aepfli · Oct 21, 2026 · Oct 21, 2026 · Oct 21, 2026 · Oct 21, 2026 · Oct 21, 2026
@@ -20,20 +20,14 @@ In vscode for instance, the following settings are recommended:
 
 ## End-to-End Tests
 
-The continuous integration runs a set of [gherkin e2e tests](https://github.com/open-feature/test-harness/blob/main/features/evaluation.feature) using [`flagd`](https://github.com/open-feature/flagd). These tests do not run with the default maven profile. If you'd like to run them locally, you can start the flagd testbed with
+The [flagd-testbed](https://github.com/open-feature/flagd-testbed) scenarios run through the
+Provider TCK, which starts the testbed itself (Docker required):
 
 ```
-docker compose up
-```
-and then run 
-```
-mvn test -P e2e
+mvn -Ptck -pl providers/flagd -am test
 ```
 
-Note if your docker compose networking does not resolve to localhost, you can set a custom host
-```
-FLAGD_HOST=192.168.100.1 mvn test -P e2e
-```
+See [Provider conformance (TCK)](README.md#provider-conformance-tck) for what runs and how.
 
 If test-harness is being updated, commit the changes before running `mvn`.
 Otherwise, uncommitted changes will be overridden by maven test plugins.
@@ -375,9 +375,11 @@ mvn -Ptck -pl providers/flagd -am -Dtest=InProcessTest     -Dsurefire.failIfNoSp
 
 The root pom's `tck` profile runs only the `tck` package, so `-am` adds no other module's tests.
 
-**`-Pe2e` does not run these suites.** The root pom excludes every provider's `tck` package from
-the default build and from `-Pe2e`, and its `tck` profile runs only that package. This module
-excludes just `**/e2e/*.java`, which the `e2e` profile clears.
+**These suites also carry the rest of flagd-testbed's scenarios**: its `tck-extensions/`, read
+from the `test-harness` submodule and run through the TCK's extension hook, so they replace the
+former `e2e` suites. `ConfigCucumberTest` still reads `config.feature` from the `test-harness`
+submodule; it needs no backend and runs in the default build. The root pom excludes every
+provider's `tck` package from the default build and from `-Pe2e`.
 
 **CI runs them in a job of their own**, `Provider TCK` in `ci.yml`, on every pull request. It sits
 apart from the `main` matrix so a conformance failure cannot be read as a regression there. Set

@@ -14,12 +14,6 @@
     <version>0.15.0</version> <!--x-release-please-version -->
 
     <properties>
-        <!--
-            Exclusion for the legacy e2e suites over the test-harness submodule; the `e2e` profile
-            below clears it. The Provider TCK adoption in tck needs nothing here: the root pom
-            excludes every provider's tck package and its `tck` profile selects them.
-        -->
-        <testExclusions>**/e2e/*.java</testExclusions>
         <io.grpc.version>1.85.0</io.grpc.version>
         <!-- caution - updating this will break compatibility with older protobuf-java versions -->
         <protobuf-java.min.version>3.25.6</protobuf-java.min.version>
@@ -182,6 +176,20 @@
             </extension>
         </extensions>
 
+        <testResources>
+            <testResource>
+                <directory>src/test/resources</directory>
+            </testResource>
+            <!-- flagd-testbed's TCK extensions, on the classpath where the TCK looks for them. -->
+            <testResource>
+                <directory>test-harness/tck-extensions</directory>
+                <targetPath>extensions</targetPath>
+                <includes>
+                    <include>*.feature</include>
+                </includes>
+            </testResource>
+        </testResources>
+
         <plugins>
             <plugin>
                 <groupId>org.codehaus.mojo</groupId>
@@ -269,16 +277,12 @@
             </build>
         </profile>
         <profile>
-            <!-- this profile handles running the flagd e2e tests -->
+            <!--
+                Initialises the test-harness submodule: ConfigCucumberTest reads its config.feature
+                and the Provider TCK suites under tck/ its compose file. The flagd-testbed
+                scenarios themselves run as TCK extensions, through the root pom's `tck` profile.
+            -->
             <id>e2e</id>
-            <properties>
-                <!--
-                    Run the legacy flagd e2e suites (Run*Test, over the test-harness submodule).
-                    The Provider TCK suites stay out: the root pom excludes them outside its `tck`
-                    profile, which ci.yml's `tck` job runs on its own.
-                -->
-                <testExclusions/>
-            </properties>
             <build>
                 <plugins>
                     <!-- pull the gherkin tests as a git submodule  -->

