feat(executor): add HttpRequest method with credential injection for GitHub Copilot and Kiro executors

feat(codex): add subscription date fields to ID token claims

README.md

@@ -19,17 +19,6 @@ This project only accepts pull requests that relate to third-party provider supp
 
 If you need to submit any non-third-party provider changes, please open them against the mainline repository.
 
-## More choices
-
-Those projects are ports of CLIProxyAPI or inspired by it:
-
-### [9Router](https://github.com/decolua/9router)
-
-A Next.js implementation inspired by CLIProxyAPI, easy to install and use, built from scratch with format translation (OpenAI/Claude/Gemini/Ollama), combo system with auto-fallback, multi-account management with exponential backoff, a Next.js web dashboard, and support for CLI tools (Cursor, Claude Code, Cline, RooCode) - no API keys needed.
-
-> [!NOTE]  
-> If you have developed a port of CLIProxyAPI or a project inspired by it, please open a PR to add it to this list.
-
 ## License
 
 This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.

README_CN.md

@@ -19,17 +19,6 @@
 
 如果需要提交任何非第三方供应商支持的 Pull Request，请提交到主线版本。
 
-## 更多选择
-
-以下项目是 CLIProxyAPI 的移植版或受其启发：
-
-### [9Router](https://github.com/decolua/9router)
-
-基于 Next.js 的实现，灵感来自 CLIProxyAPI，易于安装使用；自研格式转换（OpenAI/Claude/Gemini/Ollama）、组合系统与自动回退、多账户管理（指数退避）、Next.js Web 控制台，并支持 Cursor、Claude Code、Cline、RooCode 等 CLI 工具，无需 API 密钥。
-
-> [!NOTE]  
-> 如果你开发了 CLIProxyAPI 的移植或衍生项目，请提交 PR 将其添加到此列表中。
-
 ## 许可证
 
 此项目根据 MIT 许可证授权 - 有关详细信息，请参阅 [LICENSE](LICENSE) 文件。

config.example.yaml

@@ -215,7 +215,7 @@ ws-auth: false
 
 # Global OAuth model name mappings (per channel)
 # These mappings rename model IDs for both model listing and request routing.
-# Supported channels: gemini-cli, vertex, aistudio, antigravity, claude, codex, qwen, iflow.
+# Supported channels: gemini-cli, vertex, aistudio, antigravity, claude, codex, qwen, iflow, kiro.
 # NOTE: Mappings do not apply to gemini-api-key, codex-api-key, claude-api-key, openai-compatibility, vertex-api-key, or ampcode.
 # oauth-model-mappings:
 #   gemini-cli:
@@ -243,6 +243,9 @@ ws-auth: false
 #   iflow:
 #     - name: "glm-4.7"
 #       alias: "glm-god"
+#   kiro:
+#     - name: "kiro-claude-opus-4-5"
+#       alias: "op45"
 
 # OAuth provider excluded models
 # oauth-excluded-models:

docker-build.sh

@@ -152,16 +152,16 @@ case "$choice" in
     # Build and start the services with a local-only image tag
     export CLI_PROXY_IMAGE="cli-proxy-api:local"
 
-    if [[ "${WITH_USAGE}" == "true" ]]; then
-      export_stats
-    fi
-
     echo "Building the Docker image..."
     docker compose build \
       --build-arg VERSION="${VERSION}" \
       --build-arg COMMIT="${COMMIT}" \
       --build-arg BUILD_DATE="${BUILD_DATE}"
 
+    if [[ "${WITH_USAGE}" == "true" ]]; then
+      export_stats
+    fi
+
     echo "Starting the services..."
     docker compose up -d --remove-orphans --pull never
 

examples/custom-provider/main.go

@@ -14,6 +14,7 @@ import (
 	"bytes"
 	"context"
 	"errors"
+	"fmt"
 	"io"
 	"net/http"
 	"net/url"
@@ -122,7 +123,9 @@ func (MyExecutor) Execute(ctx context.Context, a *coreauth.Auth, req clipexec.Re
 	httpReq.Header.Set("Content-Type", "application/json")
 
 	// Inject credentials via PrepareRequest hook.
-	_ = (MyExecutor{}).PrepareRequest(httpReq, a)
+	if errPrep := (MyExecutor{}).PrepareRequest(httpReq, a); errPrep != nil {
+		return clipexec.Response{}, errPrep
+	}
 
 	resp, errDo := client.Do(httpReq)
 	if errDo != nil {
@@ -130,13 +133,28 @@ func (MyExecutor) Execute(ctx context.Context, a *coreauth.Auth, req clipexec.Re
 	}
 	defer func() {
 		if errClose := resp.Body.Close(); errClose != nil {
-			// Best-effort close; log if needed in real projects.
+			fmt.Fprintf(os.Stderr, "close response body error: %v\n", errClose)
 		}
 	}()
 	body, _ := io.ReadAll(resp.Body)
 	return clipexec.Response{Payload: body}, nil
 }
 
+func (MyExecutor) HttpRequest(ctx context.Context, a *coreauth.Auth, req *http.Request) (*http.Response, error) {
+	if req == nil {
+		return nil, fmt.Errorf("myprov executor: request is nil")
+	}
+	if ctx == nil {
+		ctx = req.Context()
+	}
+	httpReq := req.WithContext(ctx)
+	if errPrep := (MyExecutor{}).PrepareRequest(httpReq, a); errPrep != nil {
+		return nil, errPrep
+	}
+	client := buildHTTPClient(a)
+	return client.Do(httpReq)
+}
+
 func (MyExecutor) CountTokens(context.Context, *coreauth.Auth, clipexec.Request, clipexec.Options) (clipexec.Response, error) {
 	return clipexec.Response{}, errors.New("count tokens not implemented")
 }
@@ -199,8 +217,8 @@ func main() {
 	ctx, cancel := context.WithCancel(context.Background())
 	defer cancel()
 
-	if err := svc.Run(ctx); err != nil && !errors.Is(err, context.Canceled) {
-		panic(err)
+	if errRun := svc.Run(ctx); errRun != nil && !errors.Is(errRun, context.Canceled) {
+		panic(errRun)
 	}
 	_ = os.Stderr // keep os import used (demo only)
 	_ = time.Second

examples/http-request/main.go

@@ -0,0 +1,140 @@
+// Package main demonstrates how to use coreauth.Manager.HttpRequest/NewHttpRequest
+// to execute arbitrary HTTP requests with provider credentials injected.
+//
+// This example registers a minimal custom executor that injects an Authorization
+// header from auth.Attributes["api_key"], then performs two requests against
+// httpbin.org to show the injected headers.
+package main
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+
+	coreauth "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/auth"
+	clipexec "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/executor"
+	log "github.com/sirupsen/logrus"
+)
+
+const providerKey = "echo"
+
+// EchoExecutor is a minimal provider implementation for demonstration purposes.
+type EchoExecutor struct{}
+
+func (EchoExecutor) Identifier() string { return providerKey }
+
+func (EchoExecutor) PrepareRequest(req *http.Request, auth *coreauth.Auth) error {
+	if req == nil || auth == nil {
+		return nil
+	}
+	if auth.Attributes != nil {
+		if apiKey := strings.TrimSpace(auth.Attributes["api_key"]); apiKey != "" {
+			req.Header.Set("Authorization", "Bearer "+apiKey)
+		}
+	}
+	return nil
+}
+
+func (EchoExecutor) HttpRequest(ctx context.Context, auth *coreauth.Auth, req *http.Request) (*http.Response, error) {
+	if req == nil {
+		return nil, fmt.Errorf("echo executor: request is nil")
+	}
+	if ctx == nil {
+		ctx = req.Context()
+	}
+	httpReq := req.WithContext(ctx)
+	if errPrep := (EchoExecutor{}).PrepareRequest(httpReq, auth); errPrep != nil {
+		return nil, errPrep
+	}
+	return http.DefaultClient.Do(httpReq)
+}
+
+func (EchoExecutor) Execute(context.Context, *coreauth.Auth, clipexec.Request, clipexec.Options) (clipexec.Response, error) {
+	return clipexec.Response{}, errors.New("echo executor: Execute not implemented")
+}
+
+func (EchoExecutor) ExecuteStream(context.Context, *coreauth.Auth, clipexec.Request, clipexec.Options) (<-chan clipexec.StreamChunk, error) {
+	return nil, errors.New("echo executor: ExecuteStream not implemented")
+}
+
+func (EchoExecutor) Refresh(context.Context, *coreauth.Auth) (*coreauth.Auth, error) {
+	return nil, errors.New("echo executor: Refresh not implemented")
+}
+
+func (EchoExecutor) CountTokens(context.Context, *coreauth.Auth, clipexec.Request, clipexec.Options) (clipexec.Response, error) {
+	return clipexec.Response{}, errors.New("echo executor: CountTokens not implemented")
+}
+
+func main() {
+	log.SetLevel(log.InfoLevel)
+
+	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
+	defer cancel()
+
+	core := coreauth.NewManager(nil, nil, nil)
+	core.RegisterExecutor(EchoExecutor{})
+
+	auth := &coreauth.Auth{
+		ID:       "demo-echo",
+		Provider: providerKey,
+		Attributes: map[string]string{
+			"api_key": "demo-api-key",
+		},
+	}
+
+	// Example 1: Build a prepared request and execute it using your own http.Client.
+	reqPrepared, errReqPrepared := core.NewHttpRequest(
+		ctx,
+		auth,
+		http.MethodGet,
+		"https://httpbin.org/anything",
+		nil,
+		http.Header{"X-Example": []string{"prepared"}},
+	)
+	if errReqPrepared != nil {
+		panic(errReqPrepared)
+	}
+	respPrepared, errDoPrepared := http.DefaultClient.Do(reqPrepared)
+	if errDoPrepared != nil {
+		panic(errDoPrepared)
+	}
+	defer func() {
+		if errClose := respPrepared.Body.Close(); errClose != nil {
+			log.Errorf("close response body error: %v", errClose)
+		}
+	}()
+	bodyPrepared, errReadPrepared := io.ReadAll(respPrepared.Body)
+	if errReadPrepared != nil {
+		panic(errReadPrepared)
+	}
+	fmt.Printf("Prepared request status: %d\n%s\n\n", respPrepared.StatusCode, bodyPrepared)
+
+	// Example 2: Execute a raw request via core.HttpRequest (auto inject + do).
+	rawBody := []byte(`{"hello":"world"}`)
+	rawReq, errRawReq := http.NewRequestWithContext(ctx, http.MethodPost, "https://httpbin.org/anything", bytes.NewReader(rawBody))
+	if errRawReq != nil {
+		panic(errRawReq)
+	}
+	rawReq.Header.Set("Content-Type", "application/json")
+	rawReq.Header.Set("X-Example", "executed")
+
+	respExec, errDoExec := core.HttpRequest(ctx, auth, rawReq)
+	if errDoExec != nil {
+		panic(errDoExec)
+	}
+	defer func() {
+		if errClose := respExec.Body.Close(); errClose != nil {
+			log.Errorf("close response body error: %v", errClose)
+		}
+	}()
+	bodyExec, errReadExec := io.ReadAll(respExec.Body)
+	if errReadExec != nil {
+		panic(errReadExec)
+	}
+	fmt.Printf("Manager HttpRequest status: %d\n%s\n", respExec.StatusCode, bodyExec)
+}

internal/api/handlers/management/auth_files.go

@@ -464,6 +464,12 @@ func extractCodexIDTokenClaims(auth *coreauth.Auth) gin.H {
 	if v := strings.TrimSpace(claims.CodexAuthInfo.ChatgptPlanType); v != "" {
 		result["plan_type"] = v
 	}
+	if v := claims.CodexAuthInfo.ChatgptSubscriptionActiveStart; v != nil {
+		result["chatgpt_subscription_active_start"] = v
+	}
+	if v := claims.CodexAuthInfo.ChatgptSubscriptionActiveUntil; v != nil {
+		result["chatgpt_subscription_active_until"] = v
+	}
 
 	if len(result) == 0 {
 		return nil

internal/api/handlers/management/handler.go

@@ -24,8 +24,15 @@ import (
 type attemptInfo struct {
 	count        int
 	blockedUntil time.Time
+	lastActivity time.Time // track last activity for cleanup
 }
 
+// attemptCleanupInterval controls how often stale IP entries are purged
+const attemptCleanupInterval = 1 * time.Hour
+
+// attemptMaxIdleTime controls how long an IP can be idle before cleanup
+const attemptMaxIdleTime = 2 * time.Hour
+
 // Handler aggregates config reference, persistence path and helpers.
 type Handler struct {
 	cfg                 *config.Config
@@ -47,7 +54,7 @@ func NewHandler(cfg *config.Config, configFilePath string, manager *coreauth.Man
 	envSecret, _ := os.LookupEnv("MANAGEMENT_PASSWORD")
 	envSecret = strings.TrimSpace(envSecret)
 
