I think this could be difficult for Bors. For example, if a batch with two or more pull requests fail, how does Bors know whether it should auto-retry the batch or proceed to bisect the batch?
An easier approach will be for you to implement the auto-retry logic in your test script. Here’s an example (taken from here) of a bounded auto-retry loop. You would replace npm test with the test command(s) relevant for your project.
#!/bin/bash
sleep 1
try_count=0
while [ $try_count != 100 ]; do
npm test
if [ $? = 0 ]; then exit 0; fi
try_count=`expr $try_count + 1`
sleep 1
done
exit 1
One thing that has helped me with nasty CI setups (WebDriver, in my case) is to wrap the test suite in a bounded retry-loop. If your tests are flaky, that's probably a good idea no matter what you do with bors.