refactor(executor): optimize translation logic and improve plugin invocation handling

internal/runtime/executor/gemini_executor.go

@@ -885,14 +885,14 @@ func translateGeminiInteractionsRequestBody(ctx context.Context, cfg *config.Con
 }
 
 // translateGeminiInteractionsRequestPair translates the working payload and the
-// payload-config baseline. Identical inputs are translated once when no plugin
-// hooks are installed. The baseline is captured before model and thinking
-// mutations, and the working buffer is a separate copy so those mutations cannot
-// change it. Distinct inputs and plugin hooks keep the existing order: working
-// payload first, then the payload-config source.
+// payload-config baseline. Identical inputs are translated once, including plugin
+// hooks. The baseline is captured before model and thinking mutations, and the
+// working buffer is a separate copy so those mutations cannot change it. Distinct
+// inputs keep the existing order: working payload first, then the payload-config
+// source.
 func translateGeminiInteractionsRequestPair(ctx context.Context, cfg *config.Config, model string, payload []byte, opts cliproxyexecutor.Options, stream, isCompat bool) (original, working []byte) {
 	source := geminiInteractionsPayloadConfigInput(opts, payload)
-	if geminiInteractionsSameByteSlice(payload, source) && !sdktranslator.HasPluginHooks() {
+	if geminiInteractionsSameByteSlice(payload, source) {
 		original = translateGeminiInteractionsRequestBody(ctx, cfg, model, payload, opts, stream, isCompat)
 		return original, bytes.Clone(original)
 	}

internal/runtime/executor/gemini_interactions_translate_test.go

@@ -76,9 +76,6 @@ func TestTranslateGeminiInteractionsRequestPairTranslatesSameSliceOnce(t *testin
 	if sdktranslator.HasRequestTransformer(from, to) {
 		t.Fatalf("request transformer %s -> %s is already registered", from, to)
 	}
-	if sdktranslator.HasPluginHooks() {
-		t.Fatal("plugin hooks are installed and disable translation reuse")
-	}
 
 	const model = "gemini-interactions-count"
 	var calls int
@@ -156,7 +153,7 @@ func TestTranslateGeminiInteractionsRequestPairTranslatesDistinctInputs(t *testi
 	}
 }
 
-func TestTranslateGeminiInteractionsRequestPairPreservesHookOrder(t *testing.T) {
+func TestTranslateGeminiInteractionsRequestPairInvokesPluginOncePerInput(t *testing.T) {
 	hooks := &interactionsTranslateHooks{}
 	sdktranslator.SetPluginHooks(hooks)
 	t.Cleanup(func() { sdktranslator.SetPluginHooks(nil) })
@@ -166,22 +163,27 @@ func TestTranslateGeminiInteractionsRequestPairPreservesHookOrder(t *testing.T)
 	const model = "gemini-3.1-flash-lite"
 	payload := []byte(`{"model":"gemini-3.1-flash-lite","messages":[{"role":"user","content":"same"}]}`)
 	opts := cliproxyexecutor.Options{SourceFormat: sdktranslator.FormatOpenAI, OriginalRequest: payload}
-	base, work := translateGeminiInteractionsRequestPair(ctx, cfg, model, payload, opts, false, true)
-	if hooks.calls != 2 {
-		t.Fatalf("plugin hook calls = %d, want 2", hooks.calls)
+	for _, stream := range []bool{false, true} {
+		for _, compat := range []bool{false, true} {
+			hooks.calls = 0
+			base, work := translateGeminiInteractionsRequestPair(ctx, cfg, model, payload, opts, stream, compat)
+			if hooks.calls != 1 {
+				t.Fatalf("stream=%v compat=%v: plugin calls = %d, want 1", stream, compat, hooks.calls)
+			}
+			if got := gjson.GetBytes(work, "plugin_call").Int(); got != 1 {
+				t.Fatalf("working plugin_call = %d, want 1", got)
+			}
+			if !bytes.Equal(base, work) {
+				t.Fatal("same input produced different plugin results")
+			}
+			assertIndependentGeminiInteractionsBuffers(t, payload, base, work)
+		}
 	}
-	if got := gjson.GetBytes(work, "plugin_call").Int(); got != 1 {
-		t.Fatalf("working plugin_call = %d, want 1", got)
-	}
-	if got := gjson.GetBytes(base, "plugin_call").Int(); got != 2 {
-		t.Fatalf("baseline plugin_call = %d, want 2", got)
-	}
-	assertIndependentGeminiInteractionsBuffers(t, payload, base, work)
 
 	detached := bytes.Clone(payload)
 	opts.OriginalRequest = detached
 	before := hooks.calls
-	base, work = translateGeminiInteractionsRequestPair(ctx, cfg, model, payload, opts, true, false)
+	base, work := translateGeminiInteractionsRequestPair(ctx, cfg, model, payload, opts, true, false)
 	if hooks.calls != before+2 {
 		t.Fatalf("distinct plugin hook calls = %d, want %d", hooks.calls-before, 2)
 	}

internal/runtime/executor/helps/codex_multi_agent_v2.go

@@ -59,12 +59,10 @@ func TranslateRequestEnvelopeWithCodexMultiAgentV2(ctx context.Context, headers
 