-	return &Handler{
+	h := &Handler{
 		cfg:                 cfg,
 		configFilePath:      configFilePath,
 		failedAttempts:      make(map[string]*attemptInfo),
@@ -57,6 +64,38 @@ func NewHandler(cfg *config.Config, configFilePath string, manager *coreauth.Man
 		allowRemoteOverride: envSecret != "",
 		envSecret:           envSecret,
 	}
+	h.startAttemptCleanup()
+	return h
+}
+
+// startAttemptCleanup launches a background goroutine that periodically
+// removes stale IP entries from failedAttempts to prevent memory leaks.
+func (h *Handler) startAttemptCleanup() {
+	go func() {
+		ticker := time.NewTicker(attemptCleanupInterval)
+		defer ticker.Stop()
+		for range ticker.C {
+			h.purgeStaleAttempts()
+		}
+	}()
+}
+
+// purgeStaleAttempts removes IP entries that have been idle beyond attemptMaxIdleTime
+// and whose ban (if any) has expired.
+func (h *Handler) purgeStaleAttempts() {
+	now := time.Now()
+	h.attemptsMu.Lock()
+	defer h.attemptsMu.Unlock()
+	for ip, ai := range h.failedAttempts {
+		// Skip if still banned
+		if !ai.blockedUntil.IsZero() && now.Before(ai.blockedUntil) {
+			continue
+		}
+		// Remove if idle too long
+		if now.Sub(ai.lastActivity) > attemptMaxIdleTime {
+			delete(h.failedAttempts, ip)
+		}
+	}
 }
 
 // NewHandler creates a new management handler instance.
@@ -149,6 +188,7 @@ func (h *Handler) Middleware() gin.HandlerFunc {
 					h.failedAttempts[clientIP] = aip
 				}
 				aip.count++
+				aip.lastActivity = time.Now()
 				if aip.count >= maxFailures {
 					aip.blockedUntil = time.Now().Add(banDuration)
 					aip.count = 0

internal/api/modules/amp/response_rewriter.go

@@ -125,7 +125,30 @@ func (rw *ResponseRewriter) Flush() {
 var modelFieldPaths = []string{"model", "modelVersion", "response.modelVersion", "message.model"}
 
 // rewriteModelInResponse replaces all occurrences of the mapped model with the original model in JSON
+// It also suppresses "thinking" blocks if "tool_use" is present to ensure Amp client compatibility
 func (rw *ResponseRewriter) rewriteModelInResponse(data []byte) []byte {
+	// 1. Amp Compatibility: Suppress thinking blocks if tool use is detected
+	// The Amp client struggles when both thinking and tool_use blocks are present
+	if gjson.GetBytes(data, `content.#(type=="tool_use")`).Exists() {
+		filtered := gjson.GetBytes(data, `content.#(type!="thinking")#`)
+		if filtered.Exists() {
+			originalCount := gjson.GetBytes(data, "content.#").Int()
+			filteredCount := filtered.Get("#").Int()
+
+			if originalCount > filteredCount {
+				var err error
+				data, err = sjson.SetBytes(data, "content", filtered.Value())
+				if err != nil {
+					log.Warnf("Amp ResponseRewriter: failed to suppress thinking blocks: %v", err)
+				} else {
+					log.Debugf("Amp ResponseRewriter: Suppressed %d thinking blocks due to tool usage", originalCount-filteredCount)
+					// Log the result for verification
+					log.Debugf("Amp ResponseRewriter: Resulting content: %s", gjson.GetBytes(data, "content").String())
+				}
+			}
+		}
+	}
+
 	if rw.originalModel == "" {
 		return data
 	}

internal/cache/signature_cache.go

@@ -26,11 +26,17 @@ const (
 
 	// MinValidSignatureLen is the minimum length for a signature to be considered valid
 	MinValidSignatureLen = 50
+
+	// SessionCleanupInterval controls how often stale sessions are purged
+	SessionCleanupInterval = 10 * time.Minute
 )
 
 // signatureCache stores signatures by sessionId -> textHash -> SignatureEntry
 var signatureCache sync.Map
 
+// sessionCleanupOnce ensures the background cleanup goroutine starts only once
+var sessionCleanupOnce sync.Once
+
 // sessionCache is the inner map type
 type sessionCache struct {
 	mu      sync.RWMutex
@@ -45,6 +51,9 @@ func hashText(text string) string {
 
 // getOrCreateSession gets or creates a session cache
 func getOrCreateSession(sessionID string) *sessionCache {
+	// Start background cleanup on first access
+	sessionCleanupOnce.Do(startSessionCleanup)
+
 	if val, ok := signatureCache.Load(sessionID); ok {
 		return val.(*sessionCache)
 	}
@@ -53,6 +62,40 @@ func getOrCreateSession(sessionID string) *sessionCache {
 	return actual.(*sessionCache)
 }
 
+// startSessionCleanup launches a background goroutine that periodically
+// removes sessions where all entries have expired.
+func startSessionCleanup() {
+	go func() {
+		ticker := time.NewTicker(SessionCleanupInterval)
+		defer ticker.Stop()
+		for range ticker.C {
+			purgeExpiredSessions()
+		}
+	}()
+}
+
+// purgeExpiredSessions removes sessions with no valid (non-expired) entries.
+func purgeExpiredSessions() {
+	now := time.Now()
+	signatureCache.Range(func(key, value any) bool {
+		sc := value.(*sessionCache)
+		sc.mu.Lock()
+		// Remove expired entries
+		for k, entry := range sc.entries {
+			if now.Sub(entry.Timestamp) > SignatureCacheTTL {
+				delete(sc.entries, k)
+			}
+		}
+		isEmpty := len(sc.entries) == 0
+		sc.mu.Unlock()
+		// Remove session if empty
+		if isEmpty {
+			signatureCache.Delete(key)
+		}
+		return true
+	})
+}
+
 // CacheSignature stores a thinking signature for a given session and text.
 // Used for Claude models that require signed thinking blocks in multi-turn conversations.
 func CacheSignature(sessionID, text, signature string) {

internal/runtime/executor/aistudio_executor.go

@@ -8,6 +8,7 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
+	"io"
 	"net/http"
 	"net/url"
 	"strings"
@@ -50,6 +51,64 @@ func (e *AIStudioExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth)
 	return nil
 }
 
+// HttpRequest forwards an arbitrary HTTP request through the websocket relay.
+func (e *AIStudioExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth, req *http.Request) (*http.Response, error) {
+	if req == nil {
+		return nil, fmt.Errorf("aistudio executor: request is nil")
+	}
+	if ctx == nil {
+		ctx = req.Context()
+	}
+	if e.relay == nil {
+		return nil, fmt.Errorf("aistudio executor: ws relay is nil")
+	}
+	if auth == nil || auth.ID == "" {
+		return nil, fmt.Errorf("aistudio executor: missing auth")
+	}
+	httpReq := req.WithContext(ctx)
+	if httpReq.URL == nil || strings.TrimSpace(httpReq.URL.String()) == "" {
+		return nil, fmt.Errorf("aistudio executor: request URL is empty")
+	}
+
+	var body []byte
+	if httpReq.Body != nil {
+		b, errRead := io.ReadAll(httpReq.Body)
+		if errRead != nil {
+			return nil, errRead
+		}
+		body = b
+		httpReq.Body = io.NopCloser(bytes.NewReader(b))
+	}
+
+	wsReq := &wsrelay.HTTPRequest{
+		Method:  httpReq.Method,
+		URL:     httpReq.URL.String(),
+		Headers: httpReq.Header.Clone(),
+		Body:    body,
+	}
+	wsResp, errRelay := e.relay.NonStream(ctx, auth.ID, wsReq)
+	if errRelay != nil {
+		return nil, errRelay
+	}
+	if wsResp == nil {
+		return nil, fmt.Errorf("aistudio executor: ws response is nil")
+	}
+
+	statusText := http.StatusText(wsResp.Status)
+	if statusText == "" {
+		statusText = "Unknown"
+	}
+	resp := &http.Response{
+		StatusCode:    wsResp.Status,
+		Status:        fmt.Sprintf("%d %s", wsResp.Status, statusText),
+		Header:        wsResp.Headers.Clone(),
+		Body:          io.NopCloser(bytes.NewReader(wsResp.Body)),
+		ContentLength: int64(len(wsResp.Body)),
+		Request:       httpReq,
+	}
+	return resp, nil
+}
+
 // Execute performs a non-streaming request to the AI Studio API.
 func (e *AIStudioExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)

internal/runtime/executor/antigravity_executor.go

@@ -10,6 +10,7 @@ import (
 	"crypto/sha256"
 	"encoding/binary"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"io"
 	"math/rand"
@@ -45,7 +46,7 @@ const (
 	defaultAntigravityAgent        = "antigravity/1.104.0 darwin/arm64"
 	antigravityAuthType            = "antigravity"
 	refreshSkew                    = 3000 * time.Second
-	tokenRefreshTimeout            = 30 * time.Second
+	systemInstruction              = "You are Antigravity, a powerful agentic AI coding assistant designed by the Google Deepmind team working on Advanced Agentic Coding.You are pair programming with a USER to solve their coding task. The task may require creating a new codebase, modifying or debugging an existing codebase, or simply answering a question.**Absolute paths only****Proactiveness**"
 )
 
 var (
@@ -72,13 +73,42 @@ func NewAntigravityExecutor(cfg *config.Config) *AntigravityExecutor {
 // Identifier returns the executor identifier.
 func (e *AntigravityExecutor) Identifier() string { return antigravityAuthType }
 
-// PrepareRequest prepares the HTTP request for execution (no-op for Antigravity).
-func (e *AntigravityExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error { return nil }
+// PrepareRequest injects Antigravity credentials into the outgoing HTTP request.
+func (e *AntigravityExecutor) PrepareRequest(req *http.Request, auth *cliproxyauth.Auth) error {
+	if req == nil {
+		return nil
+	}
+	token, _, errToken := e.ensureAccessToken(req.Context(), auth)
+	if errToken != nil {
+		return errToken
+	}
+	if strings.TrimSpace(token) == "" {
+		return statusErr{code: http.StatusUnauthorized, msg: "missing access token"}
+	}
+	req.Header.Set("Authorization", "Bearer "+token)
+	return nil
+}
+
+// HttpRequest injects Antigravity credentials into the request and executes it.
+func (e *AntigravityExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth, req *http.Request) (*http.Response, error) {
+	if req == nil {
+		return nil, fmt.Errorf("antigravity executor: request is nil")
+	}
+	if ctx == nil {
+		ctx = req.Context()
+	}
+	httpReq := req.WithContext(ctx)
+	if err := e.PrepareRequest(httpReq, auth); err != nil {
+		return nil, err
+	}
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	return httpClient.Do(httpReq)
+}
 
 // Execute performs a non-streaming request to the Antigravity API.
 func (e *AntigravityExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
 	isClaude := strings.Contains(strings.ToLower(req.Model), "claude")
-	if isClaude {
+	if isClaude || strings.Contains(req.Model, "gemini-3-pro") {
 		return e.executeClaudeNonStream(ctx, auth, req, opts)
 	}
 
@@ -125,6 +155,9 @@ func (e *AntigravityExecutor) Execute(ctx context.Context, auth *cliproxyauth.Au
 		httpResp, errDo := httpClient.Do(httpReq)
 		if errDo != nil {
 			recordAPIResponseError(ctx, e.cfg, errDo)
+			if errors.Is(errDo, context.Canceled) || errors.Is(errDo, context.DeadlineExceeded) {
+				return resp, errDo
+			}
 			lastStatus = 0
 			lastBody = nil
 			lastErr = errDo
@@ -157,7 +190,13 @@ func (e *AntigravityExecutor) Execute(ctx context.Context, auth *cliproxyauth.Au
 				log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
 				continue
 			}
-			err = statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+			sErr := statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+			if httpResp.StatusCode == http.StatusTooManyRequests {
+				if retryAfter, parseErr := parseRetryDelay(bodyBytes); parseErr == nil && retryAfter != nil {
+					sErr.retryAfter = retryAfter
+				}
+			}
+			err = sErr
 			return resp, err
 		}
 
