From techinical point of view, most of the lints study Typedtree, so the input for linter should be a fully compiled dune project. There is a support of linters that study untyped tree, but I don't expect them to be widely used. These two abstract trees don't allow to find all possible code quirks. Currently, we delegate job of detecting something like f(1)(2)(3) to ocamlformat.
My experience in teaching OCaml at my local university.
But I'm aware that there are other linters like Camelot for OCaml and Clippy Rust.
Usage
Examples of 'zanuda' usage could be found in the 'tests' directory. But in short:
Compile your dune project and run this linter via
dune build @check @runtest -j3
zanuda -dir .
It will report found issues using OCaml's alerts
You could read the documentation about supported lints via zanuda -dump. CI runs regularly uploads information about available lints to GitHub Pages.
You could run linter and dump the results in short JSON form: zanuda -dir . -ordjsonl /tmp/1.rdjsonl. They could be processed later, for example as review comment via GitHub API. (This reporting is not implemented yet.)
Developping
To get dependecies you could use standard dune/opam machinery: