test

refactor(executor): remove `prompt_cache_retention` from request payloads

config.example.yaml

@@ -202,6 +202,7 @@ ws-auth: false
 # These mappings rename model IDs for both model listing and request routing.
 # Supported channels: gemini-cli, vertex, aistudio, antigravity, claude, codex, qwen, iflow.
 # NOTE: Mappings do not apply to gemini-api-key, codex-api-key, claude-api-key, openai-compatibility, vertex-api-key, or ampcode.
+# You can repeat the same name with different aliases to expose multiple client model names.
 # oauth-model-mappings:
 #   gemini-cli:
 #     - name: "gemini-2.5-pro"          # original model name under this channel

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

@@ -460,6 +460,12 @@ 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/middleware/request_logging.go

@@ -115,7 +115,7 @@ func shouldLogRequest(path string) bool {
 	}
 
 	if strings.HasPrefix(path, "/api") {
-		return strings.HasPrefix(path, "/api/provider")
+		return strings.HasPrefix(path, "/api/provider") || strings.HasPrefix(path, "/api/internal")
 	}
 
 	return true

internal/cache/signature_cache.go

@@ -3,7 +3,6 @@ package cache
 import (
 	"crypto/sha256"
 	"encoding/hex"
-	"sort"
 	"sync"
 	"time"
 )
@@ -16,21 +15,24 @@ type SignatureEntry struct {
 
 const (
 	// SignatureCacheTTL is how long signatures are valid
-	SignatureCacheTTL = 1 * time.Hour
-
-	// MaxEntriesPerSession limits memory usage per session
-	MaxEntriesPerSession = 100
+	SignatureCacheTTL = 3 * time.Hour
 
 	// SignatureTextHashLen is the length of the hash key (16 hex chars = 64-bit key space)
 	SignatureTextHashLen = 16
 
 	// 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 +47,9 @@ 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 +58,40 @@ 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) {
@@ -69,43 +108,6 @@ func CacheSignature(sessionID, text, signature string) {
 	sc.mu.Lock()
 	defer sc.mu.Unlock()
 
-	// Evict expired entries if at capacity
-	if len(sc.entries) >= MaxEntriesPerSession {
-		now := time.Now()
-		for key, entry := range sc.entries {
-			if now.Sub(entry.Timestamp) > SignatureCacheTTL {
-				delete(sc.entries, key)
-			}
-		}
-		// If still at capacity, remove oldest entries
-		if len(sc.entries) >= MaxEntriesPerSession {
-			// Find and remove oldest quarter
-			oldest := make([]struct {
-				key string
-				ts  time.Time
-			}, 0, len(sc.entries))
-			for key, entry := range sc.entries {
-				oldest = append(oldest, struct {
-					key string
-					ts  time.Time
-				}{key, entry.Timestamp})
-			}
-			// Sort by timestamp (oldest first) using sort.Slice
-			sort.Slice(oldest, func(i, j int) bool {
-				return oldest[i].ts.Before(oldest[j].ts)
-			})
-
-			toRemove := len(oldest) / 4
-			if toRemove < 1 {
-				toRemove = 1
-			}
-
-			for i := 0; i < toRemove; i++ {
-				delete(sc.entries, oldest[i].key)
-			}
-		}
-	}
-
 	sc.entries[textHash] = SignatureEntry{
 		Signature: signature,
 		Timestamp: time.Now(),
@@ -127,22 +129,25 @@ func GetCachedSignature(sessionID, text string) string {
 
 	textHash := hashText(text)
 
-	sc.mu.RLock()
-	entry, exists := sc.entries[textHash]
-	sc.mu.RUnlock()
+	now := time.Now()
 
+	sc.mu.Lock()
+	entry, exists := sc.entries[textHash]
 	if !exists {
+		sc.mu.Unlock()
 		return ""
 	}
-
-	// Check if expired
-	if time.Since(entry.Timestamp) > SignatureCacheTTL {
-		sc.mu.Lock()
+	if now.Sub(entry.Timestamp) > SignatureCacheTTL {
 		delete(sc.entries, textHash)
 		sc.mu.Unlock()
 		return ""
 	}
 
