diff --git a/images/prow-tests/Dockerfile b/images/prow-tests/Dockerfile
index 5b55c56bd75..a78fa8bbbec 100644
--- a/images/prow-tests/Dockerfile
+++ b/images/prow-tests/Dockerfile
@@ -28,3 +28,4 @@ RUN go get github.com/golang/dep/cmd/dep
 ADD scripts/library.sh .
 ADD scripts/presubmit-tests.sh .
 ADD scripts/release.sh .
+ADD scripts/e2e-tests.sh .
diff --git a/images/prow-tests/scripts/e2e-tests.sh b/images/prow-tests/scripts/e2e-tests.sh
new file mode 100755
index 00000000000..bfbda491f65
--- /dev/null
+++ b/images/prow-tests/scripts/e2e-tests.sh
@@ -0,0 +1,255 @@
+#!/bin/bash
+
+# Copyright 2018 The Knative Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This is a helper script for Knative E2E test scripts. To use it:
+# 1. Source this script.
+# 2. Write the teardown() function, which will tear down your test resources.
+# 3. [optional] Write the dump_extra_cluster_state() function. It will be called
+#    when a test fails, and can dump extra information about the current state of
+#    the cluster (tipically using kubectl).
+# 4. Call the initialize() function passing $@ (without quotes).
+# 5. Write logic for the end-to-end tests. Run all go tests using report_go_test()
+#    and call fail_test() or success() if any of them failed. The envitronment
+#    variables DOCKER_REPO_OVERRIDE, K8S_CLUSTER_OVERRIDE and K8S_USER_OVERRIDE
+#    will be set accordingly to the test cluster. You can also use the following
+#    boolean (0 is false, 1 is true) environment variables for the logic:
+#    EMIT_METRICS: true if --emit-metrics is passed.
+#    USING_EXISTING_CLUSTER: true if the test cluster is an already existing one,
+#                            and not a temporary cluster created by kubetest.
+#    All environment variables above are marked read-only.
+# Notes:
+# 1. Calling your script without arguments will create a new cluster in the GCP
+#    project $PROJECT_ID and run the tests against it.
+# 2. Calling your script with --run-tests and the variables K8S_CLUSTER_OVERRIDE,
+#    K8S_USER_OVERRIDE and DOCKER_REPO_OVERRIDE set will immediately start the
+#    tests against the cluster.
+
+# Load github.com/knative/test-infra/images/prow-tests/scripts/library.sh
+[ -f /workspace/library.sh ] \
+  && source /workspace/library.sh \
+  || eval "$(docker run --entrypoint sh gcr.io/knative-tests/test-infra/prow-tests -c 'cat library.sh')"
+[ -v KNATIVE_TEST_INFRA ] || exit 1
+
+# Test cluster parameters
+readonly E2E_BASE_NAME=$(basename ${REPO_ROOT_DIR})
+readonly E2E_CLUSTER_NAME=${E2E_BASE_NAME}-e2e-cls${BUILD_NUMBER}
+readonly E2E_NETWORK_NAME=${E2E_BASE_NAME}-e2e-net${BUILD_NUMBER}
+readonly E2E_CLUSTER_ZONE=us-central1-a
+readonly E2E_CLUSTER_NODES=3
+readonly E2E_CLUSTER_MACHINE=n1-standard-4
+readonly TEST_RESULT_FILE=/tmp/${E2E_BASE_NAME}-e2e-result
+
+# Tear down the test resources.
+function teardown_test_resources() {
+  header "Tearing down test environment"
+  # Free resources in GCP project.
+  if (( ! USING_EXISTING_CLUSTER )); then
+    teardown
+  fi
+
+  # Delete Knative Serving images when using prow.
+  if (( IS_PROW )); then
+    echo "Images in ${DOCKER_REPO_OVERRIDE}:"
+    gcloud container images list --repository=${DOCKER_REPO_OVERRIDE}
+    delete_gcr_images ${DOCKER_REPO_OVERRIDE}
+  else
+    # Delete the kubernetes source downloaded by kubetest
+    rm -fr kubernetes kubernetes.tar.gz
+  fi
+}
+
+# Exit test if the previous command failed, dumping current state info.
+# Parameters: $1 - error message (optional).
