My apologies if this has been discussed here before and eluded me.
I'm trying to come up with a system to give the ability to use sa-learn to users who do not have shell access to their accounts on our mail server. My initial thought was to have users forward spam and false positives to two per-user aliases ([EMAIL PROTECTED] and [EMAIL PROTECTED] ) and give everyone a .procmailrc file that runs sa-learn as appropriate for email forwarded to the aliases.
This seems relatively easy to accomplish, but my concern is that sa-learn will not learn properly because the message it receives is not the original, but a copy forwarded by the user. Will sa-learn conclude that the user is the spammer because the From: line of the forwarded message has the user rather than the spammer?
If so, I've thought of a possible workaround described below:
1) append a special message ID number to the body or custom header of each incoming email message if it is small enough to be considered spam (under 100Kb or so)
2) save a copy of each message that is tagged with an ID number into a temp directory on the server (and keep it around for 1-2 weeks).
3) write a procmail script that will check for mail addressed to the aforementioned user+spam and user+falsepositive aliases.
If it finds one, retrieve the original message out of the temp directory and run sa-learn on that, then delete the message from the temp directory.
If this seems overly complicated, I welcome suggestions on how to improve this. Am I mistaken by thinking that sa-learn will get confused by spam that is forwarded to it by the end user?
TIA, Ryan
