feat: declare ResponseBodyMode on api-translation and nemo-response-guard

- api-translation: configurable via "responseBodyMode" parameter
  (none/chunked/full, defaults to full)
- nemo-response-guard: always BodyFull (needs complete body for
  content inspection)

Depends on opendatahub-io#331 (framework migration)

Author: noyitz

.gitignore

@@ -3,3 +3,4 @@ bin/*
 
 # Downloaded CRDs for envtest (regenerated by hack/download-test-crds.sh)
 test/testdata/crds/
+.claude/

go.mod

@@ -13,13 +13,13 @@ require (
 	k8s.io/kubectl v0.35.5
 	sigs.k8s.io/controller-runtime v0.23.3
 	sigs.k8s.io/gateway-api v1.5.1
+	sigs.k8s.io/gateway-api-inference-extension v1.5.0
 )
 
 require (
 	github.com/Masterminds/semver/v3 v3.4.0 // indirect
 	github.com/aws/smithy-go v1.25.1 // indirect
 	github.com/go-task/slim-sprig/v3 v3.0.0 // indirect
-	github.com/goccy/go-yaml v1.19.2 // indirect
 	github.com/google/pprof v0.0.0-20260402051712-545e8a4df936 // indirect
 	github.com/gorilla/websocket v1.5.4-0.20250319132907-e064f32e3674 // indirect
 	github.com/moby/spdystream v0.5.1 // indirect
@@ -125,4 +125,4 @@ require (
 	sigs.k8s.io/yaml v1.6.0 // indirect
 )
 
-replace github.com/llm-d/llm-d-inference-payload-processor => ../llm-d-inference-payload-processor
+replace github.com/llm-d/llm-d-inference-payload-processor => github.com/noyitz/llm-d-inference-payload-processor v0.0.0-20260616213505-bf93834c3cf2

go.sum

@@ -143,6 +143,8 @@ github.com/munnerz/goautoneg v0.0.0-20191010083416-a7dc8b61c822 h1:C3w9PqII01/Oq
 github.com/munnerz/goautoneg v0.0.0-20191010083416-a7dc8b61c822/go.mod h1:+n7T8mK8HuQTcFwEeznm/DIxMOiR9yIdICNftLE1DvQ=
 github.com/mxk/go-flowrate v0.0.0-20140419014527-cca7078d478f h1:y5//uYreIhSUg3J1GEMiLbxo1LJaP8RfCpH6pymGZus=
 github.com/mxk/go-flowrate v0.0.0-20140419014527-cca7078d478f/go.mod h1:ZdcZmHo+o7JKHSa8/e818NopupXU1YMK5fe1lsApnBw=
+github.com/noyitz/llm-d-inference-payload-processor v0.0.0-20260616213505-bf93834c3cf2 h1:4ih3EdzAhpcA70oELfGv0qx5EMYR8RHPgF2Bgl4j1jA=
+github.com/noyitz/llm-d-inference-payload-processor v0.0.0-20260616213505-bf93834c3cf2/go.mod h1:UZldktV/E/LtaFHrtZdPOhy1d7WdazMRe3Q9OW7f9jI=
 github.com/onsi/ginkgo/v2 v2.30.0 h1:zxM/9XneXFIy64j6/wAmBIX4zRC7Hu6U8XFNZvDnCQc=
 github.com/onsi/ginkgo/v2 v2.30.0/go.mod h1:+aXOY+vzZ5mu2iI2HpTZUPmM//oQfsNFX6gU9kNcA44=
 github.com/onsi/gomega v1.41.0 h1:OwKp4pXNgVxf6sCplzYo794OFNuoL2q2SBMU5NSWOjA=

pkg/plugins/api-translation/plugin.go

@@ -47,10 +47,12 @@ const (
 // compile-time type validation
 var _ requesthandling.RequestProcessor = &APITranslationPlugin{}
 var _ requesthandling.ResponseProcessor = &APITranslationPlugin{}
+var _ requesthandling.ResponseBodyRequirement = &APITranslationPlugin{}
 
 // apiTranslationConfig holds configuration for provider-specific translators.
 type apiTranslationConfig struct {
-	VertexOpenAI *vertexOpenAIConfig `json:"vertexOpenAI,omitempty"`
+	VertexOpenAI     *vertexOpenAIConfig `json:"vertexOpenAI,omitempty"`
+	BodyProcessingMode string              `json:"responseBodyMode,omitempty"`
 }
 
 type vertexOpenAIConfig struct {
@@ -100,27 +102,34 @@ func NewAPITranslationPlugin(ctx context.Context, config apiTranslationConfig) (
 		)
 	}
 
+	bodyMode, err := parseBodyProcessingMode(config.BodyProcessingMode)
+	if err != nil {
+		return nil, err
+	}
+
 	keys := make([]string, 0, len(providers))
 	for key := range providers {
 		keys = append(keys, key)
 	}
 
-	log.FromContext(ctx).V(logutil.VERBOSE).Info("plugin initialized", "providers", strings.Join(keys, ","))
+	log.FromContext(ctx).V(logutil.VERBOSE).Info("plugin initialized", "providers", strings.Join(keys, ","), "responseBodyMode", bodyMode)
 
 	return &APITranslationPlugin{
 		typedName: plugin.TypedName{
 			Type: APITranslationPluginType,
 			Name: APITranslationPluginType,
 		},
-		providers: providers,
+		providers:        providers,
+		responseBodyMode: bodyMode,
 	}, nil
 }
 