@@ -171,7 +210,13 @@ func (e *AntigravityExecutor) Execute(ctx context.Context, auth *cliproxyauth.Au
 
 	switch {
 	case lastStatus != 0:
-		err = statusErr{code: lastStatus, msg: string(lastBody)}
+		sErr := statusErr{code: lastStatus, msg: string(lastBody)}
+		if lastStatus == http.StatusTooManyRequests {
+			if retryAfter, parseErr := parseRetryDelay(lastBody); parseErr == nil && retryAfter != nil {
+				sErr.retryAfter = retryAfter
+			}
+		}
+		err = sErr
 	case lastErr != nil:
 		err = lastErr
 	default:
@@ -225,6 +270,9 @@ func (e *AntigravityExecutor) executeClaudeNonStream(ctx context.Context, auth *
 		httpResp, errDo := httpClient.Do(httpReq)
 		if errDo != nil {
 			recordAPIResponseError(ctx, e.cfg, errDo)
+			if errors.Is(errDo, context.Canceled) || errors.Is(errDo, context.DeadlineExceeded) {
+				return resp, errDo
+			}
 			lastStatus = 0
 			lastBody = nil
 			lastErr = errDo
@@ -243,6 +291,14 @@ func (e *AntigravityExecutor) executeClaudeNonStream(ctx context.Context, auth *
 			}
 			if errRead != nil {
 				recordAPIResponseError(ctx, e.cfg, errRead)
+				if errors.Is(errRead, context.Canceled) || errors.Is(errRead, context.DeadlineExceeded) {
+					err = errRead
+					return resp, err
+				}
+				if errCtx := ctx.Err(); errCtx != nil {
+					err = errCtx
+					return resp, err
+				}
 				lastStatus = 0
 				lastBody = nil
 				lastErr = errRead
@@ -261,7 +317,13 @@ func (e *AntigravityExecutor) executeClaudeNonStream(ctx context.Context, auth *
 				log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
 				continue
 			}
-			err = statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+			sErr := statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+			if httpResp.StatusCode == http.StatusTooManyRequests {
+				if retryAfter, parseErr := parseRetryDelay(bodyBytes); parseErr == nil && retryAfter != nil {
+					sErr.retryAfter = retryAfter
+				}
+			}
+			err = sErr
 			return resp, err
 		}
 
@@ -326,7 +388,13 @@ func (e *AntigravityExecutor) executeClaudeNonStream(ctx context.Context, auth *
 
 	switch {
 	case lastStatus != 0:
-		err = statusErr{code: lastStatus, msg: string(lastBody)}
+		sErr := statusErr{code: lastStatus, msg: string(lastBody)}
+		if lastStatus == http.StatusTooManyRequests {
+			if retryAfter, parseErr := parseRetryDelay(lastBody); parseErr == nil && retryAfter != nil {
+				sErr.retryAfter = retryAfter
+			}
+		}
+		err = sErr
 	case lastErr != nil:
 		err = lastErr
 	default:
@@ -566,6 +634,9 @@ func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxya
 		httpResp, errDo := httpClient.Do(httpReq)
 		if errDo != nil {
 			recordAPIResponseError(ctx, e.cfg, errDo)
+			if errors.Is(errDo, context.Canceled) || errors.Is(errDo, context.DeadlineExceeded) {
+				return nil, errDo
+			}
 			lastStatus = 0
 			lastBody = nil
 			lastErr = errDo
@@ -584,6 +655,14 @@ func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxya
 			}
 			if errRead != nil {
 				recordAPIResponseError(ctx, e.cfg, errRead)
+				if errors.Is(errRead, context.Canceled) || errors.Is(errRead, context.DeadlineExceeded) {
+					err = errRead
+					return nil, err
+				}
+				if errCtx := ctx.Err(); errCtx != nil {
+					err = errCtx
+					return nil, err
+				}
 				lastStatus = 0
 				lastBody = nil
 				lastErr = errRead
@@ -602,7 +681,13 @@ func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxya
 				log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
 				continue
 			}
-			err = statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+			sErr := statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+			if httpResp.StatusCode == http.StatusTooManyRequests {
+				if retryAfter, parseErr := parseRetryDelay(bodyBytes); parseErr == nil && retryAfter != nil {
+					sErr.retryAfter = retryAfter
+				}
+			}
+			err = sErr
 			return nil, err
 		}
 
@@ -657,7 +742,13 @@ func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxya
 
 	switch {
 	case lastStatus != 0:
-		err = statusErr{code: lastStatus, msg: string(lastBody)}
+		sErr := statusErr{code: lastStatus, msg: string(lastBody)}
+		if lastStatus == http.StatusTooManyRequests {
+			if retryAfter, parseErr := parseRetryDelay(lastBody); parseErr == nil && retryAfter != nil {
+				sErr.retryAfter = retryAfter
+			}
+		}
+		err = sErr
 	case lastErr != nil:
 		err = lastErr
 	default:
@@ -760,6 +851,9 @@ func (e *AntigravityExecutor) CountTokens(ctx context.Context, auth *cliproxyaut
 		httpResp, errDo := httpClient.Do(httpReq)
 		if errDo != nil {
 			recordAPIResponseError(ctx, e.cfg, errDo)
+			if errors.Is(errDo, context.Canceled) || errors.Is(errDo, context.DeadlineExceeded) {
+				return cliproxyexecutor.Response{}, errDo
+			}
 			lastStatus = 0
 			lastBody = nil
 			lastErr = errDo
@@ -794,12 +888,24 @@ func (e *AntigravityExecutor) CountTokens(ctx context.Context, auth *cliproxyaut
 			log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
 			continue
 		}
-		return cliproxyexecutor.Response{}, statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+		sErr := statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+		if httpResp.StatusCode == http.StatusTooManyRequests {
+			if retryAfter, parseErr := parseRetryDelay(bodyBytes); parseErr == nil && retryAfter != nil {
+				sErr.retryAfter = retryAfter
+			}
+		}
+		return cliproxyexecutor.Response{}, sErr
 	}
 
 	switch {
 	case lastStatus != 0:
-		return cliproxyexecutor.Response{}, statusErr{code: lastStatus, msg: string(lastBody)}
+		sErr := statusErr{code: lastStatus, msg: string(lastBody)}
+		if lastStatus == http.StatusTooManyRequests {
+			if retryAfter, parseErr := parseRetryDelay(lastBody); parseErr == nil && retryAfter != nil {
+				sErr.retryAfter = retryAfter
+			}
+		}
+		return cliproxyexecutor.Response{}, sErr
 	case lastErr != nil:
 		return cliproxyexecutor.Response{}, lastErr
 	default:
@@ -836,6 +942,9 @@ func FetchAntigravityModels(ctx context.Context, auth *cliproxyauth.Auth, cfg *c
 
 		httpResp, errDo := httpClient.Do(httpReq)
 		if errDo != nil {
+			if errors.Is(errDo, context.Canceled) || errors.Is(errDo, context.DeadlineExceeded) {
+				return nil
+			}
 			if idx+1 < len(baseURLs) {
 				log.Debugf("antigravity executor: models request error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
 				continue
@@ -951,7 +1060,7 @@ func (e *AntigravityExecutor) refreshToken(ctx context.Context, auth *cliproxyau
 	httpReq.Header.Set("User-Agent", defaultAntigravityAgent)
 	httpReq.Header.Set("Content-Type", "application/x-www-form-urlencoded")
 
-	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, tokenRefreshTimeout)
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
 	httpResp, errDo := httpClient.Do(httpReq)
 	if errDo != nil {
 		return auth, errDo
@@ -968,7 +1077,13 @@ func (e *AntigravityExecutor) refreshToken(ctx context.Context, auth *cliproxyau
 	}
 
 	if httpResp.StatusCode < http.StatusOK || httpResp.StatusCode >= http.StatusMultipleChoices {
-		return auth, statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+		sErr := statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+		if httpResp.StatusCode == http.StatusTooManyRequests {
+			if retryAfter, parseErr := parseRetryDelay(bodyBytes); parseErr == nil && retryAfter != nil {
+				sErr.retryAfter = retryAfter
+			}
+		}
+		return auth, sErr
 	}
 
 	var tokenResp struct {
@@ -1048,6 +1163,19 @@ func (e *AntigravityExecutor) buildRequest(ctx context.Context, auth *cliproxyau
 		payload = []byte(strJSON)
 	}
 
+	if strings.Contains(modelName, "claude") || strings.Contains(modelName, "gemini-3-pro-preview") {
+		systemInstructionPartsResult := gjson.GetBytes(payload, "request.systemInstruction.parts")
+		payload, _ = sjson.SetBytes(payload, "request.systemInstruction.role", "user")
+		payload, _ = sjson.SetBytes(payload, "request.systemInstruction.parts.0.text", systemInstruction)
+		payload, _ = sjson.SetBytes(payload, "request.systemInstruction.parts.1.text", fmt.Sprintf("Please ignore following [ignore]%s[/ignore]", systemInstruction))
+
+		if systemInstructionPartsResult.Exists() && systemInstructionPartsResult.IsArray() {
+			for _, partResult := range systemInstructionPartsResult.Array() {
+				payload, _ = sjson.SetRawBytes(payload, "request.systemInstruction.parts.-1", []byte(partResult.Raw))
+			}
+		}
+	}
+
 	httpReq, errReq := http.NewRequestWithContext(ctx, http.MethodPost, requestURL.String(), bytes.NewReader(payload))
 	if errReq != nil {
 		return nil, errReq
@@ -1182,8 +1310,8 @@ func antigravityBaseURLFallbackOrder(auth *cliproxyauth.Auth) []string {
 		return []string{base}
 	}
 	return []string{
-		antigravityBaseURLDaily,
 		antigravitySandboxBaseURLDaily,
+		antigravityBaseURLDaily,
 		antigravityBaseURLProd,
 	}
 }
@@ -1211,6 +1339,7 @@ func resolveCustomAntigravityBaseURL(auth *cliproxyauth.Auth) string {
 func geminiToAntigravity(modelName string, payload []byte, projectID string) []byte {
 	template, _ := sjson.Set(string(payload), "model", modelName)
 	template, _ = sjson.Set(template, "userAgent", "antigravity")
+	template, _ = sjson.Set(template, "requestType", "agent")
 
 	// Use real project ID from auth if available, otherwise generate random (legacy fallback)
 	if projectID != "" {

internal/runtime/executor/cache_helpers.go

@@ -10,29 +10,69 @@ type codexCache struct {
 	Expire time.Time
 }
 
+// codexCacheMap stores prompt cache IDs keyed by model+user_id.
+// Protected by codexCacheMu. Entries expire after 1 hour.
 var (
-	codexCacheMap   = map[string]codexCache{}
-	codexCacheMutex sync.RWMutex
+	codexCacheMap = make(map[string]codexCache)
+	codexCacheMu  sync.RWMutex
 )
 
-// getCodexCache safely retrieves a cache entry
+// codexCacheCleanupInterval controls how often expired entries are purged.
+const codexCacheCleanupInterval = 15 * time.Minute
+
+// codexCacheCleanupOnce ensures the background cleanup goroutine starts only once.
+var codexCacheCleanupOnce sync.Once
+
+// startCodexCacheCleanup launches a background goroutine that periodically
+// removes expired entries from codexCacheMap to prevent memory leaks.
+func startCodexCacheCleanup() {
+	go func() {
+		ticker := time.NewTicker(codexCacheCleanupInterval)
+		defer ticker.Stop()
+
+		for range ticker.C {
+			purgeExpiredCodexCache()
+		}
+	}()
+}
+
+// purgeExpiredCodexCache removes entries that have expired.
+func purgeExpiredCodexCache() {
+	now := time.Now()
+
+	codexCacheMu.Lock()
+	defer codexCacheMu.Unlock()
+
+	for key, cache := range codexCacheMap {
+		if cache.Expire.Before(now) {
+			delete(codexCacheMap, key)
+		}
+	}
+}
+
+// getCodexCache retrieves a cached entry, returning ok=false if not found or expired.
 func getCodexCache(key string) (codexCache, bool) {
-	codexCacheMutex.RLock()
-	defer codexCacheMutex.RUnlock()
+	codexCacheCleanupOnce.Do(startCodexCacheCleanup)
+	codexCacheMu.RLock()
 	cache, ok := codexCacheMap[key]
-	return cache, ok
+	codexCacheMu.RUnlock()
+	if !ok || cache.Expire.Before(time.Now()) {
+		return codexCache{}, false
+	}
+	return cache, true
 }
 
-// setCodexCache safely sets a cache entry
+// setCodexCache stores a cache entry.
 func setCodexCache(key string, cache codexCache) {
-	codexCacheMutex.Lock()
-	defer codexCacheMutex.Unlock()
+	codexCacheCleanupOnce.Do(startCodexCacheCleanup)
+	codexCacheMu.Lock()
 	codexCacheMap[key] = cache
+	codexCacheMu.Unlock()
 }
 