+function fail_test() {
+  [[ $? -eq 0 ]] && return 0
+  [[ -n $1 ]] && echo "ERROR: $1"
+  dump_k8s_info
+  exit 1
+}
+
+# Dump info about the test cluster. If dump_extra_cluster_info() is defined, calls it too.
+# This is intended to be called when a test fails to provide debugging information.
+function dump_cluster_state() {
+  echo "***************************************"
+  echo "***           TEST FAILED           ***"
+  echo "***    Start of information dump    ***"
+  echo "***************************************"
+  echo ">>> All resources:"
+  kubectl get all --all-namespaces
+  echo ">>> Services:"
+  kubectl get services --all-namespaces
+  echo ">>> Events:"
+  kubectl get events --all-namespaces
+  [[ "$(type -t dump_extra_cluster_state)" == "function" ]] && dump_extra_cluster_state
+  echo "***************************************"
+  echo "***           TEST FAILED           ***"
+  echo "***     End of information dump     ***"
+  echo "***************************************"
+}
+
+# Create a test cluster with kubetest and call the current script again.
+function create_test_cluster() {
+  header "Creating test cluster"
+  # Smallest cluster required to run the end-to-end-tests
+  local CLUSTER_CREATION_ARGS=(
+    --gke-create-args="--enable-autoscaling --min-nodes=1 --max-nodes=${E2E_CLUSTER_NODES} --scopes=cloud-platform"
+    --gke-shape={\"default\":{\"Nodes\":${E2E_CLUSTER_NODES}\,\"MachineType\":\"${E2E_CLUSTER_MACHINE}\"}}
+    --provider=gke
+    --deployment=gke
+    --cluster="${E2E_CLUSTER_NAME}"
+    --gcp-zone="${E2E_CLUSTER_ZONE}"
+    --gcp-network="${E2E_NETWORK_NAME}"
+    --gke-environment=prod
+  )
+  if (( ! IS_PROW )); then
+    CLUSTER_CREATION_ARGS+=(--gcp-project=${PROJECT_ID:?"PROJECT_ID must be set to the GCP project where the tests are run."})
+  fi
+  # SSH keys are not used, but kubetest checks for their existence.
+  # Touch them so if they don't exist, empty files are create to satisfy the check.
+  touch $HOME/.ssh/google_compute_engine.pub
+  touch $HOME/.ssh/google_compute_engine
+  # Clear user and cluster variables, so they'll be set to the test cluster.
+  # DOCKER_REPO_OVERRIDE is not touched because when running locally it must
+  # be a writeable docker repo.
+  export K8S_USER_OVERRIDE=
+  export K8S_CLUSTER_OVERRIDE=
+  # Assume test failed (see more details at the end of this script).
+  echo -n "1"> ${TEST_RESULT_FILE}
+  local test_cmd_args="--run-tests"
+  (( EMIT_METRICS )) && test_cmd_args+=" --emit-metrics"
+  # Normalize script path; we can't use readlink because it's not available everywhere
+  local script=$0
+  [[ ${script} =~ ^[\./].* ]] || script="./$0"
+  script="$(cd ${script%/*} && echo $PWD/${script##*/})"
+  kubetest "${CLUSTER_CREATION_ARGS[@]}" \
+    --up \
+    --down \
+    --extract "gke-${SERVING_GKE_VERSION}" \
+    --gcp-node-image ${SERVING_GKE_IMAGE} \
+    --test-cmd "${script}" \
+    --test-cmd-args "${test_cmd_args}"
+  # Delete target pools and health checks that might have leaked.
+  # See https://github.com/knative/serving/issues/959 for details.
+  # TODO(adrcunha): Remove once the leak issue is resolved.
