๐ข๐ป๐ฏ๐ผ๐ฎ๐ฟ๐ฑ๐ถ๐ป๐ด ๐๐ป๐๐ผ ๐๐ฒ๐ด๐ฎ๐ฐ๐ ๐๐ผ๐ฑ๐ฒ๐ฏ๐ฎ๐๐ฒ๐ ๐ช๐ถ๐๐ต ๐๐๐ฟ๐๐ผ๐ฟ
Most AI guides focus on writing code faster. But joining a new project is different. Your main problem is not writing code. Your problem is not knowing how the system works.
Asking AI to analyze a whole project is a mistake. This leads to generic answers. It wastes tokens. It loses context.
You need a knowledge base. Do not ask random questions. Give the AI a specific job.
Create a folder called docs/onboarding/. The main file is project-memory.md. This file stores:
- Repository purposes
- Architecture notes
- Service relationships
- Coding rules
- Important files
This acts as your internal wiki. Add other files for:
- System architecture
- Learning order
- Progress tracking
Save tokens with these steps:
- Read project-memory.md first.
- Analyze only the affected repositories.
- Avoid scanning the whole workspace.
Use these steps for every ticket:
- /ticket: Read memory and make a plan.
- /implement: Modify only required files.
- /review: Check for edge cases.
- /update-docs: Save new knowledge to the memory file.
Stop using AI as a coding assistant. Use it as a project knowledge assistant.
The results:
- Faster onboarding
- Better architecture understanding
- Lower token usage
- Consistent documentation
Do not start by generating code. Start by building knowledge. Map the system. Document the system. Then write code.
Source: https://dev.to/muhammad_usman_dev/how-i-use-cursor-to-onboard-into-massive-legacy-codebases-with-almost-zero-project-knowledge-2lhf Optional learning community: https://t.me/GyaanSetuAi