 // TranslateRequestPairWithCodexMultiAgentV2 translates the untouched baseline
 // payload and the working payload that later stages mutate in place. Executors
-// normally assign the original payload to the request before translating, so both
-// translations would rescan the same bytes and produce the same result. Built-in
-// request translation is deterministic, so that case is translated once and
-// duplicated when no plugin hooks are installed. Hooks retain two invocations
-// because they may have request-scoped output or side effects. This removes a
-// full extra pass over payloads that can reach tens of megabytes.
+// normally assign the original payload to the request before translating, so the
+// same input is translated once, including plugin hooks, then copied into a
+// separate working buffer. This avoids duplicate plugin side effects and a full
+// extra pass over payloads that can reach tens of megabytes.
 func TranslateRequestPairWithCodexMultiAgentV2(ctx context.Context, headers http.Header, cfg *config.Config, from, to sdktranslator.Format, model string, originalPayload, requestPayload []byte, stream bool) (original, working []byte) {
 	req := sdktranslator.RequestEnvelope{Format: from, Model: model, Stream: stream}
 	return TranslateRequestEnvelopePairWithCodexMultiAgentV2(ctx, headers, cfg, from, to, req, originalPayload, requestPayload)
@@ -76,7 +74,7 @@ func TranslateRequestEnvelopePairWithCodexMultiAgentV2(ctx context.Context, head
 	originalReq := req
 	originalReq.Body = originalPayload
 	original = TranslateRequestEnvelopeWithCodexMultiAgentV2(ctx, headers, cfg, from, to, originalReq).Body
-	if sameByteSlice(originalPayload, requestPayload) && !sdktranslator.HasPluginHooks() {
+	if sameByteSlice(originalPayload, requestPayload) {
 		// The caller mutates the working copy, so it must not share the baseline array.
 		return original, append([]byte(nil), original...)
 	}
@@ -99,7 +97,7 @@ func sameByteSlice(a, b []byte) bool {
 }
 
 // TranslateRequestPairWithAPIKeyModelCompatibility avoids translating identical
-// inputs twice while retaining separate buffers and stateful plugin invocations.
+// inputs twice while retaining separate buffers and invoking plugins once per input.
 func TranslateRequestPairWithAPIKeyModelCompatibility(ctx context.Context, headers http.Header, cfg *config.Config, from, to sdktranslator.Format, model string, originalPayload, requestPayload []byte, stream, isCompat bool) (original, working []byte) {
 	original, working, _ = TranslateRequestPairWithAPIKeyModelCompatibilityAndUpdateIntent(ctx, headers, cfg, from, to, model, originalPayload, requestPayload, stream, isCompat)
 	return original, working
@@ -108,9 +106,9 @@ func TranslateRequestPairWithAPIKeyModelCompatibility(ctx context.Context, heade
 // TranslateRequestPairWithAPIKeyModelCompatibilityAndUpdateIntent returns the
 // normalizer decision for the working payload, not for the baseline payload.
 func TranslateRequestPairWithAPIKeyModelCompatibilityAndUpdateIntent(ctx context.Context, headers http.Header, cfg *config.Config, from, to sdktranslator.Format, model string, originalPayload, requestPayload []byte, stream, isCompat bool) (original, working []byte, updatesChanged bool) {
-	original, _ = TranslateRequestWithAPIKeyModelCompatibilityAndUpdateIntent(ctx, headers, cfg, from, to, model, originalPayload, stream, isCompat)
-	if sameByteSlice(originalPayload, requestPayload) && !sdktranslator.HasPluginHooks() {
-		return original, append([]byte(nil), original...), false
+	original, updatesChanged = TranslateRequestWithAPIKeyModelCompatibilityAndUpdateIntent(ctx, headers, cfg, from, to, model, originalPayload, stream, isCompat)
+	if sameByteSlice(originalPayload, requestPayload) {
+		return original, append([]byte(nil), original...), updatesChanged
 	}
 	working, updatesChanged = TranslateRequestWithAPIKeyModelCompatibilityAndUpdateIntent(ctx, headers, cfg, from, to, model, requestPayload, stream, isCompat)
 	return original, working, updatesChanged

internal/runtime/executor/helps/codex_multi_agent_v2_test.go

@@ -132,32 +132,45 @@ func TestTranslateRequestPairTranslatesDistinctPayloads(t *testing.T) {
 	}
 }
 