+	// Refresh TTL on access (sliding expiration).
+	entry.Timestamp = now
+	sc.entries[textHash] = entry
+	sc.mu.Unlock()
+
 	return entry.Signature
 }
 

internal/config/config.go

@@ -521,7 +521,7 @@ func LoadConfigOptional(configFile string, optional bool) (*Config, error) {
 
 // SanitizeOAuthModelMappings normalizes and deduplicates global OAuth model name mappings.
 // It trims whitespace, normalizes channel keys to lower-case, drops empty entries,
-// and ensures (From, To) pairs are unique within each channel.
+// allows multiple aliases per upstream name, and ensures aliases are unique within each channel.
 func (cfg *Config) SanitizeOAuthModelMappings() {
 	if cfg == nil || len(cfg.OAuthModelMappings) == 0 {
 		return
@@ -532,7 +532,6 @@ func (cfg *Config) SanitizeOAuthModelMappings() {
 		if channel == "" || len(mappings) == 0 {
 			continue
 		}
-		seenName := make(map[string]struct{}, len(mappings))
 		seenAlias := make(map[string]struct{}, len(mappings))
 		clean := make([]ModelNameMapping, 0, len(mappings))
 		for _, mapping := range mappings {
@@ -544,15 +543,10 @@ func (cfg *Config) SanitizeOAuthModelMappings() {
 			if strings.EqualFold(name, alias) {
 				continue
 			}
-			nameKey := strings.ToLower(name)
 			aliasKey := strings.ToLower(alias)
-			if _, ok := seenName[nameKey]; ok {
-				continue
-			}
 			if _, ok := seenAlias[aliasKey]; ok {
 				continue
 			}
-			seenName[nameKey] = struct{}{}
 			seenAlias[aliasKey] = struct{}{}
 			clean = append(clean, ModelNameMapping{Name: name, Alias: alias, Fork: mapping.Fork})
 		}

internal/config/oauth_model_mappings_test.go

@@ -25,3 +25,32 @@ func TestSanitizeOAuthModelMappings_PreservesForkFlag(t *testing.T) {
 		t.Fatalf("expected second mapping to be gpt-6->g6 fork=false, got name=%q alias=%q fork=%v", mappings[1].Name, mappings[1].Alias, mappings[1].Fork)
 	}
 }
+
+func TestSanitizeOAuthModelMappings_AllowsMultipleAliasesForSameName(t *testing.T) {
+	cfg := &Config{
+		OAuthModelMappings: map[string][]ModelNameMapping{
+			"antigravity": {
+				{Name: "gemini-claude-opus-4-5-thinking", Alias: "claude-opus-4-5-20251101", Fork: true},
+				{Name: "gemini-claude-opus-4-5-thinking", Alias: "claude-opus-4-5-20251101-thinking", Fork: true},
+				{Name: "gemini-claude-opus-4-5-thinking", Alias: "claude-opus-4-5", Fork: true},
+			},
+		},
+	}
+
+	cfg.SanitizeOAuthModelMappings()
+
+	mappings := cfg.OAuthModelMappings["antigravity"]
+	expected := []ModelNameMapping{
+		{Name: "gemini-claude-opus-4-5-thinking", Alias: "claude-opus-4-5-20251101", Fork: true},
+		{Name: "gemini-claude-opus-4-5-thinking", Alias: "claude-opus-4-5-20251101-thinking", Fork: true},
+		{Name: "gemini-claude-opus-4-5-thinking", Alias: "claude-opus-4-5", Fork: true},
+	}
+	if len(mappings) != len(expected) {
+		t.Fatalf("expected %d sanitized mappings, got %d", len(expected), len(mappings))
+	}
+	for i, exp := range expected {
+		if mappings[i].Name != exp.Name || mappings[i].Alias != exp.Alias || mappings[i].Fork != exp.Fork {
+			t.Fatalf("expected mapping %d to be name=%q alias=%q fork=%v, got name=%q alias=%q fork=%v", i, exp.Name, exp.Alias, exp.Fork, mappings[i].Name, mappings[i].Alias, mappings[i].Fork)
+		}
+	}
+}

internal/misc/codex_instructions.go

@@ -7,12 +7,77 @@ import (
 	"embed"
 	_ "embed"
 	"strings"
+
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
 )
 
