@danielbachhuber, @gitlost The above test was actually meant to fail to have a regression test for #44 .

However, it seems like the Phar that is built to be deployed is different than the Phar that is used when built during the tests. Shouldn't these two scenarios give the exact same results? At what point (and in which workflows) is the Phar stripped for size (as I assume this casues #44)?

The phar built by FeatureContext::build_phar() is totally fake really, particularly when not run from wp-cli, as it just runs utils/make-phar.php on the current repo directory. To make this work in the command repos, this https://github.com/wp-cli/wp-cli/blob/v1.4.0/utils/make-phar.php#L183 is used, which was originally added in wp-cli/wp-cli#3923, temporarily removed in wp-cli/wp-cli#4185, and added back in wp-cli/wp-cli#4257 when it was discovered that the package-command tests were failing in #22 (comment).

So as package-command's composer.json doesn't include the phpcs stuff itself and Composer doesn't recursively add them (it seems - I hesitate to say anything definite about Composer), they aren't in the vendor dirs and the autoload*s in package-command won't mention them.