-// deleteCodexCache safely deletes a cache entry
+// deleteCodexCache deletes a cache entry.
 func deleteCodexCache(key string) {
-	codexCacheMutex.Lock()
-	defer codexCacheMutex.Unlock()
+	codexCacheMu.Lock()
 	delete(codexCacheMap, key)
+	codexCacheMu.Unlock()
 }

internal/runtime/executor/claude_executor.go

@@ -35,11 +35,53 @@ type ClaudeExecutor struct {
 	cfg *config.Config
 }
 
+const claudeToolPrefix = "proxy_"
+
 func NewClaudeExecutor(cfg *config.Config) *ClaudeExecutor { return &ClaudeExecutor{cfg: cfg} }
 
 func (e *ClaudeExecutor) Identifier() string { return "claude" }
 
-func (e *ClaudeExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error { return nil }
+// PrepareRequest injects Claude credentials into the outgoing HTTP request.
+func (e *ClaudeExecutor) PrepareRequest(req *http.Request, auth *cliproxyauth.Auth) error {
+	if req == nil {
+		return nil
+	}
+	apiKey, _ := claudeCreds(auth)
+	if strings.TrimSpace(apiKey) == "" {
+		return nil
+	}
+	useAPIKey := auth != nil && auth.Attributes != nil && strings.TrimSpace(auth.Attributes["api_key"]) != ""
+	isAnthropicBase := req.URL != nil && strings.EqualFold(req.URL.Scheme, "https") && strings.EqualFold(req.URL.Host, "api.anthropic.com")
+	if isAnthropicBase && useAPIKey {
+		req.Header.Del("Authorization")
+		req.Header.Set("x-api-key", apiKey)
+	} else {
+		req.Header.Del("x-api-key")
+		req.Header.Set("Authorization", "Bearer "+apiKey)
+	}
+	var attrs map[string]string
+	if auth != nil {
+		attrs = auth.Attributes
+	}
+	util.ApplyCustomHeadersFromAttrs(req, attrs)
+	return nil
+}
+
+// HttpRequest injects Claude credentials into the request and executes it.
+func (e *ClaudeExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth, req *http.Request) (*http.Response, error) {
+	if req == nil {
+		return nil, fmt.Errorf("claude executor: request is nil")
+	}
+	if ctx == nil {
+		ctx = req.Context()
+	}
+	httpReq := req.WithContext(ctx)
+	if err := e.PrepareRequest(httpReq, auth); err != nil {
+		return nil, err
+	}
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	return httpClient.Do(httpReq)
+}
 
 func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
 	apiKey, baseURL := claudeCreds(auth)
@@ -81,9 +123,14 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	// Extract betas from body and convert to header
 	var extraBetas []string
 	extraBetas, body = extractAndRemoveBetas(body)
+	bodyForTranslation := body
+	bodyForUpstream := body
+	if isClaudeOAuthToken(apiKey) {
+		bodyForUpstream = applyClaudeToolPrefix(body, claudeToolPrefix)
+	}
 
 	url := fmt.Sprintf("%s/v1/messages?beta=true", baseURL)
-	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
+	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(bodyForUpstream))
 	if err != nil {
 		return resp, err
 	}
@@ -98,7 +145,7 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 		URL:       url,
 		Method:    http.MethodPost,
 		Headers:   httpReq.Header.Clone(),
-		Body:      body,
+		Body:      bodyForUpstream,
 		Provider:  e.Identifier(),
 		AuthID:    authID,
 		AuthLabel: authLabel,
@@ -152,8 +199,20 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	} else {
 		reporter.publish(ctx, parseClaudeUsage(data))
 	}
+	if isClaudeOAuthToken(apiKey) {
+		data = stripClaudeToolPrefixFromResponse(data, claudeToolPrefix)
+	}
 	var param any
-	out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), body, data, &param)
+	out := sdktranslator.TranslateNonStream(
+		ctx,
+		to,
+		from,
+		req.Model,
+		bytes.Clone(opts.OriginalRequest),
+		bodyForTranslation,
+		data,
+		&param,
+	)
 	resp = cliproxyexecutor.Response{Payload: []byte(out)}
 	return resp, nil
 }
@@ -193,9 +252,14 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	// Extract betas from body and convert to header
 	var extraBetas []string
 	extraBetas, body = extractAndRemoveBetas(body)
+	bodyForTranslation := body
+	bodyForUpstream := body
+	if isClaudeOAuthToken(apiKey) {
+		bodyForUpstream = applyClaudeToolPrefix(body, claudeToolPrefix)
+	}
 
 	url := fmt.Sprintf("%s/v1/messages?beta=true", baseURL)
-	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
+	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(bodyForUpstream))
 	if err != nil {
 		return nil, err
 	}
@@ -210,7 +274,7 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 		URL:       url,
 		Method:    http.MethodPost,
 		Headers:   httpReq.Header.Clone(),
-		Body:      body,
+		Body:      bodyForUpstream,
 		Provider:  e.Identifier(),
 		AuthID:    authID,
 		AuthLabel: authLabel,
@@ -263,6 +327,9 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 				if detail, ok := parseClaudeStreamUsage(line); ok {
 					reporter.publish(ctx, detail)
 				}
+				if isClaudeOAuthToken(apiKey) {
+					line = stripClaudeToolPrefixFromStreamLine(line, claudeToolPrefix)
+				}
 				// Forward the line as-is to preserve SSE format
 				cloned := make([]byte, len(line)+1)
 				copy(cloned, line)
@@ -287,7 +354,19 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 			if detail, ok := parseClaudeStreamUsage(line); ok {
 				reporter.publish(ctx, detail)
 			}
-			chunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), body, bytes.Clone(line), &param)
+			if isClaudeOAuthToken(apiKey) {
+				line = stripClaudeToolPrefixFromStreamLine(line, claudeToolPrefix)
+			}
+			chunks := sdktranslator.TranslateStream(
+				ctx,
+				to,
+				from,
+				req.Model,
+				bytes.Clone(opts.OriginalRequest),
+				bodyForTranslation,
+				bytes.Clone(line),
+				&param,
+			)
 			for i := range chunks {
 				out <- cliproxyexecutor.StreamChunk{Payload: []byte(chunks[i])}
 			}
@@ -326,6 +405,9 @@ func (e *ClaudeExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Aut
 	// Extract betas from body and convert to header (for count_tokens too)
 	var extraBetas []string
 	extraBetas, body = extractAndRemoveBetas(body)
+	if isClaudeOAuthToken(apiKey) {
+		body = applyClaudeToolPrefix(body, claudeToolPrefix)
+	}
 
 	url := fmt.Sprintf("%s/v1/messages/count_tokens?beta=true", baseURL)
 	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
@@ -770,3 +852,107 @@ func checkSystemInstructions(payload []byte) []byte {
 	}
 	return payload
 }
+
+func isClaudeOAuthToken(apiKey string) bool {
+	return strings.Contains(apiKey, "sk-ant-oat")
+}
+
+func applyClaudeToolPrefix(body []byte, prefix string) []byte {
+	if prefix == "" {
+		return body
+	}
+
+	if tools := gjson.GetBytes(body, "tools"); tools.Exists() && tools.IsArray() {
+		tools.ForEach(func(index, tool gjson.Result) bool {
+			name := tool.Get("name").String()
+			if name == "" || strings.HasPrefix(name, prefix) {
+				return true
+			}
+			path := fmt.Sprintf("tools.%d.name", index.Int())
+			body, _ = sjson.SetBytes(body, path, prefix+name)
+			return true
+		})
+	}
+
+	if gjson.GetBytes(body, "tool_choice.type").String() == "tool" {
+		name := gjson.GetBytes(body, "tool_choice.name").String()
+		if name != "" && !strings.HasPrefix(name, prefix) {
+			body, _ = sjson.SetBytes(body, "tool_choice.name", prefix+name)
+		}
+	}
+
+	if messages := gjson.GetBytes(body, "messages"); messages.Exists() && messages.IsArray() {
+		messages.ForEach(func(msgIndex, msg gjson.Result) bool {
+			content := msg.Get("content")
+			if !content.Exists() || !content.IsArray() {
+				return true
+			}
+			content.ForEach(func(contentIndex, part gjson.Result) bool {
+				if part.Get("type").String() != "tool_use" {
+					return true
+				}
+				name := part.Get("name").String()
+				if name == "" || strings.HasPrefix(name, prefix) {
+					return true
+				}
+				path := fmt.Sprintf("messages.%d.content.%d.name", msgIndex.Int(), contentIndex.Int())
+				body, _ = sjson.SetBytes(body, path, prefix+name)
+				return true
+			})
+			return true
+		})
+	}
+
+	return body
+}
+
+func stripClaudeToolPrefixFromResponse(body []byte, prefix string) []byte {
+	if prefix == "" {
+		return body
+	}
+	content := gjson.GetBytes(body, "content")
+	if !content.Exists() || !content.IsArray() {
+		return body
+	}
+	content.ForEach(func(index, part gjson.Result) bool {
+		if part.Get("type").String() != "tool_use" {
+			return true
+		}
+		name := part.Get("name").String()
+		if !strings.HasPrefix(name, prefix) {
+			return true
+		}
+		path := fmt.Sprintf("content.%d.name", index.Int())
+		body, _ = sjson.SetBytes(body, path, strings.TrimPrefix(name, prefix))
+		return true
+	})
+	return body
+}
+
+func stripClaudeToolPrefixFromStreamLine(line []byte, prefix string) []byte {
+	if prefix == "" {
+		return line
+	}
+	payload := jsonPayload(line)
+	if len(payload) == 0 || !gjson.ValidBytes(payload) {
+		return line
+	}
+	contentBlock := gjson.GetBytes(payload, "content_block")
+	if !contentBlock.Exists() || contentBlock.Get("type").String() != "tool_use" {
+		return line
+	}
+	name := contentBlock.Get("name").String()
+	if !strings.HasPrefix(name, prefix) {
+		return line
+	}
+	updated, err := sjson.SetBytes(payload, "content_block.name", strings.TrimPrefix(name, prefix))
+	if err != nil {
+		return line
+	}
+
+	trimmed := bytes.TrimSpace(line)
+	if bytes.HasPrefix(trimmed, []byte("data:")) {
+		return append([]byte("data: "), updated...)
+	}
+	return updated
+}

internal/runtime/executor/claude_executor_test.go

@@ -0,0 +1,51 @@
+package executor
+
+import (
+	"bytes"
+	"testing"
+
+	"github.com/tidwall/gjson"
+)
+
+func TestApplyClaudeToolPrefix(t *testing.T) {
+	input := []byte(`{"tools":[{"name":"alpha"},{"name":"proxy_bravo"}],"tool_choice":{"type":"tool","name":"charlie"},"messages":[{"role":"assistant","content":[{"type":"tool_use","name":"delta","id":"t1","input":{}}]}]}`)
+	out := applyClaudeToolPrefix(input, "proxy_")
+
+	if got := gjson.GetBytes(out, "tools.0.name").String(); got != "proxy_alpha" {
+		t.Fatalf("tools.0.name = %q, want %q", got, "proxy_alpha")
+	}
+	if got := gjson.GetBytes(out, "tools.1.name").String(); got != "proxy_bravo" {
+		t.Fatalf("tools.1.name = %q, want %q", got, "proxy_bravo")
+	}
+	if got := gjson.GetBytes(out, "tool_choice.name").String(); got != "proxy_charlie" {
+		t.Fatalf("tool_choice.name = %q, want %q", got, "proxy_charlie")
+	}
+	if got := gjson.GetBytes(out, "messages.0.content.0.name").String(); got != "proxy_delta" {
+		t.Fatalf("messages.0.content.0.name = %q, want %q", got, "proxy_delta")
+	}
+}
+
+func TestStripClaudeToolPrefixFromResponse(t *testing.T) {
+	input := []byte(`{"content":[{"type":"tool_use","name":"proxy_alpha","id":"t1","input":{}},{"type":"tool_use","name":"bravo","id":"t2","input":{}}]}`)
+	out := stripClaudeToolPrefixFromResponse(input, "proxy_")
+
+	if got := gjson.GetBytes(out, "content.0.name").String(); got != "alpha" {
+		t.Fatalf("content.0.name = %q, want %q", got, "alpha")
+	}
+	if got := gjson.GetBytes(out, "content.1.name").String(); got != "bravo" {
+		t.Fatalf("content.1.name = %q, want %q", got, "bravo")
+	}
+}
+
+func TestStripClaudeToolPrefixFromStreamLine(t *testing.T) {
+	line := []byte(`data: {"type":"content_block_start","content_block":{"type":"tool_use","name":"proxy_alpha","id":"t1"},"index":0}`)
+	out := stripClaudeToolPrefixFromStreamLine(line, "proxy_")
+
+	payload := bytes.TrimSpace(out)
+	if bytes.HasPrefix(payload, []byte("data:")) {
+		payload = bytes.TrimSpace(payload[len("data:"):])
+	}
+	if got := gjson.GetBytes(payload, "content_block.name").String(); got != "alpha" {
+		t.Fatalf("content_block.name = %q, want %q", got, "alpha")
+	}
+}

internal/runtime/executor/codex_executor.go

@@ -38,7 +38,38 @@ func NewCodexExecutor(cfg *config.Config) *CodexExecutor { return &CodexExecutor
 