 // APITranslationPlugin translates inference API requests and responses between
 // OpenAI Chat Completions format and provider-native formats (e.g., Anthropic Messages API).
 type APITranslationPlugin struct {
-	typedName plugin.TypedName
-	providers map[string]translator.Translator // map from provider name to translator interface
+	typedName        plugin.TypedName
+	providers        map[string]translator.Translator // map from provider name to translator interface
+	responseBodyMode requesthandling.BodyProcessingMode
 }
 
 // TypedName returns the type and name tuple of this plugin instance.
@@ -134,6 +143,25 @@ func (p *APITranslationPlugin) WithName(name string) *APITranslationPlugin {
 	return p
 }
 
+// BodyProcessingMode returns the configured response body mode.
+// Defaults to Full. Configurable via the "responseBodyMode" parameter.
+func (p *APITranslationPlugin) BodyProcessingMode() requesthandling.BodyProcessingMode {
+	return p.responseBodyMode
+}
+
+func parseBodyProcessingMode(s string) (requesthandling.BodyProcessingMode, error) {
+	switch strings.ToLower(s) {
+	case "", "full":
+		return requesthandling.Full, nil
+	case "chunked":
+		return requesthandling.Chunks, nil
+	case "none":
+		return requesthandling.Skip, nil
+	default:
+		return 0, fmt.Errorf("invalid responseBodyMode %q (valid: none, chunked, full)", s)
+	}
+}
+
 // ProcessRequest reads the provider from CycleState (set by an upstream plugin) and translates
 // the request body from OpenAI format to the provider's native format if needed.
 // When the incoming client format matches the upstream API format (passthrough mode),

pkg/plugins/apikey-injection/auth-generator/simple_auth_generator_test.go

@@ -19,10 +19,12 @@ package authgenerator
 import (
 	"testing"
 
+	"github.com/llm-d/llm-d-inference-payload-processor/pkg/framework/interface/plugin"
+	"github.com/llm-d/llm-d-inference-payload-processor/pkg/framework/interface/requesthandling"
+
 	"github.com/google/go-cmp/cmp"
 	"github.com/google/go-cmp/cmp/cmpopts"
 	"github.com/stretchr/testify/require"
-	"sigs.k8s.io/gateway-api-inference-extension/pkg/bbr/framework"
 
 	"github.com/opendatahub-io/ai-gateway-payload-processing/pkg/plugins/common/auth"
 	"github.com/opendatahub-io/ai-gateway-payload-processing/pkg/plugins/common/state"
@@ -154,11 +156,11 @@ func TestSimpleExtractRequestData(t *testing.T) {
 
 	for _, test := range tests {
 		t.Run(test.name, func(t *testing.T) {
-			cs := framework.NewCycleState()
+			cs := plugin.NewCycleState()
 			cs.Write(state.ModelConfigKey, test.config)
 
 			generator := NewSimpleAuthGenerator()
-			result, err := generator.ExtractRequestData(cs, framework.NewInferenceRequest())
+			result, err := generator.ExtractRequestData(cs, requesthandling.NewInferenceRequest())
 
 			require.NoError(t, err)
 			require.Equal(t, test.wantHeaderName, result[auth.SimpleAuthHeaderName])
@@ -169,7 +171,7 @@ func TestSimpleExtractRequestData(t *testing.T) {
 
 func TestSimpleExtractRequestData_MissingModelConfigKey(t *testing.T) {
 	generator := NewSimpleAuthGenerator()
-	_, err := generator.ExtractRequestData(framework.NewCycleState(), framework.NewInferenceRequest())
+	_, err := generator.ExtractRequestData(plugin.NewCycleState(), requesthandling.NewInferenceRequest())
 	require.Error(t, err)
 	require.Contains(t, err.Error(), "failed to extract config from cycle state")
 }

pkg/plugins/nemo/response_guard.go

@@ -33,6 +33,7 @@ const (
 
 // compile-time type validation
 var _ requesthandling.ResponseProcessor = &NemoResponseGuardPlugin{}
+var _ requesthandling.ResponseBodyRequirement = &NemoResponseGuardPlugin{}
 
 // NemoResponseGuardPlugin calls a NeMo Guardrails service over HTTP to check model output
 // using output rails. It implements ResponseProcessor to inspect responses before returning
@@ -85,6 +86,12 @@ func (p *NemoResponseGuardPlugin) WithName(name string) *NemoResponseGuardPlugin
 	return p
 }
 
+// BodyProcessingMode returns Full because NeMo guardrails need the complete response
+// body to inspect the full model output.
+func (p *NemoResponseGuardPlugin) BodyProcessingMode() requesthandling.BodyProcessingMode {
+	return requesthandling.Full
+}
+
 // ProcessResponse calls NeMo Guardrails to evaluate output rails on the model response.
 // It extracts assistant messages from the OpenAI-style response body, POSTs them to the
 // configured NeMo endpoint, and returns an errcommon.Error with Forbidden (403) if NeMo