Sieve Email Filtering: Include Extension
RFC 6609

I think it would be helpful if you spent more time explaining what you
mean by recursion. Direct recursion is (of course) a lot easier to spot
than general (by menas of indeirections) recursion. But it is the 
general case you need to protect against and that is potentially 
considerably more complicated for an implementation to police. Since
the consequences are just as bad, and since it may be harder for the 
coder to realise that they have transgressed, I think you should call it
out explicitly.

Perhaps add a definition of recurssion to Section 2, and then note the
implication for tracking all nested script names in 3.1

Section 3.1 states:

   An error MUST be generated when such a script is executed.

However, Section 3.2 states:

   Implementations MUST NOT generate an error if an "include :once"
   command names a script whose inclusion would be recursive; in this
   case, the script MUST be considered previously included and therefore
   "include :once" will not include it again.

These two are in conflict. I suggest changing the text in Section 3.1 as follows:

   An error MUST be generated when such a script is executed, except 
   as noted under Section 3.2.

In Section 3.2:

   The ":optional" parameter indicates that the script may be missing.

I'd change "may" to "MAY" or (to avoid confusion with "might be missing") say "is allowed to be absent" or somesuch.

Section 3.4.1 states:

   If a "global" command is given the name of a variable that has
   previously been defined in the immediate script with "set", an error
   MUST be generated either when the script is uploaded or at execution
   time.

Does that impose an ordering requirement on script uploads?

1) s3.2: Contains the following:

   The "include" command takes an optional "location" parameter, an
   optional ":once" parameter, an optional ":optional" parameter, and a
   single string argument representing the name of the script to include
   for processing at that point.

Shouldn't the "optional"s be "OPTIONAL" to ensure they match they ABNF?

2) s3.2: Contains the following:

   Note: It is RECOMMENDED that script authors / generators use the
   ":once" parameter only when including a script that performs general
   duties such as declaring global variables and making sanity checks of
   the environment.

Can you rephrase this so the requirement isn't in a "note"?

Is SIEVE turning into a fully-fledged programming language? 
I guess I wonder if that's happening by design or by accident.
(Not that I know which would be better;-)