-func TestTranslateRequestPairPreservesPluginHookInvocations(t *testing.T) {
+func TestTranslateRequestPairInvokesPluginOncePerInput(t *testing.T) {
 	hooks := &pairRequestPluginHooks{}
 	sdktranslator.SetPluginHooks(hooks)
 	t.Cleanup(func() { sdktranslator.SetPluginHooks(nil) })
 
 	payload := geminiToolHistoryPayload(1)
-	base, work := TranslateRequestPairWithCodexMultiAgentV2(
-		context.Background(),
-		http.Header{},
-		&config.Config{},
-		sdktranslator.FormatGemini,
-		sdktranslator.FromString("antigravity"),
-		"gemini-3.6-flash-high",
-		payload,
-		payload,
-		true,
-	)
-
-	if hooks.calls != 2 {
-		t.Fatalf("plugin hook calls = %d, want 2", hooks.calls)
-	}
-	if got := gjson.GetBytes(base, "plugin_call").Int(); got != 1 {
-		t.Fatalf("baseline plugin_call = %d, want 1", got)
-	}
-	if got := gjson.GetBytes(work, "plugin_call").Int(); got != 2 {
-		t.Fatalf("working plugin_call = %d, want 2", got)
+	for _, stream := range []bool{false, true} {
+		for _, distinct := range []bool{false, true} {
+			hooks.calls = 0
+			request := payload
+			wantCalls := int64(1)
+			if distinct {
+				request = geminiToolHistoryPayload(2)
+				wantCalls = 2
+			}
+			base, work := TranslateRequestPairWithCodexMultiAgentV2(
+				context.Background(), http.Header{}, &config.Config{},
+				sdktranslator.FormatGemini, sdktranslator.FormatAntigravity,
+				"gemini-3.6-flash-high", payload, request, stream,
+			)
+			if hooks.calls != wantCalls {
+				t.Fatalf("stream=%v distinct=%v: plugin calls = %d, want %d", stream, distinct, hooks.calls, wantCalls)
+			}
+			if got := gjson.GetBytes(base, "plugin_call").Int(); got != 1 {
+				t.Fatalf("baseline plugin_call = %d, want 1", got)
+			}
+			if got := gjson.GetBytes(work, "plugin_call").Int(); got != wantCalls {
+				t.Fatalf("working plugin_call = %d, want %d", got, wantCalls)
+			}
+			if !distinct && !bytes.Equal(base, work) {
+				t.Fatal("same input produced different plugin results")
+			}
+			baselineBefore := bytes.Clone(base)
+			inputBefore := bytes.Clone(request)
+			work[0] = 'X'
+			if !bytes.Equal(base, baselineBefore) || !bytes.Equal(request, inputBefore) {
+				t.Fatal("working buffer aliases the baseline or input")
+			}
+		}
 	}
 }
 

internal/runtime/executor/helps/configuration_update_plugin_test.go

@@ -1,9 +1,11 @@
 package helps_test
 
 import (
+	"bytes"
 	"context"
 	"testing"
 
+	"github.com/router-for-me/CLIProxyAPI/v8/internal/config"
 	"github.com/router-for-me/CLIProxyAPI/v8/internal/registry"
 	helps "github.com/router-for-me/CLIProxyAPI/v8/internal/runtime/executor/helps"
 	_ "github.com/router-for-me/CLIProxyAPI/v8/internal/thinking/provider/claude"
@@ -18,9 +20,11 @@ import (
 type editUpdatePluginHooks struct {
 	summaryRemovingPluginHooks
 	effort string
+	calls  int
 }
 
 func (h *editUpdatePluginHooks) NormalizeRequest(_ context.Context, _, _ sdktranslator.Format, _ string, body []byte, _ bool) []byte {
+	h.calls++
 	if h.effort == "" {
 		out, _ := sjson.DeleteBytes(body, "input.0")
 		return out
@@ -57,6 +61,41 @@ func TestPluginConfigurationUpdateNormalizationIsAuthoritative(t *testing.T) {
 	}
 }
 
