Generate project documentation and prompts for executing code tasks as a way to reduce AI programming tool tokens consumption and task execution errors

Writing the original project documentation prompts

You are my software architect. You will help me write down specific user stories and **functional requirements** based on the project description. Do not provide code.

We will be using a tool called bolt.new - to build this entire project. Imagine bolt.new to be like an LLM - where you give instructions to it, and it will write the code for you.




I need you to be my software architect and help me **write down all functional requirements**. This document will be sent to bolt (an LLM which will write code), so you have to be specific about the functional requirements.

Try to write the requirements as detailed as possible, but if it exceeds 200 words, then split it further into multiple functional requirements (so that you don't overwhelm the LLM).

You should **write only functional requirements** and not include the tech stack needed.

Write translations of prompt words for project documentation

你是我的软件架构师。你将根据项目描述帮助我编写具体的用户故事和**功能需求**。不要提供代码。

我们将使用一个名为 bolt.new 的工具来构建整个项目。可以将 bolt.new 想象成一个大语言模型（LLM）——你向它提供指令，它会为你编写代码。

我需要你担任我的软件架构师，帮助我**编写所有功能需求**。该文档将发送给 bolt（一个会编写代码的 LLM），因此你必须明确具体的功能需求。

尝试尽可能详细地编写需求，但如果超过 200 个单词，请进一步拆分为多个功能需求（以免让 LLM 过于负担）。

你应该**只编写功能需求**，而不要包括所需的技术栈。

Original text of the prompt for the implementation of the code-writing task

## Project Overview

I've uploaded the project file structure in project knowledge - this is what we've built so far. I need you to go through it and understand the complete flow, based on the functional requirements document also uploaded to project knowledge.

## Functional Requirements Components

I have also added the different components of the Functional Requirements into separate files, for you to have more context:
- `<component 1>`
- `<component 2>`
- `<component 3>`

## Development Environment

I am working with `bolt.new` (which is like an LLM which writes the code and executes based on prompts that I give).

## Current Version and Next Steps

I have built the first version of the `<your product>`. Here's what we need to do now:
- [Describe the issue you're facing, or the new functionality you'd want to implement]

## Important Notes

Especially if you're a non-dev and struggle to pinpoint which file causes the issue, in a large project:

Please tell me which files do you need the code to review, from the project structure. I need you to ask me all the info you need, to be able to fix this. We do not want to add new features - we should just fix this issue alone. You need to do a code review and fix the existing implementation, use the current structure, variables used and then tell me how to fix this.

Translation of prompts for performing code-writing tasks

## 项目概述

我已将项目文件结构上传到项目知识中——这是我们目前构建的内容。我需要你浏览它并根据上传到项目知识中的功能需求文档，理解完整的流程。

## 功能需求组件

我还将功能需求的不同组件分成了单独的文件，以便你获得更多上下文：
- `<组件 1>`
- `<组件 2>`
- `<组件 3>`

## 开发环境

我正在使用 `bolt.new`（类似于一个根据我提供的提示生成代码并执行的 LLM）。

## 当前版本和下一步工作

我已经构建了 `<你的产品>` 的第一个版本。以下是我们现在需要完成的任务：
- [描述你遇到的问题，或者需要实现的新功能]

## 重要说明

特别是当你不是开发人员并且在大型项目中难以确定哪个文件导致问题时：

请告诉我你需要查看哪些文件中的代码（根据项目结构）。我需要你询问所有需要的信息，以便解决这个问题。我们不想添加新功能——我们只需要解决这个问题。你需要进行代码审查并修复现有实现，使用当前的结构和所用的变量，然后告诉我如何修复此问题。

Instructions for use

A http://bolt.new tip
When building a complex project, I'll http://bolt.new the Token Usage reduced by 70% (background: my current project has 35 pages of PRD and 16 database tables)!
From: 1M Tokens Processing 3-4 tips
To: The same 1M Tokens can now handle 10-12 tips!

In my experience with http://bolt.new, the key to successful implementation is precise problem solving - knowing exactly what the problem is and how to fix it. As a developer, this ability is enhanced because it's easier to locate the problem and fix it. But if you're a non-developer like me, I've found that it will be Claude Setting up as your "software architect" is the key to achieving this level of precision.

Building on my previous tip about a detailed FRD (Functional Requirements Document), here is the structured system I developed:

File and folder structure in Bolt

Start with a file structure chart. I had bolt create a "http://fileNames.md" that lists each file and maintains a folder hierarchy. Each entry contains a one line description of the purpose and function of that component. This became the map for our project.

Claude Project:

Set up a dedicated "problem solving" project in Claude. I created a Claude project dedicated to handling fixes and updates. In Project Knowledge, I added:

• Complete file structure (from http://fileNames.md)
• Main Functional Requirements Document
• FRD by component (based on user flow)
• Documentation explaining the functionality of http://bolt.new

Streamlining problem solving:

For each fix or new feature, I go into this Claude project and use a specific prompt structure. This is my workflow:

• First, I set the context with my "System Tip".
• Then, for each fix/feature request, I use the "Execution Prompt".

The specific format I use to describe the problem/function can help Claude write optimized tips for http://bolt.new, identify relevant files, suggest the most Token-saving approach, and even provide concrete steps to solve the problem.

Use .bolt/ignore:

I worked with Claude to identify files that didn't need to be in the LLM context and added them to .bolt/ignore. This significantly reduced Token usage while maintaining development efficiency. Note that depending on what you're fixing, we'll need to do this multiple times.

Results?

I actually created a two-tier system:

• Claude acts as a "software architect", analyzing problems and designing solutions.
• http://bolt.new Become a "developer" and implement these solutions efficiently!

This approach has revolutionized my development process. Instead of being stuck with Token restrictions or unclear prompts, I can focus on building and improving features.

Yes, the initial setup will take time. Yes, you'll be challenged with Token limits and error loops. But by choosing to give up when things get complicated, you're missing out on the true potential of http://bolt.new. This structure is worth it in terms of lower Token usage and a clearer development path.

@stackblitz
Features and optimizations are already being released at a breakneck pace - all we need is to find a way to solve most of the problems.

If you would like to implement this method for your project or need clarification, please feel free to reply or private message me.

P.S. Sometimes I still wonder if I should make a video for this setup. Please let me know if this would be helpful to you.