Agent doesn't fix Linter errors

It seems like in older versions of Cursor this worked. I would write “fix errors” or “linter errors” to it, and it would make corrections. Now, when I write that, it starts guessing what the error might be. I have to copy the error messages manually. Has something broken?

I want to emphasize that I’m talking about situations where I tell the agent about the errors, not automatic recognition (that doesn’t work for me because I need to rebuild the project for some errors to disappear).

In other words, when I write “fix linter errors” to it, it has stopped seeing them, whereas before it did.

In recent versions, we’ve made the Linter fix a more formal step in Composer, which you can enable in the settings.

We found that providing linter errors as context usually broke more than it fixed, but doing it as a standalone step once all other changes were made worked a lot better!

Give these settings a go, and see if that resolve things for you:

As I said before, automatic iterations don’t work for me because I need to rebuild the project for some errors to disappear.

What do you mean by “as a standalone step” - how can I ask for it as a new step when the agent doesn’t see Linter errors?

I was ignorant to linting in general so I went on a small path to understand how the agent does linting.

You may find thi thread I had with Composer interesting with respect to guiding it to the linting behavior you want:

I think this highlights a few things Ive mentioned in thread here on this forum WRT better and much more enhanced config/behavior mgmt of Cursor.

Whereby, there would be benefit in a workspace, with a standard project structure with standard file defenitions for cursor logging, context tracking, linting rules… Basically .cusorrules on steroids that isnt just a files, but it points to a repo/directory of style_guide.rmd, development_diary.json, db_diary.rmd, etc…

And as I have been directing my agents to begin using postgres for source of truth, docs, style_guide, logging, persona directives etc.

Does the cursor team feel the need for such? How does Cursor Team do it internally (managing of their agents)

WHat multi-agent tooling can you share - either direct code, files or even best practices, pitfalls etc?

Again - I also ask that my agents have full access to my profile usage stats for all calls and resources it touches/consumes/depends/requires… may this be accomplished?