added .cursor/rules folder and update .cursorrules promts

[skulidropek]

Issue

Towards #57 .

Progress

I updated the prompt system and integrated key components to improve the structure and functionality of the project:

• Added dedicated prompts for TypeScript and Tact
• Included essential contracts from the Tact examples
• Integrated all TEPs from the TON Blockchain TEPs
• Completed three examples from Tact by Example, with more to be added
• Incorporated Jetton implementation from tact-lang/jetton into prompts

[skulidropek]

This is an example, he always starts writing code by viewing the available documentation. I am still finalizing the documentation that he has. If you have useful information, drop it here and I will add it to the agent's knowledge

[novusnota]

LGTM, @i582 @Shvandre would you add/remove anything?

[skulidropek]

LGTM, @i582 @Shvandre would you add/remove anything?

This is not the final version yet. I will let you know when I finish with the prompts. I noticed that the context window has shrunk a lot due to the addition of a large number of prompts.

[skulidropek]

LGTM, @i582 @Shvandre would you add/remove anything?

This is not the final version yet. I will let you know when I finish with the prompts. I noticed that the context window has shrunk a lot due to the addition of a large number of prompts.

I think I'll have to remove most of the unimportant hints. Make them more universal.

[skulidropek]

I think we'll also need to tell the AI ​​about CLI commands that will allow it to browse node module folders to see additional information about the TS and TACT library source code.

ls, cat, tree

[skulidropek]

I think that we should leave information about all standards (tep). About what the source code of Jeton and NFT looks like.

Everything else is not that important information. He can get it by decompiling and thinking in agent mode.

[novusnota]

Sure, I'll mark this as a draft. Also, let's remove this contract.

[skulidropek]

I think we'll also need to tell the AI ​​about CLI commands that will allow it to browse node module folders to see additional information about the TS and TACT library source code.

ls, cat, tree

Add support in System Promt for commands for working with a GitHub repository and a local project.

git log -p -G "hello" --no-merges --pretty=format:"COMMIT: %h | %s%nAUTHOR: %an | DATE: %ad%nDESCRIPTION: %b%n" | grep -C3 "hello"

grep -r "hello" .

git blame

[skulidropek]

I think we'll also need to tell the AI ​​about CLI commands that will allow it to browse node module folders to see additional information about the TS and TACT library source code.

ls, cat, tree

Add support in System Promt for commands for working with a GitHub repository and a local project.

git log -p -G "hello" --no-merges --pretty=format:"COMMIT: %h | %s%nAUTHOR: %an | DATE: %ad%nDESCRIPTION: %b%n" | grep -C3 "hello"

grep -r "hello" .

git blame

For now, the current plan is as follows: All documentation/contracts must be kept in a separate folder. AI has access to work with the project. Access to git changes and search across the entire project.

AI always checks the relevance of its information by comparing data with local folders in which documentation and an example of best contract practices will be located.

Two additional folders will be created: .docs .contrasts

or .information or and it will have subfolders like .docs and .contrasts

For a regular user, these are optional folders, they will serve for the CLI AI to work with the agent mode

[anton-trunov]

@skulidropek btw, not sure if this helps, maybe you can tell: we have some recommendations on gas optimizations for Tact contracts, is it possible to somehow incorporate these: tact-lang/tact#1868?

[skulidropek]

@skulidropek btw, not sure if this helps, maybe you can tell: we have some recommendations on gas optimizations for Tact contracts, is it possible to somehow incorporate these: tact-lang/tact#1868?

Yes, we can write system prompts taking into account these preferences. But we must understand that AI is a probability. If we want to get rid of certain bad practices in the code, it is best to write a static analyzer that will scold the AI ​​for any bad misdeed.