+func TestRequestPairPreservesPluginConfigurationUpdateIntent(t *testing.T) {
+	for _, stream := range []bool{false, true} {
+		for _, compat := range []bool{false, true} {
+			for _, distinct := range []bool{false, true} {
+				hooks := &editUpdatePluginHooks{}
+				sdktranslator.SetPluginHooks(hooks)
+				t.Cleanup(func() { sdktranslator.SetPluginHooks(nil) })
+				original := []byte(`{"reasoning":{"effort":"medium"},"input":[{"type":"configuration_update","reasoning":{"effort":"low"}},{"role":"user","content":"hi"}]}`)
+				request := original
+				wantCalls := 1
+				if distinct {
+					request = []byte(`{"input":[{"role":"user","content":"different"}]}`)
+					wantCalls = 2
+				}
+				inputBefore := bytes.Clone(request)
+				base, work, updatesChanged := helps.TranslateRequestPairWithAPIKeyModelCompatibilityAndUpdateIntent(t.Context(), nil, &config.Config{}, sdktranslator.FormatOpenAIResponse, sdktranslator.FormatCodex, "opaque-route", original, request, stream, compat)
+				if hooks.calls != wantCalls {
+					t.Fatalf("stream=%v compat=%v distinct=%v: plugin calls = %d, want %d", stream, compat, distinct, hooks.calls, wantCalls)
+				}
+				if updatesChanged != !distinct {
+					t.Fatalf("stream=%v compat=%v distinct=%v: working update intent = %v, want %v", stream, compat, distinct, updatesChanged, !distinct)
+				}
+				if !distinct && !bytes.Equal(base, work) {
+					t.Fatal("same input produced different plugin results")
+				}
+				baselineBefore := bytes.Clone(base)
+				work[0] = 'X'
+				if !bytes.Equal(base, baselineBefore) || !bytes.Equal(request, inputBefore) {
+					t.Fatal("working buffer aliases the baseline or input")
+				}
+			}
+		}
+	}
+}
+
 func TestNativeCrossProtocolTranslationWithoutPluginRetainsSourceUpdate(t *testing.T) {
 	const source = `{"model":"private-chat","reasoning":{"effort":"medium"},"input":[{"type":"configuration_update","reasoning":{"effort":"high"}},{"role":"user","content":"hi"}]}`
 	translated := sdktranslator.TranslateRequestEnvelope(t.Context(), sdktranslator.FormatOpenAIResponse, sdktranslator.FormatOpenAI, sdktranslator.RequestEnvelope{Model: "private-chat", Body: []byte(source)})

internal/runtime/executor/helps/request_pair_compat_test.go

@@ -45,9 +45,6 @@ func TestTranslateRequestPairWithAPIKeyModelCompatibilityCountsTranslations(t *t
 	if sdktranslator.HasRequestTransformer(from, to) {
 		t.Fatalf("request transformer %s -> %s is already registered", from, to)
 	}
-	if sdktranslator.HasPluginHooks() {
-		t.Fatal("plugin hooks are installed and disable translation reuse")
-	}
 
 	const model = "compat-count-model"
 	var calls int
@@ -103,13 +100,35 @@ func TestTranslateRequestPairWithAPIKeyModelCompatibilityCountsTranslations(t *t
 	}
 }
 
-func TestCompatibilityRequestPairPreservesHooks(t *testing.T) {
+func TestCompatibilityRequestPairInvokesPluginOncePerInput(t *testing.T) {
 	hooks := &pairRequestPluginHooks{}
 	sdktranslator.SetPluginHooks(hooks)
 	t.Cleanup(func() { sdktranslator.SetPluginHooks(nil) })
-	request := []byte(`{"model":"test","input":"hello"}`)
-	base, work := TranslateRequestPairWithAPIKeyModelCompatibility(context.Background(), nil, &config.Config{}, sdktranslator.FormatOpenAIResponse, sdktranslator.FormatOpenAI, "test", request, request, true, true)
-	if hooks.calls != 2 || bytes.Equal(base, work) {
-		t.Fatalf("stateful plugin calls must remain independent, calls=%d", hooks.calls)
+	payload := []byte(`{"model":"test","input":"hello"}`)
+	for _, stream := range []bool{false, true} {
+		for _, compat := range []bool{false, true} {
+			for _, distinct := range []bool{false, true} {
+				hooks.calls = 0
+				request := payload
+				wantCalls := int64(1)
+				if distinct {
+					request = []byte(`{"model":"test","input":"changed"}`)
+					wantCalls = 2
+				}
+				base, work := TranslateRequestPairWithAPIKeyModelCompatibility(context.Background(), nil, &config.Config{}, sdktranslator.FormatOpenAIResponse, sdktranslator.FormatOpenAI, "test", payload, request, stream, compat)
+				if hooks.calls != wantCalls {
+					t.Fatalf("stream=%v compat=%v distinct=%v: plugin calls = %d, want %d", stream, compat, distinct, hooks.calls, wantCalls)
+				}
+				if bytes.Equal(base, work) == distinct {
+					t.Fatal("plugin results do not match input identity")
+				}
+				baselineBefore := bytes.Clone(base)
+				inputBefore := bytes.Clone(request)
+				work[0] = 'X'
+				if !bytes.Equal(base, baselineBefore) || !bytes.Equal(request, inputBefore) {
+					t.Fatal("working buffer aliases the baseline or input")
+				}
+			}
+		}
 	}
 }

