ChatGPT “Project” feature practice notes

This page organizes how to incorporate the ChatGPT-4o/o3 series “project” function into a problem-solving workflow centered on Python development. This is a memo that serves as both a personal memo and a blog article, from an overview of functions to model selection, handling of large repositories, limitations and countermeasures.

What is the project function?

• A “work room” that bundles chat history, files, and custom instructions into one theme

• Share uploaded documents to all chats within the same project

• Available for paid plans of Plus/Teams and above (as of 2025-06-04)

Utilization scenario

• Research Memo – Includes paper PDF + code snippets and weekly Q&A

• Repository improvement – Post Git diffs and discuss refactoring strategies

• Event Planning – Consolidate spreadsheets and meeting minutes in one place

Model selection guide

model comparison

model	Strengths	Assumed task
GPT-4o	Fast, cheap, images/audio available	Specifications PDF + Blueprint review
O3	Top class inference accuracy	Difficult bug analysis, mathematical optimization
o3-mini / o4-mini	Light and fast	Standard generation, CI script

Actual operation sample

1. Initial investigation → GPT-4o

2. If it gets stuck → switch to o3

3. YAML Small items such as mass production → mini series

“Latest source” synchronization technique for huge projects

There are only three sources of information that ChatGPT can reference: (1) Chat history, (2) Upload files, and (3) URLs that are explicitly loaded. As the scale expands, “how to give” becomes more important.

sync pattern

• Full ZIP regular uploads – Entire git archive output once a week

• Incremental diff – Paste git diff main..HEAD for each PR

• Summary Document – Condenses the responsibilities of major modules into a single Markdown sheet

• Automatically generated metadata – Attach dependency graphs and ERDs as SVG

Recommended operation flow

Operational template

timing	action	the purpose
When merging PR	Let ChatGPT summarize the diff and add summary.md	“Brain Wiki” updated
weekly	Resynchronize main with ZIP	Removing leaks/updating cache
When starting a new chat	summary.md + latest diff first attached	Initializing a new context
When the function completes	Regenerate diagrams such as “Blueprint with PlantUML”	Visualization document preparation

Example prompt for comprehension test

あなたが現在把握しているフォルダ構造と各モジュールの責務を
300 words 以内で要約して下さい。

If the response is outdated, resubmit the additional file at that point.

Limitations and precautions

• Token cap – above 128k, naturally falls off from older history → compress information in summary layer

• GitHub deep research is Read-Only – Does not follow automatically → Prompts re-fetch after changes

• Security – Secret code must be mocked, API key must be managed as Secret

• Humans are ultimately responsible – Always commit and merge at hand CI/review

summary

1. Project = “Workspace for AI”. Organizing materials and instructions increases reuse efficiency.

2. The model is based on GPT-4o and uses o3 as a trump card.

3. Synchronization scales with full ZIP+diff+summary cascade.

4. Always have prompts to check for understanding and update immediately if gaps appear.

With this setup, you can make ChatGPT useful even in a large codebase while keeping track of how much context the AI actually has.

Article information

author:

Mr. Takagi

Release date:

2025-06-04

Consideration of introducing Markdown and decision of continuing reStructuredText   Steps to rebuild a broken WSL environment after using Rancher