The condition hierarchy tree doesn't show &i/o-operation-error and its
descendant &i/o-operation-not-available-error. They are specified
between &i/o-malformed-filename-error and &i/o-file-protection-error, so
you might also want to change the order of them in the specification.

The layer-specific condition types are also not included in the list.
This may be intentional.

file-options-include? is specified to return true or #f, not #t or #f as
  the other predicates in this SRFI (the other predicates are type
predicates).

reader? exists but writer? does not exist.

reader-close and writer-close are not specified to take any arguments.

The documentation of make-i/o-buffer refers to a function names
writer-write!, but the specification specifies a function named
writer-write!! (with two exclamation marks).

open-file-reader+writer is not specified to return results equivalent to
open-file-reader and open-file-writer. Is this intentional? Also, there
is a typo in the third statement of the definition: "This writer and the
writer may ...".

More documentation on open-u8vector-writer and writer-u8vector is called
for. At the very least, create a little section that states explicitly
that the created writer uses a private buffer that is unshared with
writer-u8vector's result (if these are indeed the semantics).

This is a great I/O framework. I wish I knew more, and could comment on
the real, technical issues. That said, I'm a little worried about the
octets vs. bits decision -- there is a use for bitwise I/O.

- Amir