@@ -12,17 +12,17 @@
 import org.junit.platform.suite.api.Suite;
 
 /**
- * Class for running the tests associated with "stable" e2e tests (no fake disconnection) for the
- * in-process provider
+ * Runs the flagd-testbed's config.feature, which checks option defaults, environment variables
+ * and builder validation. No backend is involved.
  */
 @Order(value = Integer.MAX_VALUE)
 @Suite
 @IncludeEngines("cucumber")
 @SelectFile("test-harness/gherkin/config.feature")
 @ConfigurationParameter(key = PLUGIN_PROPERTY_NAME, value = "summary")
-@ConfigurationParameter(key = GLUE_PROPERTY_NAME, value = "dev.openfeature.contrib.providers.flagd.e2e.steps.config")
+@ConfigurationParameter(key = GLUE_PROPERTY_NAME, value = "dev.openfeature.contrib.providers.flagd.config")
 // The TCK ships its own Cucumber ObjectFactory, so two are on the test classpath and Cucumber
-// refuses to guess. The e2e suites name the same one; this suite needs it for the same reason.
+// refuses to guess.
 @ConfigurationParameter(key = OBJECT_FACTORY_PROPERTY_NAME, value = "io.cucumber.picocontainer.PicoFactory")
 // Config scenarios read System env vars in FlagdOptions.build() and some scenarios also
 // mutate them. Parallel execution causes env-var races (e.g. FLAGD_PORT=3456 leaking into

@@ -0,0 +1,10 @@
+package dev.openfeature.contrib.providers.flagd.config;
+
+import dev.openfeature.contrib.providers.flagd.FlagdOptions;
+
+/** Scenario state shared by the config steps, one instance per scenario. */
+public class ConfigState {
+    public FlagdOptions options;
+    public FlagdOptions.FlagdOptionsBuilder builder = FlagdOptions.builder();
+    public boolean hasError;
+}
@@ -1,12 +1,8 @@
-package dev.openfeature.contrib.providers.flagd.e2e.steps.config;
+package dev.openfeature.contrib.providers.flagd.config;
 
 import static org.assertj.core.api.Assertions.assertThat;
 
 import dev.openfeature.contrib.providers.flagd.Config;
-import dev.openfeature.contrib.providers.flagd.e2e.State;
-import dev.openfeature.contrib.providers.flagd.e2e.steps.AbstractSteps;
-import dev.openfeature.contrib.providers.flagd.e2e.steps.EnvironmentVariableUtils;
-import dev.openfeature.contrib.providers.flagd.e2e.steps.Utils;
 import io.cucumber.java.en.Given;
 import io.cucumber.java.en.Then;
 import io.cucumber.java.en.When;
@@ -19,7 +15,9 @@
 import lombok.extern.slf4j.Slf4j;
 
 @Slf4j
-public class ConfigSteps extends AbstractSteps {
+public class ConfigSteps {
+    private final ConfigState state;
+
     /**
      * Not all properties are correctly implemented, hence that we need to ignore them till this is
      * fixed
@@ -30,8 +28,8 @@ public class ConfigSteps extends AbstractSteps {
         }
     };
 
-    public ConfigSteps(State state) {
-        super(state);
+    public ConfigSteps(ConfigState state) {
+        this.state = state;
     }
 
     @When("a config was initialized")

@@ -1,4 +1,4 @@
-package dev.openfeature.contrib.providers.flagd.e2e.steps;
+package dev.openfeature.contrib.providers.flagd.config;
 
 /*
  * Copy of JUnit Pioneer's EnvironmentVariable Utils

@@ -1,4 +1,4 @@
-package dev.openfeature.contrib.providers.flagd.e2e.steps;
+package dev.openfeature.contrib.providers.flagd.config;
 
 import dev.openfeature.contrib.providers.flagd.Config;
 import dev.openfeature.contrib.providers.flagd.resolver.rpc.cache.CacheType;