 //go:embed codex_instructions
 var codexInstructionsDir embed.FS
 
-func CodexInstructionsForModel(modelName, systemInstructions string) (bool, string) {
+//go:embed opencode_codex_instructions.txt
+var opencodeCodexInstructions string
+
+const (
+	codexUserAgentKey  = "__cpa_user_agent"
+	userAgentOpenAISDK = "ai-sdk/openai/"
+)
+
+func InjectCodexUserAgent(raw []byte, userAgent string) []byte {
+	if len(raw) == 0 {
+		return raw
+	}
+	trimmed := strings.TrimSpace(userAgent)
+	if trimmed == "" {
+		return raw
+	}
+	updated, err := sjson.SetBytes(raw, codexUserAgentKey, trimmed)
+	if err != nil {
+		return raw
+	}
+	return updated
+}
+
+func ExtractCodexUserAgent(raw []byte) string {
+	if len(raw) == 0 {
+		return ""
+	}
+	return strings.TrimSpace(gjson.GetBytes(raw, codexUserAgentKey).String())
+}
+
+func StripCodexUserAgent(raw []byte) []byte {
+	if len(raw) == 0 {
+		return raw
+	}
+	if !gjson.GetBytes(raw, codexUserAgentKey).Exists() {
+		return raw
+	}
+	updated, err := sjson.DeleteBytes(raw, codexUserAgentKey)
+	if err != nil {
+		return raw
+	}
+	return updated
+}
+
+func codexInstructionsForOpenCode(systemInstructions string) (bool, string) {
+	if opencodeCodexInstructions == "" {
+		return false, ""
+	}
+	if strings.HasPrefix(systemInstructions, opencodeCodexInstructions) {
+		return true, ""
+	}
+	return false, opencodeCodexInstructions
+}
+
+func useOpenCodeInstructions(userAgent string) bool {
+	return strings.Contains(strings.ToLower(userAgent), userAgentOpenAISDK)
+}
+
+func IsOpenCodeUserAgent(userAgent string) bool {
+	return useOpenCodeInstructions(userAgent)
+}
+
+func codexInstructionsForCodex(modelName, systemInstructions string) (bool, string) {
 	entries, _ := codexInstructionsDir.ReadDir("codex_instructions")
 
 	lastPrompt := ""
@@ -57,3 +122,10 @@ func CodexInstructionsForModel(modelName, systemInstructions string) (bool, stri
 		return false, lastPrompt
 	}
 }
+
+func CodexInstructionsForModel(modelName, systemInstructions, userAgent string) (bool, string) {
+	if IsOpenCodeUserAgent(userAgent) {
+		return codexInstructionsForOpenCode(systemInstructions)
+	}
+	return codexInstructionsForCodex(modelName, systemInstructions)
+}

