack
- user flow
- core features
- ui details
- backend schema
- security guidelines
- regulations

Here's how you can create PRDs:
Before you create a PRD, brainstorm with ChatGPT voice.

This step will help you finalize what features you want in your version 1.

In the end ask ChatGPT: "now add all approved features in one document and create a detailed draft)

Now it's time to create a PRD.
I use @CodeGuidedev to create all my coding docs including PRD

It uses 03 + 04-mini + Sonnet 4 model to create extensive set of docs

Flow is simple

Step 1: Open "New Project" and paste the feature draft in the input field

Step 2: Now select AI tools & AI models you'll use
Read 8 tweets
Apr 18, 2025
What are AI Agents?

OpenAI just published 30-page guide on AI agents.

I explained the guide in 10 points. Bookmark this: Image
1. Agents ≠ basic workflows

A normal “workflow” just runs a fixed list of steps with an LLM; it doesn’t think for itself.

On the other hand, Agents are systems that independently accomplish tasks on your behalf.

Workflows are linear. Agents are dynamic & autonomous. Image
2. Use agents for the complex jobs

Agents handle changing rules, tricky choices, and unstructured data better than any fixed script.

Pick jobs that need judgment, change a lot, or rely on messy text, images, or PDFs. Image
Read 12 tweets
Apr 10, 2025
I tested Gemini Pro 2.5 as my main coding model for 40+ hours.

Here're 2 documents that are working brilliantly well with Gemini.

"App flow document + App flowchart."

This made my Cursor workflow 10x better. Here's why it is working: ↓
Gemini Pro 2.5 has a 1 million token context window.

In simple terms it can scan upto 100,000 lines of code instantly to understand full codebase.

Claude & GPT models has 200k context window and that's why they struggle with large codebases.
But large context window doesn't mean Gemini doesn't hallucinate or assume things.

That issue is still there but I'm able to fix hallucinations by 95% in Gemini with my 'Context Boundary' technique.

You need to provide context with coding docs:
- PRD
- App flow doc
- Tech stack doc
etc

But here's how 2 docs are working really good with Gemini.Image
Read 7 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us!

:(