So by default it pulls out recent PRs, grabs only the tests that were commited and then checks to see how well an agent with different claude.mds can finish the test suite?