internal/misc/opencode_codex_instructions.txt

@@ -0,0 +1,318 @@
+You are a coding agent running in the opencode, a terminal-based coding assistant. opencode is an open source project. You are expected to be precise, safe, and helpful.
+
+Your capabilities:
+
+- Receive user prompts and other context provided by the harness, such as files in the workspace.
+- Communicate with the user by streaming thinking & responses, and by making & updating plans.
+- Emit function calls to run terminal commands and apply edits. Depending on how this specific run is configured, you can request that these function calls be escalated to the user for approval before running. More on this in the "Sandbox and approvals" section.
+
+Within this context, Codex refers to the open-source agentic coding interface (not the old Codex language model built by OpenAI).
+
+# How you work
+
+## Personality
+
+Your default personality and tone is concise, direct, and friendly. You communicate efficiently, always keeping the user clearly informed about ongoing actions without unnecessary detail. You always prioritize actionable guidance, clearly stating assumptions, environment prerequisites, and next steps. Unless explicitly asked, you avoid excessively verbose explanations about your work.
+
+# AGENTS.md spec
+- Repos often contain AGENTS.md files. These files can appear anywhere within the repository.
+- These files are a way for humans to give you (the agent) instructions or tips for working within the container.
+- Some examples might be: coding conventions, info about how code is organized, or instructions for how to run or test code.
+- Instructions in AGENTS.md files:
+    - The scope of an AGENTS.md file is the entire directory tree rooted at the folder that contains it.
+    - For every file you touch in the final patch, you must obey instructions in any AGENTS.md file whose scope includes that file.
+    - Instructions about code style, structure, naming, etc. apply only to code within the AGENTS.md file's scope, unless the file states otherwise.
+    - More-deeply-nested AGENTS.md files take precedence in the case of conflicting instructions.
+    - Direct system/developer/user instructions (as part of a prompt) take precedence over AGENTS.md instructions.
+- The contents of the AGENTS.md file at the root of the repo and any directories from the CWD up to the root are included with the developer message and don't need to be re-read. When working in a subdirectory of CWD, or a directory outside the CWD, check for any AGENTS.md files that may be applicable.
+
+## Responsiveness
+
+### Preamble messages
+
+Before making tool calls, send a brief preamble to the user explaining what you’re about to do. When sending preamble messages, follow these principles and examples:
+
+- **Logically group related actions**: if you’re about to run several related commands, describe them together in one preamble rather than sending a separate note for each.
+- **Keep it concise**: be no more than 1-2 sentences, focused on immediate, tangible next steps. (8–12 words for quick updates).
+- **Build on prior context**: if this is not your first tool call, use the preamble message to connect the dots with what’s been done so far and create a sense of momentum and clarity for the user to understand your next actions.
+- **Keep your tone light, friendly and curious**: add small touches of personality in preambles feel collaborative and engaging.
+- **Exception**: Avoid adding a preamble for every trivial read (e.g., `cat` a single file) unless it’s part of a larger grouped action.
+
+**Examples:**
+
+- “I’ve explored the repo; now checking the API route definitions.”
+- “Next, I’ll patch the config and update the related tests.”
+- “I’m about to scaffold the CLI commands and helper functions.”
+- “Ok cool, so I’ve wrapped my head around the repo. Now digging into the API routes.”
+- “Config’s looking tidy. Next up is editing helpers to keep things in sync.”
+- “Finished poking at the DB gateway. I will now chase down error handling.”
+- “Alright, build pipeline order is interesting. Checking how it reports failures.”
+- “Spotted a clever caching util; now hunting where it gets used.”
+
+## Planning
+
+You have access to an `todowrite` tool which tracks steps and progress and renders them to the user. Using the tool helps demonstrate that you've understood the task and convey how you're approaching it. Plans can help to make complex, ambiguous, or multi-phase work clearer and more collaborative for the user. A good plan should break the task into meaningful, logically ordered steps that are easy to verify as you go.
+
+Note that plans are not for padding out simple work with filler steps or stating the obvious. The content of your plan should not involve doing anything that you aren't capable of doing (i.e. don't try to test things that you can't test). Do not use plans for simple or single-step queries that you can just do or answer immediately.
+
+Do not repeat the full contents of the plan after an `todowrite` call — the harness already displays it. Instead, summarize the change made and highlight any important context or next step.
+
+Before running a command, consider whether or not you have completed the
+previous step, and make sure to mark it as completed before moving on to the
+next step. It may be the case that you complete all steps in your plan after a
+single pass of implementation. If this is the case, you can simply mark all the
+planned steps as completed. Sometimes, you may need to change plans in the
+middle of a task: call `todowrite` with the updated plan and make sure to provide an `explanation` of the rationale when doing so.
+
+Use a plan when:
+
