Entering an invalid traphandle program doesn't immediately fail. Nor
does it fail on the first trap received:
pipe([12, 13])                          = 0
pipe([14, 15])                          = 0
fork()                                  = 16528
close(12)                               = 0
close(15)                               = 0
write(13, "rtr\n10.10.10.10\nRFC1213-MIB::s"..., 327) = 327
close(13)                               = 0
waitpid(16528, [{WIFEXITED(s) && WEXITSTATUS(s) == 1}], 0) = 16528
close(14)                               = 0

But eventually snmptrapd will die:
pipe([12, 13])                          = 0
pipe([14, 15])                          = 0
fork()                                  = 16529
--- SIGCHLD (Child exited) @ 0 (0) ---
close(12)                               = 0
close(15)                               = 0
write(13, "rtr\n10.10.10.10\nRFC1213-MIB::s"..., 327) = -1 EPIPE
(Broken pipe)
--- SIGPIPE (Broken pipe) @ 0 (0) ---
+++ killed by SIGPIPE +++

Steps to reproduce: I tried entering a handler that normally could be
found via the PATH.

Workaround: Explicitly enter the full path of the trap handler. And
pray that your handler script doesn't b0rk or it will take snmptrapd
with it.