SpamAssassin is invoked from Procmail,
to analyze the content of each email, and insert
headers to summarize the analysis with
a number of keywords, corresponding to a certain
"score". If the sum of all these scores are above
a certain level, the message is flagged with the word
"Yes" in the X-Spam-Status: header.
You may then use the content of the X-Spam-Status: header
in a procmail recipe to decide if the mail should go into your default inbox,
be stored in a low-priority folder, or written to /dev/null
to discard it completely.
You can change the level that triggers the Yes-flag by changing the value
assigned to the "required_hits" keyword, in a file named
.spamAssassin/user_prefs (note the leading dot).
If you don't agree with some of the default scores that SpamAssassin uses,
you can assign any value you choose by adding lines like:
score TEST_NAME x.yz
where TEST_NAME is a symbolic name from the X-Spam-Status:
header, and x.yz are numeric values you want it to use instead.
The X-Spam-Status: header also includes a flag to indicate if
it has used the email in its "Bayesian" database of patterns that
may categorize the message as spam (or non-spam). If you notice a
"autolearn=no" in this header, you may wish to feed
the message manually to the Bayesian learner.
References: