Naser Rafinia

Senior software developer

[Naser Rafinia @nRafinia]

Navigating .NET at Scale: How csmesh Solves Indirection, Dependency Injection, and Call Graph Discovery

https://t.co/sWpmgYUwpE

[Naser Rafinia @nRafinia]

@mjovanovictech https://t.co/B4Ho27ZhcT

[Naser Rafinia @nRafinia]

@mjovanovictech Your coding agent burns ~4,000 tokens answering one structural question and almost none of it is reading logic. It's grep, wrong file, grep again, then scrolling Program.cs for the DI registration

csmesh collapses that into one command 94ms. 73 tokens

[Naser Rafinia @nRafinia]

@vrapolinario Your coding agent burns ~4,000 tokens answering one structural question and almost none of it is reading logic. It's grep, wrong file, grep again, then scrolling Program.cs for the DI registration

csmesh collapses that into one command 94ms. 73 tokens

https://t.co/B4Ho27ZhcT

[Naser Rafinia @nRafinia]

@dotnet Your coding agent burns ~4,000 tokens answering one structural question and almost none of it is reading logic. It's grep, wrong file, grep again, then scrolling Program.cs for the DI registration.

csmesh collapses that into one command. 94ms. 73 tokens.

https://t.co/B4Ho27ZhcT

[Naser Rafinia @nRafinia]

@nickchapsas https://t.co/B4Ho27ZhcT

[Naser Rafinia @nRafinia]

Your agent didn't burn 5,000 tokens understanding your code. It burned them LOOKING for it.

https://t.co/B4Ho27ZhcT

[Naser Rafinia @nRafinia]

@davidfowl Your agent didn't burn 5,000 tokens understanding your code.

It burned them LOOKING for it.

And swapping grep for a "real" tool doesn't fix it. Call hierarchy, LSP references, tree-sitter graphs — they all stop at the same wall.

https://t.co/B4Ho27ZhcT

[Naser Rafinia @nRafinia]

@githubstatus https://t.co/Fmo2my0EL5

[Naser Rafinia @nRafinia]

@mjovanovictech https://t.co/iXh5Zg9oHf

[Naser Rafinia @nRafinia]

Your agent didn't burn 5,000 tokens understanding your code.

It burned them LOOKING for it.

And swapping grep for a "real" tool doesn't fix it. Call hierarchy, LSP references, tree-sitter graphs — they all stop at the same wall.

The wall is indirection. 🧵

#csharp #dotnet

[Naser Rafinia @nRafinia]

@davidfowl https://t.co/iXh5Zg9oHf

[Naser Rafinia @nRafinia]

Your agent didn't burn 5,000 tokens understanding your code.

It burned them LOOKING for it.

And swapping grep for a "real" tool doesn't fix it. Call hierarchy, LSP references, tree-sitter graphs — they all stop at the same wall.

The wall is indirection. 🧵

#csharp #dotnet

[Naser Rafinia @nRafinia]

dotnet tool install --global CsMesh
csmesh index
csmesh skill --install

Last line writes native rules for Claude Code, Cursor, Windsurf, Cline, Copilot, Codex, Gemini. Your agent starts reaching for it unprompted.

C#/.NET. Native AOT. MIT.
https://t.co/B4Ho27ZhcT

[Naser Rafinia @nRafinia]

Your agent didn't burn 5,000 tokens understanding your code.

It burned them LOOKING for it.

And swapping grep for a "real" tool doesn't fix it. Call hierarchy, LSP references, tree-sitter graphs — they all stop at the same wall.

The wall is indirection. 🧵

#csharp #dotnet

[Naser Rafinia @nRafinia]

The [di-bound] tag is the point. Two classes implement that interface; one is a test fake. csmesh ranks the one the container actually resolves.

And --budget is a hard cap. Too big? Exit 2 with a narrower query suggestion. It never floods your context.

[Naser Rafinia @nRafinia]

@StatsWire https://t.co/41hfDjNgig

[Naser Rafinia @nRafinia]

@Mr_Salio https://t.co/41hfDjNgig