 func (e *CodexExecutor) Identifier() string { return "codex" }
 
-func (e *CodexExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error { return nil }
+// PrepareRequest injects Codex credentials into the outgoing HTTP request.
+func (e *CodexExecutor) PrepareRequest(req *http.Request, auth *cliproxyauth.Auth) error {
+	if req == nil {
+		return nil
+	}
+	apiKey, _ := codexCreds(auth)
+	if strings.TrimSpace(apiKey) != "" {
+		req.Header.Set("Authorization", "Bearer "+apiKey)
+	}
+	var attrs map[string]string
+	if auth != nil {
+		attrs = auth.Attributes
+	}
+	util.ApplyCustomHeadersFromAttrs(req, attrs)
+	return nil
+}
+
+// HttpRequest injects Codex credentials into the request and executes it.
+func (e *CodexExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth, req *http.Request) (*http.Response, error) {
+	if req == nil {
+		return nil, fmt.Errorf("codex executor: request is nil")
+	}
+	if ctx == nil {
+		ctx = req.Context()
+	}
+	httpReq := req.WithContext(ctx)
+	if err := e.PrepareRequest(httpReq, auth); err != nil {
+		return nil, err
+	}
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	return httpClient.Do(httpReq)
+}
 
 func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
 	apiKey, baseURL := codexCreds(auth)
@@ -457,9 +488,9 @@ func (e *CodexExecutor) cacheHelper(ctx context.Context, from sdktranslator.Form
 	if from == "claude" {
 		userIDResult := gjson.GetBytes(req.Payload, "metadata.user_id")
 		if userIDResult.Exists() {
-			var hasKey bool
 			key := fmt.Sprintf("%s-%s", req.Model, userIDResult.String())
-			if cache, hasKey = getCodexCache(key); !hasKey || cache.Expire.Before(time.Now()) {
+			var ok bool
+			if cache, ok = getCodexCache(key); !ok {
 				cache = codexCache{
 					ID:     uuid.New().String(),
 					Expire: time.Now().Add(1 * time.Hour),

internal/runtime/executor/gemini_cli_executor.go

@@ -63,8 +63,42 @@ func NewGeminiCLIExecutor(cfg *config.Config) *GeminiCLIExecutor {
 // Identifier returns the executor identifier.
 func (e *GeminiCLIExecutor) Identifier() string { return "gemini-cli" }
 
-// PrepareRequest prepares the HTTP request for execution (no-op for Gemini CLI).
-func (e *GeminiCLIExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error { return nil }
+// PrepareRequest injects Gemini CLI credentials into the outgoing HTTP request.
+func (e *GeminiCLIExecutor) PrepareRequest(req *http.Request, auth *cliproxyauth.Auth) error {
+	if req == nil {
+		return nil
+	}
+	tokenSource, _, errSource := prepareGeminiCLITokenSource(req.Context(), e.cfg, auth)
+	if errSource != nil {
+		return errSource
+	}
+	tok, errTok := tokenSource.Token()
+	if errTok != nil {
+		return errTok
+	}
+	if strings.TrimSpace(tok.AccessToken) == "" {
+		return statusErr{code: http.StatusUnauthorized, msg: "missing access token"}
+	}
+	req.Header.Set("Authorization", "Bearer "+tok.AccessToken)
+	applyGeminiCLIHeaders(req)
+	return nil
+}
+
+// HttpRequest injects Gemini CLI credentials into the request and executes it.
+func (e *GeminiCLIExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth, req *http.Request) (*http.Response, error) {
+	if req == nil {
+		return nil, fmt.Errorf("gemini-cli executor: request is nil")
+	}
+	if ctx == nil {
+		ctx = req.Context()
+	}
+	httpReq := req.WithContext(ctx)
+	if err := e.PrepareRequest(httpReq, auth); err != nil {
+		return nil, err
+	}
+	httpClient := newHTTPClient(ctx, e.cfg, auth, 0)
+	return httpClient.Do(httpReq)
+}
 
 // Execute performs a non-streaming request to the Gemini CLI API.
 func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {

internal/runtime/executor/gemini_executor.go

@@ -55,8 +55,38 @@ func NewGeminiExecutor(cfg *config.Config) *GeminiExecutor {
 // Identifier returns the executor identifier.
 func (e *GeminiExecutor) Identifier() string { return "gemini" }
 
-// PrepareRequest prepares the HTTP request for execution (no-op for Gemini).
-func (e *GeminiExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error { return nil }
+// PrepareRequest injects Gemini credentials into the outgoing HTTP request.
+func (e *GeminiExecutor) PrepareRequest(req *http.Request, auth *cliproxyauth.Auth) error {
+	if req == nil {
+		return nil
+	}
+	apiKey, bearer := geminiCreds(auth)
+	if apiKey != "" {
+		req.Header.Set("x-goog-api-key", apiKey)
+		req.Header.Del("Authorization")
+	} else if bearer != "" {
+		req.Header.Set("Authorization", "Bearer "+bearer)
+		req.Header.Del("x-goog-api-key")
+	}
+	applyGeminiHeaders(req, auth)
+	return nil
+}
+
+// HttpRequest injects Gemini credentials into the request and executes it.
+func (e *GeminiExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth, req *http.Request) (*http.Response, error) {
+	if req == nil {
+		return nil, fmt.Errorf("gemini executor: request is nil")
+	}
+	if ctx == nil {
+		ctx = req.Context()
+	}
+	httpReq := req.WithContext(ctx)
+	if err := e.PrepareRequest(httpReq, auth); err != nil {
+		return nil, err
+	}
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	return httpClient.Do(httpReq)
+}
 
 // Execute performs a non-streaming request to the Gemini API.
 // It translates the request to Gemini format, sends it to the API, and translates

internal/runtime/executor/gemini_vertex_executor.go

@@ -50,11 +50,49 @@ func NewGeminiVertexExecutor(cfg *config.Config) *GeminiVertexExecutor {
 // Identifier returns the executor identifier.
 func (e *GeminiVertexExecutor) Identifier() string { return "vertex" }
 
-// PrepareRequest prepares the HTTP request for execution (no-op for Vertex).
-func (e *GeminiVertexExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error {
+// PrepareRequest injects Vertex credentials into the outgoing HTTP request.
+func (e *GeminiVertexExecutor) PrepareRequest(req *http.Request, auth *cliproxyauth.Auth) error {
+	if req == nil {
+		return nil
+	}
+	apiKey, _ := vertexAPICreds(auth)
+	if strings.TrimSpace(apiKey) != "" {
+		req.Header.Set("x-goog-api-key", apiKey)
+		req.Header.Del("Authorization")
+		return nil
+	}
+	_, _, saJSON, errCreds := vertexCreds(auth)
+	if errCreds != nil {
+		return errCreds
+	}
+	token, errToken := vertexAccessToken(req.Context(), e.cfg, auth, saJSON)
+	if errToken != nil {
+		return errToken
+	}
+	if strings.TrimSpace(token) == "" {
+		return statusErr{code: http.StatusUnauthorized, msg: "missing access token"}
+	}
+	req.Header.Set("Authorization", "Bearer "+token)
+	req.Header.Del("x-goog-api-key")
 	return nil
 }
 
+// HttpRequest injects Vertex credentials into the request and executes it.
+func (e *GeminiVertexExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth, req *http.Request) (*http.Response, error) {
+	if req == nil {
+		return nil, fmt.Errorf("vertex executor: request is nil")
+	}
+	if ctx == nil {
+		ctx = req.Context()
+	}
+	httpReq := req.WithContext(ctx)
+	if err := e.PrepareRequest(httpReq, auth); err != nil {
+		return nil, err
+	}
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	return httpClient.Do(httpReq)
+}
+
 // Execute performs a non-streaming request to the Vertex AI API.
 func (e *GeminiVertexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
 	// Try API key authentication first

internal/runtime/executor/github_copilot_executor.go

@@ -63,10 +63,38 @@ func NewGitHubCopilotExecutor(cfg *config.Config) *GitHubCopilotExecutor {
 func (e *GitHubCopilotExecutor) Identifier() string { return githubCopilotAuthType }
 
 // PrepareRequest implements ProviderExecutor.
-func (e *GitHubCopilotExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error {
+func (e *GitHubCopilotExecutor) PrepareRequest(req *http.Request, auth *cliproxyauth.Auth) error {
+	if req == nil {
+		return nil
+	}
+	ctx := req.Context()
+	if ctx == nil {
+		ctx = context.Background()
+	}
+	apiToken, errToken := e.ensureAPIToken(ctx, auth)
+	if errToken != nil {
+		return errToken
+	}
+	e.applyHeaders(req, apiToken)
 	return nil
 }
 
+// HttpRequest injects GitHub Copilot credentials into the request and executes it.
+func (e *GitHubCopilotExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth, req *http.Request) (*http.Response, error) {
+	if req == nil {
+		return nil, fmt.Errorf("github-copilot executor: request is nil")
+	}
+	if ctx == nil {
+		ctx = req.Context()
+	}
+	httpReq := req.WithContext(ctx)
+	if errPrepare := e.PrepareRequest(httpReq, auth); errPrepare != nil {
+		return nil, errPrepare
+	}
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	return httpClient.Do(httpReq)
+}
+
 // Execute handles non-streaming requests to GitHub Copilot.
 func (e *GitHubCopilotExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
 	apiToken, errToken := e.ensureAPIToken(ctx, auth)

internal/runtime/executor/iflow_executor.go

@@ -37,8 +37,33 @@ func NewIFlowExecutor(cfg *config.Config) *IFlowExecutor { return &IFlowExecutor
 // Identifier returns the provider key.
 func (e *IFlowExecutor) Identifier() string { return "iflow" }
 
-// PrepareRequest implements ProviderExecutor but requires no preprocessing.
-func (e *IFlowExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error { return nil }
+// PrepareRequest injects iFlow credentials into the outgoing HTTP request.
+func (e *IFlowExecutor) PrepareRequest(req *http.Request, auth *cliproxyauth.Auth) error {
+	if req == nil {
+		return nil
+	}
+	apiKey, _ := iflowCreds(auth)
+	if strings.TrimSpace(apiKey) != "" {
+		req.Header.Set("Authorization", "Bearer "+apiKey)
+	}
+	return nil
+}
+
+// HttpRequest injects iFlow credentials into the request and executes it.
+func (e *IFlowExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth, req *http.Request) (*http.Response, error) {
+	if req == nil {
+		return nil, fmt.Errorf("iflow executor: request is nil")
+	}
+	if ctx == nil {
+		ctx = req.Context()
+	}
+	httpReq := req.WithContext(ctx)
+	if err := e.PrepareRequest(httpReq, auth); err != nil {
+		return nil, err
+	}
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	return httpClient.Do(httpReq)
+}
 
 // Execute performs a non-streaming chat completion request.
 func (e *IFlowExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {

internal/runtime/executor/kiro_executor.go

@@ -28,7 +28,6 @@ import (
 	"github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/usage"
 	sdktranslator "github.com/router-for-me/CLIProxyAPI/v6/sdk/translator"
 	log "github.com/sirupsen/logrus"
-
 )
 
 const (
@@ -218,7 +217,48 @@ func NewKiroExecutor(cfg *config.Config) *KiroExecutor {
 func (e *KiroExecutor) Identifier() string { return "kiro" }
 
 // PrepareRequest prepares the HTTP request before execution.
-func (e *KiroExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error { return nil }
+func (e *KiroExecutor) PrepareRequest(req *http.Request, auth *cliproxyauth.Auth) error {
+	if req == nil {
+		return nil
+	}
+	accessToken, _ := kiroCredentials(auth)
+	if strings.TrimSpace(accessToken) == "" {
+		return statusErr{code: http.StatusUnauthorized, msg: "missing access token"}
+	}
+	if isIDCAuth(auth) {
+		req.Header.Set("User-Agent", kiroIDEUserAgent)
+		req.Header.Set("X-Amz-User-Agent", kiroIDEAmzUserAgent)
+		req.Header.Set("x-amzn-kiro-agent-mode", kiroIDEAgentModeSpec)
+	} else {
+		req.Header.Set("User-Agent", kiroUserAgent)
+		req.Header.Set("X-Amz-User-Agent", kiroFullUserAgent)
+	}
+	req.Header.Set("Amz-Sdk-Request", "attempt=1; max=3")
+	req.Header.Set("Amz-Sdk-Invocation-Id", uuid.New().String())
+	req.Header.Set("Authorization", "Bearer "+accessToken)
+	var attrs map[string]string
+	if auth != nil {
+		attrs = auth.Attributes
+	}
+	util.ApplyCustomHeadersFromAttrs(req, attrs)
+	return nil
+}
+
+// HttpRequest injects Kiro credentials into the request and executes it.
+func (e *KiroExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth, req *http.Request) (*http.Response, error) {
+	if req == nil {
+		return nil, fmt.Errorf("kiro executor: request is nil")
+	}
+	if ctx == nil {
+		ctx = req.Context()
+	}
+	httpReq := req.WithContext(ctx)
+	if errPrepare := e.PrepareRequest(httpReq, auth); errPrepare != nil {
+		return nil, errPrepare
+	}
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	return httpClient.Do(httpReq)
+}
 
