Macros are expanded by a correct tree walker: the macro expander. It recognizes all special forms, and maintains an environment representing information about the lexical scope (at least in any Lisp dialect worth a damn, with lexical macros that can shadow lexical functions and variables and vice versa).
Macros can also receive that environment as an argument, which provides the possibility that there can be an API by which macros can query the lexical environment.
Furthermore, if macros are given an API for fully expanding an expression, with detailed information about what free references occur in that form, that creates a lot of possibilities for macros to do all sorts of clever things correctly, far beyond just "sticking pieces into a code template with a sprinkling of gensyms".
I was talking about macro-expanding the form first so that the result to be walked consists of only "core" lisp forms.
Do you mean that there are macro-expander APIs in CL that offer all ingredients to do the derivation as part of the macro-expansion step? (I'd welcome some pointer to study.)
The macro expander is a code walker. Whenever it encounters a macro, it expands it (repeatedly, if necessary). Then when all that is left is a special form or function call, or when such a thing is encountered in the first place, it walks that form in the manner appropriate to that form. So by the time macros are expanded, a code walk has taken place over all the special forms and function calls that have emerged.
CL doesn't standardize any API for doing anything with the &environment parameter, other than using it as an argument in macroexpand and macroexpand-1, which is a pity. There is information in there such as "is this symbol a lexical variable in this macro's scope".
Steele's Cltl2 describes such features, but these didn't make it into ANSI CL. It describes accessors variable-information, function-information and declaration-information. Also an augment-environment function and some other things like enclose.
Macros can also receive that environment as an argument, which provides the possibility that there can be an API by which macros can query the lexical environment.
Furthermore, if macros are given an API for fully expanding an expression, with detailed information about what free references occur in that form, that creates a lot of possibilities for macros to do all sorts of clever things correctly, far beyond just "sticking pieces into a code template with a sprinkling of gensyms".