+- The task is non-trivial and will require multiple actions over a long time horizon.
+- There are logical phases or dependencies where sequencing matters.
+- The work has ambiguity that benefits from outlining high-level goals.
+- You want intermediate checkpoints for feedback and validation.
+- When the user asked you to do more than one thing in a single prompt
+- The user has asked you to use the plan tool (aka "TODOs")
+- You generate additional steps while working, and plan to do them before yielding to the user
+
+### Examples
+
+**High-quality plans**
+
+Example 1:
+
+1. Add CLI entry with file args
+2. Parse Markdown via CommonMark library
+3. Apply semantic HTML template
+4. Handle code blocks, images, links
+5. Add error handling for invalid files
+
+Example 2:
+
+1. Define CSS variables for colors
+2. Add toggle with localStorage state
+3. Refactor components to use variables
+4. Verify all views for readability
+5. Add smooth theme-change transition
+
+Example 3:
+
+1. Set up Node.js + WebSocket server
+2. Add join/leave broadcast events
+3. Implement messaging with timestamps
+4. Add usernames + mention highlighting
+5. Persist messages in lightweight DB
+6. Add typing indicators + unread count
+
+**Low-quality plans**
+
+Example 1:
+
+1. Create CLI tool
+2. Add Markdown parser
+3. Convert to HTML
+
+Example 2:
+
+1. Add dark mode toggle
+2. Save preference
+3. Make styles look good
+
+Example 3:
+
+1. Create single-file HTML game
+2. Run quick sanity check
+3. Summarize usage instructions
+
+If you need to write a plan, only write high quality plans, not low quality ones.
+
+## Task execution
+
+You are a coding agent. Please keep going until the query is completely resolved, before ending your turn and yielding back to the user. Only terminate your turn when you are sure that the problem is solved. Autonomously resolve the query to the best of your ability, using the tools available to you, before coming back to the user. Do NOT guess or make up an answer.
+
+You MUST adhere to the following criteria when solving queries:
+
+- Working on the repo(s) in the current environment is allowed, even if they are proprietary.
+- Analyzing code for vulnerabilities is allowed.
+- Showing user code and tool call details is allowed.
+- Use the `edit` tool to edit files
+
+If completing the user's task requires writing or modifying files, your code and final answer should follow these coding guidelines, though user instructions (i.e. AGENTS.md) may override these guidelines:
+
+- Fix the problem at the root cause rather than applying surface-level patches, when possible.
+- Avoid unneeded complexity in your solution.
+- Do not attempt to fix unrelated bugs or broken tests. It is not your responsibility to fix them. (You may mention them to the user in your final message though.)
+- Update documentation as necessary.
+- Keep changes consistent with the style of the existing codebase. Changes should be minimal and focused on the task.
+- Use `git log` and `git blame` to search the history of the codebase if additional context is required.
+- NEVER add copyright or license headers unless specifically requested.
+- Do not waste tokens by re-reading files after calling `edit` on them. The tool call will fail if it didn't work. The same goes for making folders, deleting folders, etc.
+- Do not `git commit` your changes or create new git branches unless explicitly requested.
+- Do not add inline comments within code unless explicitly requested.
+- Do not use one-letter variable names unless explicitly requested.
+- NEVER output inline citations like "【F:README.md†L5-L14】" in your outputs. The CLI is not able to render these so they will just be broken in the UI. Instead, if you output valid filepaths, users will be able to click on them to open the files in their editor.
+
+## Sandbox and approvals
+
+The Codex CLI harness supports several different sandboxing, and approval configurations that the user can choose from.
+
+Filesystem sandboxing prevents you from editing files without user approval. The options are:
+
+- **read-only**: You can only read files.
+- **workspace-write**: You can read files. You can write to files in your workspace folder, but not outside it.
+- **danger-full-access**: No filesystem sandboxing.
+
+Network sandboxing prevents you from accessing network without approval. Options are
+
+- **restricted**
+- **enabled**
+
+Approvals are your mechanism to get user consent to perform more privileged actions. Although they introduce friction to the user because your work is paused until the user responds, you should leverage them to accomplish your important work. Do not let these settings or the sandbox deter you from attempting to accomplish the user's task. Approval options are
+
+- **untrusted**: The harness will escalate most commands for user approval, apart from a limited allowlist of safe "read" commands.
+- **on-failure**: The harness will allow all commands to run in the sandbox (if enabled), and failures will be escalated to the user for approval to run again without the sandbox.
+- **on-request**: Commands will be run in the sandbox by default, and you can specify in your tool call if you want to escalate a command to run without sandboxing. (Note that this mode is not always available. If it is, you'll see parameters for it in the `shell` command description.)
+- **never**: This is a non-interactive mode where you may NEVER ask the user for approval to run commands. Instead, you must always persist and work around constraints to solve the task for the user. You MUST do your utmost best to finish the task and validate your work before yielding. If this mode is pared with `danger-full-access`, take advantage of it to deliver the best outcome for the user. Further, in this mode, your default testing philosophy is overridden: Even if you don't see local patterns for testing, you may add tests and scripts to validate your work. Just remove them before yielding.