 // Execute sends the request to Kiro API and returns the response.
 // Supports automatic token refresh on 401/403 errors.
@@ -1004,7 +1044,7 @@ func findRealThinkingEndTag(content string, alreadyInCodeBlock, alreadyInInlineC
 		discussionPatterns := []string{
 			"标签", "返回", "输出", "包含", "使用", "解析", "转换", "生成", // Chinese
 			"tag", "return", "output", "contain", "use", "parse", "emit", "convert", "generate", // English
-			"<thinking>", // discussing both tags together
+			"<thinking>",    // discussing both tags together
 			"`</thinking>`", // explicitly in inline code
 		}
 		isDiscussion := false
@@ -1852,7 +1892,6 @@ func (e *KiroExecutor) extractEventTypeFromBytes(headers []byte) string {
 	return ""
 }
 
-
 // NOTE: Response building functions moved to internal/translator/kiro/claude/kiro_claude_response.go
 // The executor now uses kiroclaude.BuildClaudeResponse() and kiroclaude.ExtractThinkingFromContent() instead
 
@@ -1889,18 +1928,18 @@ func (e *KiroExecutor) streamToChannel(ctx context.Context, body io.Reader, out
 	var lastReportedOutputTokens int64   // Last reported output token count
 
 	// Upstream usage tracking - Kiro API returns credit usage and context percentage
-	var upstreamCreditUsage float64        // Credit usage from upstream (e.g., 1.458)
-	var upstreamContextPercentage float64  // Context usage percentage from upstream (e.g., 78.56)
-	var hasUpstreamUsage bool              // Whether we received usage from upstream
+	var upstreamCreditUsage float64       // Credit usage from upstream (e.g., 1.458)
+	var upstreamContextPercentage float64 // Context usage percentage from upstream (e.g., 78.56)
+	var hasUpstreamUsage bool             // Whether we received usage from upstream
 
 	// Translator param for maintaining tool call state across streaming events
 	// IMPORTANT: This must persist across all TranslateStream calls
 	var translatorParam any
 
 	// Thinking mode state tracking - tag-based parsing for <thinking> tags in content
-	inThinkBlock := false       // Whether we're currently inside a <thinking> block
-	isThinkingBlockOpen := false // Track if thinking content block SSE event is open
-	thinkingBlockIndex := -1     // Index of the thinking content block
+	inThinkBlock := false                          // Whether we're currently inside a <thinking> block
+	isThinkingBlockOpen := false                   // Track if thinking content block SSE event is open
+	thinkingBlockIndex := -1                       // Index of the thinking content block
 	var accumulatedThinkingContent strings.Builder // Accumulate thinking content for token counting
 
 	// Buffer for handling partial tag matches at chunk boundaries
@@ -2319,16 +2358,16 @@ func (e *KiroExecutor) streamToChannel(ctx context.Context, body io.Reader, out
 
 					lastUsageUpdateLen = accumulatedContent.Len()
 					lastUsageUpdateTime = time.Now()
-					}
+				}
 
-					// TAG-BASED THINKING PARSING: Parse <thinking> tags from content
-					// Combine pending content with new content for processing
-					pendingContent.WriteString(contentDelta)
-					processContent := pendingContent.String()
-					pendingContent.Reset()
+				// TAG-BASED THINKING PARSING: Parse <thinking> tags from content
+				// Combine pending content with new content for processing
+				pendingContent.WriteString(contentDelta)
+				processContent := pendingContent.String()
+				pendingContent.Reset()
 
-					// Process content looking for thinking tags
-					for len(processContent) > 0 {
+				// Process content looking for thinking tags
+				for len(processContent) > 0 {
 					if inThinkBlock {
 						// We're inside a thinking block, look for </thinking>
 						endIdx := strings.Index(processContent, kirocommon.ThinkingEndTag)
@@ -2503,7 +2542,7 @@ func (e *KiroExecutor) streamToChannel(ctx context.Context, body io.Reader, out
 							processContent = ""
 						}
 					}
-					}
+				}
 			}
 
 			// Handle tool uses in response (with deduplication)
@@ -2927,7 +2966,7 @@ func (e *KiroExecutor) streamToChannel(ctx context.Context, body io.Reader, out
 		// Calculate input tokens from context percentage
 		// Using 200k as the base since that's what Kiro reports against
 		calculatedInputTokens := int64(upstreamContextPercentage * 200000 / 100)
-		
+
 		// Only use calculated value if it's significantly different from local estimate
 		// This provides more accurate token counts based on upstream data
 		if calculatedInputTokens > 0 {

internal/runtime/executor/logging_helpers.go

@@ -304,11 +304,7 @@ func formatAuthInfo(info upstreamRequestLog) string {
 			parts = append(parts, "type=api_key")
 		}
 	case "oauth":
-		if authValue != "" {
-			parts = append(parts, fmt.Sprintf("type=oauth account=%s", authValue))
-		} else {
-			parts = append(parts, "type=oauth")
-		}
+		parts = append(parts, "type=oauth")
 	default:
 		if authType != "" {
 			if authValue != "" {

internal/runtime/executor/openai_compat_executor.go

@@ -35,11 +35,39 @@ func NewOpenAICompatExecutor(provider string, cfg *config.Config) *OpenAICompatE
 // Identifier implements cliproxyauth.ProviderExecutor.
 func (e *OpenAICompatExecutor) Identifier() string { return e.provider }
 
-// PrepareRequest is a no-op for now (credentials are added via headers at execution time).
-func (e *OpenAICompatExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error {
+// PrepareRequest injects OpenAI-compatible credentials into the outgoing HTTP request.
+func (e *OpenAICompatExecutor) PrepareRequest(req *http.Request, auth *cliproxyauth.Auth) error {
+	if req == nil {
+		return nil
+	}
+	_, apiKey := e.resolveCredentials(auth)
+	if strings.TrimSpace(apiKey) != "" {
+		req.Header.Set("Authorization", "Bearer "+apiKey)
+	}
+	var attrs map[string]string
+	if auth != nil {
+		attrs = auth.Attributes
+	}
+	util.ApplyCustomHeadersFromAttrs(req, attrs)
 	return nil
 }
 
+// HttpRequest injects OpenAI-compatible credentials into the request and executes it.
+func (e *OpenAICompatExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth, req *http.Request) (*http.Response, error) {
+	if req == nil {
+		return nil, fmt.Errorf("openai compat executor: request is nil")
+	}
+	if ctx == nil {
+		ctx = req.Context()
+	}
+	httpReq := req.WithContext(ctx)
+	if err := e.PrepareRequest(httpReq, auth); err != nil {
+		return nil, err
+	}
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	return httpClient.Do(httpReq)
+}
+
 func (e *OpenAICompatExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
@@ -241,6 +269,11 @@ func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
 			if len(line) == 0 {
 				continue
 			}
+
+			if !bytes.HasPrefix(line, []byte("data:")) {
+				continue
+			}
+
 			// OpenAI-compatible streams are SSE: lines typically prefixed with "data: ".
 			// Pass through translator; it yields one or more chunks for the target schema.
 			chunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, bytes.Clone(line), &param)

internal/runtime/executor/qwen_executor.go

@@ -36,7 +36,33 @@ func NewQwenExecutor(cfg *config.Config) *QwenExecutor { return &QwenExecutor{cf
 
 func (e *QwenExecutor) Identifier() string { return "qwen" }
 
-func (e *QwenExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error { return nil }
+// PrepareRequest injects Qwen credentials into the outgoing HTTP request.
+func (e *QwenExecutor) PrepareRequest(req *http.Request, auth *cliproxyauth.Auth) error {
+	if req == nil {
+		return nil
+	}
+	token, _ := qwenCreds(auth)
+	if strings.TrimSpace(token) != "" {
+		req.Header.Set("Authorization", "Bearer "+token)
+	}
+	return nil
+}
+
+// HttpRequest injects Qwen credentials into the request and executes it.
+func (e *QwenExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth, req *http.Request) (*http.Response, error) {
+	if req == nil {
+		return nil, fmt.Errorf("qwen executor: request is nil")
+	}
+	if ctx == nil {
+		ctx = req.Context()
+	}
+	httpReq := req.WithContext(ctx)
+	if err := e.PrepareRequest(httpReq, auth); err != nil {
+		return nil, err
+	}
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	return httpClient.Do(httpReq)
+}
 
 func (e *QwenExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
 	token, baseURL := qwenCreds(auth)

internal/translator/antigravity/claude/antigravity_claude_request.go

@@ -14,7 +14,6 @@ import (
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/cache"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/common"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
-	log "github.com/sirupsen/logrus"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
 )
@@ -136,14 +135,14 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 						if sessionID != "" && thinkingText != "" {
 							if cachedSig := cache.GetCachedSignature(sessionID, thinkingText); cachedSig != "" {
 								signature = cachedSig
-								log.Debugf("Using cached signature for thinking block")
+								// log.Debugf("Using cached signature for thinking block")
 							}
 						}
 
 						// Fallback to client signature only if cache miss and client signature is valid
 						if signature == "" && cache.HasValidSignature(clientSignature) {
 							signature = clientSignature
-							log.Debugf("Using client-provided signature for thinking block")
+							// log.Debugf("Using client-provided signature for thinking block")
 						}
 
 						// Store for subsequent tool_use in the same message
@@ -158,8 +157,7 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 						// Claude requires assistant messages to start with thinking blocks when thinking is enabled
 						// Converting to text would break this requirement
 						if isUnsigned {
-							// TypeScript plugin approach: drop unsigned thinking blocks entirely
-							log.Debugf("Dropping unsigned thinking block (no valid signature)")
+							// log.Debugf("Dropping unsigned thinking block (no valid signature)")
 							continue
 						}
 
@@ -183,7 +181,6 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 					} else if contentTypeResult.Type == gjson.String && contentTypeResult.String() == "tool_use" {
 						// NOTE: Do NOT inject dummy thinking blocks here.
 						// Antigravity API validates signatures, so dummy values are rejected.
-						// The TypeScript plugin removes unsigned thinking blocks instead of injecting dummies.
 
 						functionName := contentResult.Get("name").String()
 						argsResult := contentResult.Get("input")

internal/translator/antigravity/claude/antigravity_claude_response.go

@@ -136,11 +136,11 @@ func ConvertAntigravityResponseToClaude(_ context.Context, _ string, originalReq
 				// Process thinking content (internal reasoning)
 				if partResult.Get("thought").Bool() {
 					if thoughtSignature := partResult.Get("thoughtSignature"); thoughtSignature.Exists() && thoughtSignature.String() != "" {
-						log.Debug("Branch: signature_delta")
+						// log.Debug("Branch: signature_delta")
 
 						if params.SessionID != "" && params.CurrentThinkingText.Len() > 0 {
 							cache.CacheSignature(params.SessionID, params.CurrentThinkingText.String(), thoughtSignature.String())
-							log.Debugf("Cached signature for thinking block (sessionID=%s, textLen=%d)", params.SessionID, params.CurrentThinkingText.Len())
+							// log.Debugf("Cached signature for thinking block (sessionID=%s, textLen=%d)", params.SessionID, params.CurrentThinkingText.Len())
 							params.CurrentThinkingText.Reset()
 						}
 

internal/translator/claude/openai/responses/claude_openai-responses_response.go

@@ -40,6 +40,16 @@ type claudeToResponsesState struct {
 
 var dataTag = []byte("data:")
 
+func pickRequestJSON(originalRequestRawJSON, requestRawJSON []byte) []byte {
+	if len(originalRequestRawJSON) > 0 && gjson.ValidBytes(originalRequestRawJSON) {
+		return originalRequestRawJSON
+	}
+	if len(requestRawJSON) > 0 && gjson.ValidBytes(requestRawJSON) {
+		return requestRawJSON
+	}
+	return nil
+}
+
 func emitEvent(event string, payload string) string {
 	return fmt.Sprintf("event: %s\ndata: %s", event, payload)
 }
@@ -279,8 +289,9 @@ func ConvertClaudeResponseToOpenAIResponses(ctx context.Context, modelName strin
 		completed, _ = sjson.Set(completed, "response.created_at", st.CreatedAt)
 		// Inject original request fields into response as per docs/response.completed.json
 
