psycopg2 gets confused by % characters and strict pyformat mode

I've got a script that takes SQL input from the user, then tacks
something on at the end and executes it. 

If the final execution looks like this:

cursor.execute("select * from t where s like '%whatever%' and k =
%(something)s", {'something' : 'else'})

psycopg2 throws an exception

<type 'exceptions.TypeError'>: 'dict' object is unindexable

here, the script added the k = %(something)s part, and the parameters.

By trial and error, and partial reading of the source code, I discover
that doing something like:

cursor.execute("select * from t where s like '%%whatever%%' and k =
%(something)s", {'something' : 'else'})

works fine, even if it is ugly.  Also, if the parameters are not passed
in, then it doesn't convert the doubles to a single:

ie
>>> cursor.mogrify("select * from t where s like '%%whatever%%'")
"select * from t where s like '%%whatever%%'"

This isn't too much of a problem since it will match whatever was
matched before, but it is even uglier.

Is there a way to force psycopg2 to only take into account pyformat
variables and to leave the other %'s alone?  If not, would you accept a
patch to do this?