> It's straightforward to implement e-d-e on top of the existing SRFI,
> so I think it belongs in a different SRFI or in application code.

Fair enough. That's true.

> For example, should it keep track of which directories have been
> created, and then if there is a failure further down, remove them again?

Good point. As far as I know, the equivalents in other languages don't
clean up after failure. If directory A can be created, there isn't much
reason for the creation of A/B, A/B/C, etc. to fail since anyone who
successfully creates a directory controls that directory.

There can still be situations like I/O error, space/quota exceeded,
depth exceeded, etc. but those are getting quite esoteric. Cleanup
definitely wouldn't pass an 80/20 test.

>> Sounds good to me, but it definitely looks out of scope.

Defining it in another SRFI is ok. But which one?

This is different from a "Unix shell emulator" which would have
procedures like (cd), (pwd), (ls) and (mv). e-d-e is useful in ordinary
applications and probably even in libraries.