+  local gcp_project=${PROJECT_ID}
+  [[ -z ${gcp_project} ]] && gcp_project=$(gcloud config get-value project)
+  local http_health_checks="$(gcloud compute target-pools list \
+    --project=${gcp_project} --format='value(healthChecks)' --filter="instances~-${E2E_CLUSTER_NAME}-" | \
+    grep httpHealthChecks | tr '\n' ' ')"
+  local target_pools="$(gcloud compute target-pools list \
+    --project=${gcp_project} --format='value(name)' --filter="instances~-${E2E_CLUSTER_NAME}-" | \
+    tr '\n' ' ')"
+  local region="$(gcloud compute zones list --filter=name=${E2E_CLUSTER_ZONE} --format='value(region)')"
+  if [[ -n "${target_pools}" ]]; then
+    echo "Found leaked target pools, deleting"
+    gcloud compute forwarding-rules delete -q --project=${gcp_project} --region=${region} ${target_pools}
+    gcloud compute target-pools delete -q --project=${gcp_project} --region=${region} ${target_pools}
+  fi
+  if [[ -n "${http_health_checks}" ]]; then
+    echo "Found leaked health checks, deleting"
+    gcloud compute http-health-checks delete -q --project=${gcp_project} ${http_health_checks}
+  fi
+  local result="$(cat ${TEST_RESULT_FILE})"
+  echo "Test result code is $result"
+  exit ${result}
+}
+
+# Setup the test cluster for running the tests.
+function setup_test_cluster() {
+  # Fail fast during setup.
+  set -o errexit
+  set -o pipefail
+
+  # Set the required variables if necessary.
+  if [[ -z ${K8S_USER_OVERRIDE} ]]; then
+    export K8S_USER_OVERRIDE=$(gcloud config get-value core/account)
+  fi
+
+  if [[ -z ${K8S_CLUSTER_OVERRIDE} ]]; then
+    USING_EXISTING_CLUSTER=0
+    export K8S_CLUSTER_OVERRIDE=$(kubectl config current-context)
+    acquire_cluster_admin_role ${K8S_USER_OVERRIDE} ${E2E_CLUSTER_NAME} ${E2E_CLUSTER_ZONE}
+    # Make sure we're in the default namespace. Currently kubetest switches to
+    # test-pods namespace when creating the cluster.
+    kubectl config set-context $K8S_CLUSTER_OVERRIDE --namespace=default
+  fi
+  readonly USING_EXISTING_CLUSTER
+
+  if [[ -z ${DOCKER_REPO_OVERRIDE} ]]; then
+    export DOCKER_REPO_OVERRIDE=gcr.io/$(gcloud config get-value project)/${E2E_BASE_NAME}e2e-img
+  fi
+
+  echo "- Cluster is ${K8S_CLUSTER_OVERRIDE}"
+  echo "- User is ${K8S_USER_OVERRIDE}"
+  echo "- Docker is ${DOCKER_REPO_OVERRIDE}"
+
+  trap teardown_test_resources EXIT
+
+  if (( USING_EXISTING_CLUSTER )); then
+    echo "Deleting any previous Knative Serving instance"
+    teardown
+  fi
+
+  readonly K8S_CLUSTER_OVERRIDE
+  readonly K8S_USER_OVERRIDE
+  readonly DOCKER_REPO_OVERRIDE
+
+  # Handle failures ourselves, so we can dump useful info.
+  set +o errexit
+  set +o pipefail
+}
+
+function success() {
+  # kubetest teardown might fail and thus incorrectly report failure of the
+  # script, even if the tests pass.
+  # We store the real test result to return it later, ignoring any teardown
+  # failure in kubetest.
+  # TODO(adrcunha): Get rid of this workaround.
+  echo -n "0"> ${TEST_RESULT_FILE}
+  echo "**************************************"
+  echo "***        ALL TESTS PASSED        ***"
+  echo "**************************************"
+  exit 0
+}
+
+RUN_TESTS=0
+EMIT_METRICS=0
+USING_EXISTING_CLUSTER=1
+
+# Parse flags and initialize the test cluster.
+function initialize() {
+  cd ${REPO_ROOT_DIR}
+  for parameter in $@; do
+    case $parameter in
+      --run-tests) RUN_TESTS=1 ;;
+      --emit-metrics) EMIT_METRICS=1 ;;
+      *)
+        echo "error: unknown option ${parameter}"
+        echo "usage: $0 [--run-tests][--emit-metrics]"
+        exit 1
+        ;;
+    esac
+    shift
+  done
+  readonly RUN_TESTS
+  readonly EMIT_METRICS
+
+  if (( ! RUN_TESTS )); then
+    create_test_cluster
+  else
+    setup_test_cluster
+  fi
+}