internal/runtime/executor/helps/request_pair_plugin_host_test.go

@@ -0,0 +1,76 @@
+package helps_test
+
+import (
+	"bytes"
+	"fmt"
+	"strings"
+	"testing"
+
+	"github.com/router-for-me/CLIProxyAPI/v8/internal/config"
+	"github.com/router-for-me/CLIProxyAPI/v8/internal/pluginhost"
+	"github.com/router-for-me/CLIProxyAPI/v8/internal/runtime/executor/helps"
+	sdktranslator "github.com/router-for-me/CLIProxyAPI/v8/sdk/translator"
+	log "github.com/sirupsen/logrus"
+	"github.com/tidwall/gjson"
+)
+
+func TestRequestPairWithEmptyPluginHostSanitizesSignaturesOnce(t *testing.T) {
+	logger := log.StandardLogger()
+	oldOutput, oldLevel := logger.Out, logger.GetLevel()
+	var logs bytes.Buffer
+	logger.SetOutput(&logs)
+	logger.SetLevel(log.DebugLevel)
+	t.Cleanup(func() {
+		logger.SetOutput(oldOutput)
+		logger.SetLevel(oldLevel)
+		sdktranslator.SetPluginHooks(nil)
+	})
+
+	var items []string
+	for i := range 10 {
+		items = append(items, fmt.Sprintf(`{"type":"function_call","call_id":"call_%d","name":"read_file","arguments":"{}"}`, i))
+	}
+	for i := range 10 {
+		items = append(items, fmt.Sprintf(`{"type":"function_call_output","call_id":"call_%d","output":"ok"}`, i))
+	}
+	items = append(items, `{"role":"user","content":"continue"}`)
+	payload := []byte(`{"model":"gemini-3.6-flash-high","input":[` + strings.Join(items, ",") + `]}`)
+	inputBefore := bytes.Clone(payload)
+
+	for _, host := range []sdktranslator.PluginHooks{nil, pluginhost.New()} {
+		sdktranslator.SetPluginHooks(host)
+		for _, stream := range []bool{false, true} {
+			for _, route := range []string{"envelope", "api-key", "api-key-compat"} {
+				logs.Reset()
+				var base, work []byte
+				if route == "envelope" {
+					req := sdktranslator.RequestEnvelope{Format: sdktranslator.FormatOpenAIResponse, Model: "gemini-3.6-flash-high", Stream: stream}
+					base, work = helps.TranslateRequestEnvelopePairWithCodexMultiAgentV2(t.Context(), nil, &config.Config{}, sdktranslator.FormatOpenAIResponse, sdktranslator.FormatAntigravity, req, payload, payload)
+				} else {
+					base, work = helps.TranslateRequestPairWithAPIKeyModelCompatibility(t.Context(), nil, &config.Config{}, sdktranslator.FormatOpenAIResponse, sdktranslator.FormatAntigravity, "gemini-3.6-flash-high", payload, payload, stream, route == "api-key-compat")
+				}
+				const message = "gemini request: sanitized 9 thoughtSignatures"
+				if count := strings.Count(logs.String(), message); count != 1 {
+					t.Fatalf("host=%T stream=%v route=%s: sanitation logs = %d, want 1; logs=%s", host, stream, route, count, logs.String())
+				}
+				if !bytes.Equal(base, work) {
+					t.Fatal("same input produced different translations")
+				}
+				parts := gjson.GetBytes(work, "request.contents.0.parts").Array()
+				if len(parts) != 10 || parts[0].Get("thoughtSignature").String() != "skip_thought_signature_validator" {
+					t.Fatalf("unexpected parallel tool call layout: %s", work)
+				}
+				for _, part := range parts[1:] {
+					if part.Get("thoughtSignature").Exists() {
+						t.Fatal("sibling tool call still has a signature")
+					}
+				}
+				baselineBefore := bytes.Clone(base)
+				work[0] = 'X'
+				if !bytes.Equal(base, baselineBefore) || !bytes.Equal(payload, inputBefore) {
+					t.Fatal("working buffer aliases the baseline or input")
+				}
+			}
+		}
+	}
+}