But the other day I was thinking about using the tools.trace library, and realized it was a bit of a drag that you have to remember to load the code up front and then remember the exact invocation to enter in the repl to enable tracing on a given defn. It's not much, but if there's friction in between being in the zone and enabling a tool like that, you're likely to just fall back to printlns. I was looking through what it would take to toggle the tracing directly from Emacs, but at the time I wasn't really in the mood for writing a bunch of elisp, especially not if it had to be repeated for every command you'd want to add support for in nrepl.el. The worst part was that if the elisp needed to invoke any server-side code, it had to be embedded in the elisp code, usually as strings.
This got me thinking about whether we could come up with a way to make commands self-describing in such a way that the editor (whether Emacs or another) could construct the appropriate commands automatically. I ended up putting together a proof-of-concept which annotated tools.trace such that it could be invoked directly from Emacs via M-x nrepl-toggle-trace or bound to a key combination by the user. When I found that posed little difficulty I went on to extend it to add a command to run tests from clojure.test as well in a way that could mostly deprecate clojure-test-mode.
The way it works is that on the server side vars are annotated with :nrepl/op metadata that describes the command's name, documentation, and arguments. Then an initial discovery endpoint is provided which can tell the client about all known ops. In my proof-of-concept, Emacs uses this data to construct elisp defuns which prompt the user for the arguments, often in ways involving fancy completion schemes. The results can be displayed either as a simple message or as a number of other richer formats. I've described the mechanism in a slightly more formal proposal here, which I hope could be useful to others wanting to annotate their own development tools or by maintainers of the Clojure tooling for Vim, Eclipse, etc.
If you've got some piece of functionality you'd like to expose to users directly in their editor, please give it a shot. There's probably more discussion that needs to happen around the fancier response types as well as providing implementations for other clients. There's some discussion on the clojure-tools mailing list where you can chime up with suggestions or notes on how it's worked for you.