juju is apparently silently reading *.pem files in the CWD and then aborting on them:
Version: 0.5.1+bzr574-0juju2~quantal1
steve@dogmatix:~/juju$ rm *
steve@dogmatix:~/juju$ openssl genrsa -out rabbit-server-privkey.pem 2048
...
steve@dogmatix:~/juju$ openssl req -batch -new -x509 -key rabbit-server-privkey.pem -out rabbit-server-cert.pem -days 10000
steve@dogmatix:~/juju$ juju -l /dev/stderr bootstrap
2012-09-14 15:40:16,104 WARNING S3 API calls not using secure transport
2012-09-14 15:40:16,106 INFO Bootstrapping environment 'canonistack' (origin: distro type: ec2)...
2012-09-14 15:40:17,445 ERROR [('PEM routines', 'PEM_read_bio', 'no start line')]
steve@dogmatix:~/juju$ rm *.pem
steve@dogmatix:~/juju$ juju -l /dev/stderr bootstrap
2012-09-14 15:40:33,087 WARNING S3 API calls not using secure transport
2012-09-14 15:40:33,088 INFO Bootstrapping environment 'canonistack' (origin: distro type: ec2)...
2012-09-14 15:40:43,821 INFO 'bootstrap' command finished successfully
This is actually txaws doing so, and not juju directly. I'm not sure its a sane behavior for a library, but would like to perhaps have some commentary from txaws developers before patching to outright disabling it. Seems like, if nothing else, the API should allow overriding this behavior, since its possible that txaws code could be run from a hostile CWD.