I think you would need to do a content analysis with a rubric and multiple coders to achieve good inter-coder reliability.
I use rubrics adapted from the 6+1 Traits model developed by the Northwest Regional Educational Laboratory. Attached is a document with their original model for evaluating writing, and an example of how I adapted their model to oral proficiency.
I have taught college-level writing courses that are specifically designed for ELs (both domestic and international in the U.S.) and I used the rubric attached. I explain to my students what each category is about and usually do a lot if modeling to illustrate what is considered good academic writing in college.
In general, when it comes to evaluating one's writing skills, it cannot be done with a test. it is best when learners are asked to produce a text and then two or three readers/raters (who are normed) assess it. Also, one writing sample from a students does not show the whole picture. if possible, use portfolio assessment.