-		if requestRawJSON != nil {
-			req := gjson.ParseBytes(requestRawJSON)
+		reqBytes := pickRequestJSON(originalRequestRawJSON, requestRawJSON)
+		if len(reqBytes) > 0 {
+			req := gjson.ParseBytes(reqBytes)
 			if v := req.Get("instructions"); v.Exists() {
 				completed, _ = sjson.Set(completed, "response.instructions", v.String())
 			}
@@ -549,8 +560,9 @@ func ConvertClaudeResponseToOpenAIResponsesNonStream(_ context.Context, _ string
 	out, _ = sjson.Set(out, "created_at", createdAt)
 
 	// Inject request echo fields as top-level (similar to streaming variant)
-	if requestRawJSON != nil {
-		req := gjson.ParseBytes(requestRawJSON)
+	reqBytes := pickRequestJSON(originalRequestRawJSON, requestRawJSON)
+	if len(reqBytes) > 0 {
+		req := gjson.ParseBytes(reqBytes)
 		if v := req.Get("instructions"); v.Exists() {
 			out, _ = sjson.Set(out, "instructions", v.String())
 		}

internal/translator/codex/claude/codex_claude_response.go

@@ -20,6 +20,12 @@ var (
 	dataTag = []byte("data:")
 )
 
+// ConvertCodexResponseToClaudeParams holds parameters for response conversion.
+type ConvertCodexResponseToClaudeParams struct {
+	HasToolCall bool
+	BlockIndex  int
+}
+
 // ConvertCodexResponseToClaude performs sophisticated streaming response format conversion.
 // This function implements a complex state machine that translates Codex API responses
 // into Claude Code-compatible Server-Sent Events (SSE) format. It manages different response types
@@ -38,8 +44,10 @@ var (
 //   - []string: A slice of strings, each containing a Claude Code-compatible JSON response
 func ConvertCodexResponseToClaude(_ context.Context, _ string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, param *any) []string {
 	if *param == nil {
-		hasToolCall := false
-		*param = &hasToolCall
+		*param = &ConvertCodexResponseToClaudeParams{
+			HasToolCall: false,
+			BlockIndex:  0,
+		}
 	}
 
 	// log.Debugf("rawJSON: %s", string(rawJSON))
@@ -62,46 +70,49 @@ func ConvertCodexResponseToClaude(_ context.Context, _ string, originalRequestRa
 		output += fmt.Sprintf("data: %s\n\n", template)
 	} else if typeStr == "response.reasoning_summary_part.added" {
 		template = `{"type":"content_block_start","index":0,"content_block":{"type":"thinking","thinking":""}}`
-		template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
+		template, _ = sjson.Set(template, "index", (*param).(*ConvertCodexResponseToClaudeParams).BlockIndex)
 
 		output = "event: content_block_start\n"
 		output += fmt.Sprintf("data: %s\n\n", template)
 	} else if typeStr == "response.reasoning_summary_text.delta" {
 		template = `{"type":"content_block_delta","index":0,"delta":{"type":"thinking_delta","thinking":""}}`
-		template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
+		template, _ = sjson.Set(template, "index", (*param).(*ConvertCodexResponseToClaudeParams).BlockIndex)
 		template, _ = sjson.Set(template, "delta.thinking", rootResult.Get("delta").String())
 
 		output = "event: content_block_delta\n"
 		output += fmt.Sprintf("data: %s\n\n", template)
 	} else if typeStr == "response.reasoning_summary_part.done" {
 		template = `{"type":"content_block_stop","index":0}`
-		template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
+		template, _ = sjson.Set(template, "index", (*param).(*ConvertCodexResponseToClaudeParams).BlockIndex)
+		(*param).(*ConvertCodexResponseToClaudeParams).BlockIndex++
 
 		output = "event: content_block_stop\n"
 		output += fmt.Sprintf("data: %s\n\n", template)
+
 	} else if typeStr == "response.content_part.added" {
 		template = `{"type":"content_block_start","index":0,"content_block":{"type":"text","text":""}}`
-		template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
+		template, _ = sjson.Set(template, "index", (*param).(*ConvertCodexResponseToClaudeParams).BlockIndex)
 
 		output = "event: content_block_start\n"
 		output += fmt.Sprintf("data: %s\n\n", template)
 	} else if typeStr == "response.output_text.delta" {
 		template = `{"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":""}}`
-		template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
+		template, _ = sjson.Set(template, "index", (*param).(*ConvertCodexResponseToClaudeParams).BlockIndex)
 		template, _ = sjson.Set(template, "delta.text", rootResult.Get("delta").String())
 
 		output = "event: content_block_delta\n"
 		output += fmt.Sprintf("data: %s\n\n", template)
 	} else if typeStr == "response.content_part.done" {
 		template = `{"type":"content_block_stop","index":0}`
-		template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
+		template, _ = sjson.Set(template, "index", (*param).(*ConvertCodexResponseToClaudeParams).BlockIndex)
+		(*param).(*ConvertCodexResponseToClaudeParams).BlockIndex++
 
 		output = "event: content_block_stop\n"
 		output += fmt.Sprintf("data: %s\n\n", template)
 	} else if typeStr == "response.completed" {
 		template = `{"type":"message_delta","delta":{"stop_reason":"tool_use","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
-		p := (*param).(*bool)
-		if *p {
+		p := (*param).(*ConvertCodexResponseToClaudeParams).HasToolCall
+		if p {
 			template, _ = sjson.Set(template, "delta.stop_reason", "tool_use")
 		} else {
 			template, _ = sjson.Set(template, "delta.stop_reason", "end_turn")
@@ -118,10 +129,9 @@ func ConvertCodexResponseToClaude(_ context.Context, _ string, originalRequestRa
 		itemResult := rootResult.Get("item")
 		itemType := itemResult.Get("type").String()
 		if itemType == "function_call" {
-			p := true
-			*param = &p
+			(*param).(*ConvertCodexResponseToClaudeParams).HasToolCall = true
 			template = `{"type":"content_block_start","index":0,"content_block":{"type":"tool_use","id":"","name":"","input":{}}}`
-			template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
+			template, _ = sjson.Set(template, "index", (*param).(*ConvertCodexResponseToClaudeParams).BlockIndex)
 			template, _ = sjson.Set(template, "content_block.id", itemResult.Get("call_id").String())
 			{
 				// Restore original tool name if shortened
@@ -137,7 +147,7 @@ func ConvertCodexResponseToClaude(_ context.Context, _ string, originalRequestRa
 			output += fmt.Sprintf("data: %s\n\n", template)
 
 			template = `{"type":"content_block_delta","index":0,"delta":{"type":"input_json_delta","partial_json":""}}`
-			template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
+			template, _ = sjson.Set(template, "index", (*param).(*ConvertCodexResponseToClaudeParams).BlockIndex)
 
 			output += "event: content_block_delta\n"
 			output += fmt.Sprintf("data: %s\n\n", template)
@@ -147,14 +157,15 @@ func ConvertCodexResponseToClaude(_ context.Context, _ string, originalRequestRa
 		itemType := itemResult.Get("type").String()
 		if itemType == "function_call" {
 			template = `{"type":"content_block_stop","index":0}`
-			template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
+			template, _ = sjson.Set(template, "index", (*param).(*ConvertCodexResponseToClaudeParams).BlockIndex)
+			(*param).(*ConvertCodexResponseToClaudeParams).BlockIndex++
 
 			output = "event: content_block_stop\n"
 			output += fmt.Sprintf("data: %s\n\n", template)
 		}
 	} else if typeStr == "response.function_call_arguments.delta" {
 		template = `{"type":"content_block_delta","index":0,"delta":{"type":"input_json_delta","partial_json":""}}`
-		template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
+		template, _ = sjson.Set(template, "index", (*param).(*ConvertCodexResponseToClaudeParams).BlockIndex)
 		template, _ = sjson.Set(template, "delta.partial_json", rootResult.Get("delta").String())
 
 		output += "event: content_block_delta\n"

internal/translator/codex/openai/chat-completions/codex_openai_request.go

@@ -275,7 +275,15 @@ func ConvertOpenAIRequestToCodex(modelName string, inputRawJSON []byte, stream b
 		arr := tools.Array()
 		for i := 0; i < len(arr); i++ {
 			t := arr[i]
-			if t.Get("type").String() == "function" {
+			toolType := t.Get("type").String()
+			// Pass through built-in tools (e.g. {"type":"web_search"}) directly for the Responses API.
+			// Only "function" needs structural conversion because Chat Completions nests details under "function".
+			if toolType != "" && toolType != "function" && t.IsObject() {
+				out, _ = sjson.SetRaw(out, "tools.-1", t.Raw)
+				continue
+			}
+
+			if toolType == "function" {
 				item := `{}`
 				item, _ = sjson.Set(item, "type", "function")
 				fn := t.Get("function")
@@ -304,6 +312,37 @@ func ConvertOpenAIRequestToCodex(modelName string, inputRawJSON []byte, stream b
 		}
 	}
 
+	// Map tool_choice when present.
+	// Chat Completions: "tool_choice" can be a string ("auto"/"none") or an object (e.g. {"type":"function","function":{"name":"..."}}).
+	// Responses API: keep built-in tool choices as-is; flatten function choice to {"type":"function","name":"..."}.
+	if tc := gjson.GetBytes(rawJSON, "tool_choice"); tc.Exists() {
+		switch {
+		case tc.Type == gjson.String:
+			out, _ = sjson.Set(out, "tool_choice", tc.String())
+		case tc.IsObject():
+			tcType := tc.Get("type").String()
+			if tcType == "function" {
+				name := tc.Get("function.name").String()
+				if name != "" {
+					if short, ok := originalToolNameMap[name]; ok {
+						name = short
+					} else {
+						name = shortenNameIfNeeded(name)
+					}
+				}
+				choice := `{}`
+				choice, _ = sjson.Set(choice, "type", "function")
+				if name != "" {
+					choice, _ = sjson.Set(choice, "name", name)
+				}
+				out, _ = sjson.SetRaw(out, "tool_choice", choice)
+			} else if tcType != "" {
+				// Built-in tool choices (e.g. {"type":"web_search"}) are already Responses-compatible.
+				out, _ = sjson.SetRaw(out, "tool_choice", tc.Raw)
+			}
+		}
+	}
+
 	out, _ = sjson.Set(out, "store", false)
 	return []byte(out)
 }

internal/translator/openai/openai/responses/openai_openai-responses_request.go

@@ -163,6 +163,14 @@ func ConvertOpenAIResponsesRequestToOpenAIChatCompletions(modelName string, inpu
 		var chatCompletionsTools []interface{}
 
 		tools.ForEach(func(_, tool gjson.Result) bool {
+			// Built-in tools (e.g. {"type":"web_search"}) are already compatible with the Chat Completions schema.
+			// Only function tools need structural conversion because Chat Completions nests details under "function".
+			toolType := tool.Get("type").String()
+			if toolType != "" && toolType != "function" && tool.IsObject() {
+				chatCompletionsTools = append(chatCompletionsTools, tool.Value())
+				return true
+			}
+
 			chatTool := `{"type":"function","function":{}}`
 
 			// Convert tool structure from responses format to chat completions format

sdk/api/management.go

@@ -21,6 +21,7 @@ type ManagementTokenRequester interface {
 	RequestIFlowToken(*gin.Context)
 	RequestIFlowCookieToken(*gin.Context)
 	GetAuthStatus(c *gin.Context)
+	PostOAuthCallback(c *gin.Context)
 }
 
 type managementTokenRequester struct {
@@ -65,3 +66,7 @@ func (m *managementTokenRequester) RequestIFlowCookieToken(c *gin.Context) {
 func (m *managementTokenRequester) GetAuthStatus(c *gin.Context) {
 	m.handler.GetAuthStatus(c)
 }
+
+func (m *managementTokenRequester) PostOAuthCallback(c *gin.Context) {
+	m.handler.PostOAuthCallback(c)
+}

sdk/cliproxy/auth/conductor.go

@@ -1,10 +1,13 @@
 package auth
 
 import (
+	"bytes"
 	"context"
 	"encoding/json"
 	"errors"
+	"io"
 	"net/http"
+	"path/filepath"
 	"strconv"
 	"strings"
 	"sync"
@@ -31,6 +34,9 @@ type ProviderExecutor interface {
 	Refresh(ctx context.Context, auth *Auth) (*Auth, error)
 	// CountTokens returns the token count for the given request.
 	CountTokens(ctx context.Context, auth *Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error)
+	// HttpRequest injects provider credentials into the supplied HTTP request and executes it.
+	// Callers must close the response body when non-nil.
+	HttpRequest(ctx context.Context, auth *Auth, req *http.Request) (*http.Response, error)
 }
 
 // RefreshEvaluator allows runtime state to override refresh decisions.
@@ -388,22 +394,8 @@ func (m *Manager) executeWithProvider(ctx context.Context, provider string, req
 			return cliproxyexecutor.Response{}, errPick
 		}
 
-		accountType, accountInfo := auth.AccountInfo()
-		proxyInfo := auth.ProxyInfo()
 		entry := logEntryWithRequestID(ctx)
-		if accountType == "api_key" {
-			if proxyInfo != "" {
-				entry.Debugf("Use API key %s for model %s %s", util.HideAPIKey(accountInfo), req.Model, proxyInfo)
-			} else {
-				entry.Debugf("Use API key %s for model %s", util.HideAPIKey(accountInfo), req.Model)
-			}
-		} else if accountType == "oauth" {
-			if proxyInfo != "" {
-				entry.Debugf("Use OAuth %s for model %s %s", accountInfo, req.Model, proxyInfo)
-			} else {
-				entry.Debugf("Use OAuth %s for model %s", accountInfo, req.Model)
-			}
-		}
+		debugLogAuthSelection(entry, auth, provider, req.Model)
 
 		tried[auth.ID] = struct{}{}
 		execCtx := ctx
@@ -450,22 +442,8 @@ func (m *Manager) executeCountWithProvider(ctx context.Context, provider string,
 			return cliproxyexecutor.Response{}, errPick
 		}
 
-		accountType, accountInfo := auth.AccountInfo()
-		proxyInfo := auth.ProxyInfo()
 		entry := logEntryWithRequestID(ctx)
-		if accountType == "api_key" {
-			if proxyInfo != "" {
-				entry.Debugf("Use API key %s for model %s %s", util.HideAPIKey(accountInfo), req.Model, proxyInfo)
-			} else {
-				entry.Debugf("Use API key %s for model %s", util.HideAPIKey(accountInfo), req.Model)
-			}
-		} else if accountType == "oauth" {
-			if proxyInfo != "" {
-				entry.Debugf("Use OAuth %s for model %s %s", accountInfo, req.Model, proxyInfo)
-			} else {
-				entry.Debugf("Use OAuth %s for model %s", accountInfo, req.Model)
-			}
-		}
+		debugLogAuthSelection(entry, auth, provider, req.Model)
 
 		tried[auth.ID] = struct{}{}
 		execCtx := ctx
@@ -512,22 +490,8 @@ func (m *Manager) executeStreamWithProvider(ctx context.Context, provider string
 			return nil, errPick
 		}
 
-		accountType, accountInfo := auth.AccountInfo()
-		proxyInfo := auth.ProxyInfo()
 		entry := logEntryWithRequestID(ctx)
-		if accountType == "api_key" {
-			if proxyInfo != "" {
-				entry.Debugf("Use API key %s for model %s %s", util.HideAPIKey(accountInfo), req.Model, proxyInfo)
-			} else {
-				entry.Debugf("Use API key %s for model %s", util.HideAPIKey(accountInfo), req.Model)
-			}
-		} else if accountType == "oauth" {
-			if proxyInfo != "" {
-				entry.Debugf("Use OAuth %s for model %s %s", accountInfo, req.Model, proxyInfo)
-			} else {
-				entry.Debugf("Use OAuth %s for model %s", accountInfo, req.Model)
-			}
-		}
+		debugLogAuthSelection(entry, auth, provider, req.Model)
 
 		tried[auth.ID] = struct{}{}
 		execCtx := ctx
@@ -1615,6 +1579,23 @@ type RequestPreparer interface {
 	PrepareRequest(req *http.Request, auth *Auth) error
 }
 
+func executorKeyFromAuth(auth *Auth) string {
+	if auth == nil {
+		return ""
+	}
+	if auth.Attributes != nil {
+		providerKey := strings.TrimSpace(auth.Attributes["provider_key"])
+		compatName := strings.TrimSpace(auth.Attributes["compat_name"])
+		if compatName != "" {
+			if providerKey == "" {
+				providerKey = compatName
+			}
+			return strings.ToLower(providerKey)
+		}
+	}
+	return strings.ToLower(strings.TrimSpace(auth.Provider))
+}
+
 // logEntryWithRequestID returns a logrus entry with request_id field if available in context.
 func logEntryWithRequestID(ctx context.Context) *log.Entry {
 	if ctx == nil {
@@ -1626,6 +1607,59 @@ func logEntryWithRequestID(ctx context.Context) *log.Entry {
 	return log.NewEntry(log.StandardLogger())
 }
 
+func debugLogAuthSelection(entry *log.Entry, auth *Auth, provider string, model string) {
+	if !log.IsLevelEnabled(log.DebugLevel) {
+		return
+	}
+	if entry == nil || auth == nil {
+		return
+	}
+	accountType, accountInfo := auth.AccountInfo()
+	proxyInfo := auth.ProxyInfo()
+	suffix := ""
+	if proxyInfo != "" {
+		suffix = " " + proxyInfo
+	}
+	switch accountType {
+	case "api_key":
+		entry.Debugf("Use API key %s for model %s%s", util.HideAPIKey(accountInfo), model, suffix)
+	case "oauth":
+		ident := formatOauthIdentity(auth, provider, accountInfo)
+		entry.Debugf("Use OAuth %s for model %s%s", ident, model, suffix)
+	}
+}
+
+func formatOauthIdentity(auth *Auth, provider string, accountInfo string) string {
+	if auth == nil {
+		return ""
+	}
+	// Prefer the auth's provider when available.
+	providerName := strings.TrimSpace(auth.Provider)
+	if providerName == "" {
+		providerName = strings.TrimSpace(provider)
+	}
+	// Only log the basename to avoid leaking host paths.
+	// FileName may be unset for some auth backends; fall back to ID.
+	authFile := strings.TrimSpace(auth.FileName)
+	if authFile == "" {
+		authFile = strings.TrimSpace(auth.ID)
+	}
+	if authFile != "" {
+		authFile = filepath.Base(authFile)
+	}
+	parts := make([]string, 0, 3)
+	if providerName != "" {
+		parts = append(parts, "provider="+providerName)
+	}
+	if authFile != "" {
+		parts = append(parts, "auth_file="+authFile)
+	}
+	if len(parts) == 0 {
+		return accountInfo
+	}
+	return strings.Join(parts, " ")
+}
+
 // InjectCredentials delegates per-provider HTTP request preparation when supported.
 // If the registered executor for the auth provider implements RequestPreparer,
 // it will be invoked to modify the request (e.g., add headers).
@@ -1637,7 +1671,7 @@ func (m *Manager) InjectCredentials(req *http.Request, authID string) error {
 	a := m.auths[authID]
 	var exec ProviderExecutor
 	if a != nil {
-		exec = m.executors[a.Provider]
+		exec = m.executors[executorKeyFromAuth(a)]
 	}
 	m.mu.RUnlock()
 	if a == nil || exec == nil {
@@ -1648,3 +1682,80 @@ func (m *Manager) InjectCredentials(req *http.Request, authID string) error {
 	}
 	return nil
 }
+
+// PrepareHttpRequest injects provider credentials into the supplied HTTP request.
+func (m *Manager) PrepareHttpRequest(ctx context.Context, auth *Auth, req *http.Request) error {
+	if m == nil {
+		return &Error{Code: "provider_not_found", Message: "manager is nil"}
+	}
+	if auth == nil {
+		return &Error{Code: "auth_not_found", Message: "auth is nil"}
+	}
+	if req == nil {
+		return &Error{Code: "invalid_request", Message: "http request is nil"}
+	}
+	if ctx != nil {
+		*req = *req.WithContext(ctx)
+	}
+	providerKey := executorKeyFromAuth(auth)
+	if providerKey == "" {
+		return &Error{Code: "provider_not_found", Message: "auth provider is empty"}
+	}
+	exec := m.executorFor(providerKey)
+	if exec == nil {
+		return &Error{Code: "provider_not_found", Message: "executor not registered for provider: " + providerKey}
+	}
+	preparer, ok := exec.(RequestPreparer)
+	if !ok || preparer == nil {
+		return &Error{Code: "not_supported", Message: "executor does not support http request preparation"}
+	}
+	return preparer.PrepareRequest(req, auth)
+}
+
+// NewHttpRequest constructs a new HTTP request and injects provider credentials into it.
+func (m *Manager) NewHttpRequest(ctx context.Context, auth *Auth, method, targetURL string, body []byte, headers http.Header) (*http.Request, error) {
+	if ctx == nil {
+		ctx = context.Background()
+	}
+	method = strings.TrimSpace(method)
+	if method == "" {
+		method = http.MethodGet
+	}
+	var reader io.Reader
+	if body != nil {
+		reader = bytes.NewReader(body)
+	}
+	httpReq, err := http.NewRequestWithContext(ctx, method, targetURL, reader)
+	if err != nil {
+		return nil, err
+	}
+	if headers != nil {
+		httpReq.Header = headers.Clone()
+	}
+	if errPrepare := m.PrepareHttpRequest(ctx, auth, httpReq); errPrepare != nil {
+		return nil, errPrepare
+	}
+	return httpReq, nil
+}
+
+// HttpRequest injects provider credentials into the supplied HTTP request and executes it.
+func (m *Manager) HttpRequest(ctx context.Context, auth *Auth, req *http.Request) (*http.Response, error) {
+	if m == nil {
+		return nil, &Error{Code: "provider_not_found", Message: "manager is nil"}
+	}
+	if auth == nil {
+		return nil, &Error{Code: "auth_not_found", Message: "auth is nil"}
+	}
+	if req == nil {
+		return nil, &Error{Code: "invalid_request", Message: "http request is nil"}
+	}
+	providerKey := executorKeyFromAuth(auth)
+	if providerKey == "" {
+		return nil, &Error{Code: "provider_not_found", Message: "auth provider is empty"}
+	}
+	exec := m.executorFor(providerKey)
+	if exec == nil {
+		return nil, &Error{Code: "provider_not_found", Message: "executor not registered for provider: " + providerKey}
+	}
+	return exec.HttpRequest(ctx, auth, req)
+}

sdk/cliproxy/auth/model_name_mappings.go

@@ -165,6 +165,8 @@ func OAuthModelMappingChannel(provider, authKind string) string {
 		return "codex"
 	case "gemini-cli", "aistudio", "antigravity", "qwen", "iflow":
 		return provider
+	case "kiro":
+		return provider
 	default:
 		return ""
 	}

test/builtin_tools_translation_test.go

@@ -0,0 +1,54 @@
+package test
+
+import (
+	"testing"
+
+	_ "github.com/router-for-me/CLIProxyAPI/v6/internal/translator"
+
+	sdktranslator "github.com/router-for-me/CLIProxyAPI/v6/sdk/translator"
+	"github.com/tidwall/gjson"
+)
+
+func TestOpenAIToCodex_PreservesBuiltinTools(t *testing.T) {
+	in := []byte(`{
+		"model":"gpt-5",
+		"messages":[{"role":"user","content":"hi"}],
+		"tools":[{"type":"web_search","search_context_size":"high"}],
+		"tool_choice":{"type":"web_search"}
+	}`)
+
+	out := sdktranslator.TranslateRequest(sdktranslator.FormatOpenAI, sdktranslator.FormatCodex, "gpt-5", in, false)
+
+	if got := gjson.GetBytes(out, "tools.#").Int(); got != 1 {
+		t.Fatalf("expected 1 tool, got %d: %s", got, string(out))
+	}
+	if got := gjson.GetBytes(out, "tools.0.type").String(); got != "web_search" {
+		t.Fatalf("expected tools[0].type=web_search, got %q: %s", got, string(out))
+	}
+	if got := gjson.GetBytes(out, "tools.0.search_context_size").String(); got != "high" {
+		t.Fatalf("expected tools[0].search_context_size=high, got %q: %s", got, string(out))
+	}
+	if got := gjson.GetBytes(out, "tool_choice.type").String(); got != "web_search" {
+		t.Fatalf("expected tool_choice.type=web_search, got %q: %s", got, string(out))
+	}
+}
+
+func TestOpenAIResponsesToOpenAI_PreservesBuiltinTools(t *testing.T) {
+	in := []byte(`{
+		"model":"gpt-5",
+		"input":[{"role":"user","content":[{"type":"input_text","text":"hi"}]}],
+		"tools":[{"type":"web_search","search_context_size":"low"}]
+	}`)
+
+	out := sdktranslator.TranslateRequest(sdktranslator.FormatOpenAIResponse, sdktranslator.FormatOpenAI, "gpt-5", in, false)
+
+	if got := gjson.GetBytes(out, "tools.#").Int(); got != 1 {
+		t.Fatalf("expected 1 tool, got %d: %s", got, string(out))
+	}
+	if got := gjson.GetBytes(out, "tools.0.type").String(); got != "web_search" {
+		t.Fatalf("expected tools[0].type=web_search, got %q: %s", got, string(out))
+	}
+	if got := gjson.GetBytes(out, "tools.0.search_context_size").String(); got != "low" {
+		t.Fatalf("expected tools[0].search_context_size=low, got %q: %s", got, string(out))
+	}
+}