+
+When you are running with approvals `on-request`, and sandboxing enabled, here are scenarios where you'll need to request approval:
+
+- You need to run a command that writes to a directory that requires it (e.g. running tests that write to /tmp)
+- You need to run a GUI app (e.g., open/xdg-open/osascript) to open browsers or files.
+- You are running sandboxed and need to run a command that requires network access (e.g. installing packages)
+- If you run a command that is important to solving the user's query, but it fails because of sandboxing, rerun the command with approval.
+- You are about to take a potentially destructive action such as an `rm` or `git reset` that the user did not explicitly ask for
+- (For all of these, you should weigh alternative paths that do not require approval.)
+
+Note that when sandboxing is set to read-only, you'll need to request approval for any command that isn't a read.
+
+You will be told what filesystem sandboxing, network sandboxing, and approval mode are active in a developer or user message. If you are not told about this, assume that you are running with workspace-write, network sandboxing ON, and approval on-failure.
+
+## Validating your work
+
+If the codebase has tests or the ability to build or run, consider using them to verify that your work is complete.
+
+When testing, your philosophy should be to start as specific as possible to the code you changed so that you can catch issues efficiently, then make your way to broader tests as you build confidence. If there's no test for the code you changed, and if the adjacent patterns in the codebases show that there's a logical place for you to add a test, you may do so. However, do not add tests to codebases with no tests.
+
+Similarly, once you're confident in correctness, you can suggest or use formatting commands to ensure that your code is well formatted. If there are issues you can iterate up to 3 times to get formatting right, but if you still can't manage it's better to save the user time and present them a correct solution where you call out the formatting in your final message. If the codebase does not have a formatter configured, do not add one.
+
+For all of testing, running, building, and formatting, do not attempt to fix unrelated bugs. It is not your responsibility to fix them. (You may mention them to the user in your final message though.)
+
+Be mindful of whether to run validation commands proactively. In the absence of behavioral guidance:
+
+- When running in non-interactive approval modes like **never** or **on-failure**, proactively run tests, lint and do whatever you need to ensure you've completed the task.
+- When working in interactive approval modes like **untrusted**, or **on-request**, hold off on running tests or lint commands until the user is ready for you to finalize your output, because these commands take time to run and slow down iteration. Instead suggest what you want to do next, and let the user confirm first.
+- When working on test-related tasks, such as adding tests, fixing tests, or reproducing a bug to verify behavior, you may proactively run tests regardless of approval mode. Use your judgement to decide whether this is a test-related task.
+
+## Ambition vs. precision
+
+For tasks that have no prior context (i.e. the user is starting something brand new), you should feel free to be ambitious and demonstrate creativity with your implementation.
+
+If you're operating in an existing codebase, you should make sure you do exactly what the user asks with surgical precision. Treat the surrounding codebase with respect, and don't overstep (i.e. changing filenames or variables unnecessarily). You should balance being sufficiently ambitious and proactive when completing tasks of this nature.
+
+You should use judicious initiative to decide on the right level of detail and complexity to deliver based on the user's needs. This means showing good judgment that you're capable of doing the right extras without gold-plating. This might be demonstrated by high-value, creative touches when scope of the task is vague; while being surgical and targeted when scope is tightly specified.
+
+## Sharing progress updates
+
+For especially longer tasks that you work on (i.e. requiring many tool calls, or a plan with multiple steps), you should provide progress updates back to the user at reasonable intervals. These updates should be structured as a concise sentence or two (no more than 8-10 words long) recapping progress so far in plain language: this update demonstrates your understanding of what needs to be done, progress so far (i.e. files explores, subtasks complete), and where you're going next.
+
+Before doing large chunks of work that may incur latency as experienced by the user (i.e. writing a new file), you should send a concise message to the user with an update indicating what you're about to do to ensure they know what you're spending time on. Don't start editing or writing large files before informing the user what you are doing and why.
+
+The messages you send before tool calls should describe what is immediately about to be done next in very concise language. If there was previous work done, this preamble message should also include a note about the work done so far to bring the user along.
+
+## Presenting your work and final message
+
+Your final message should read naturally, like an update from a concise teammate. For casual conversation, brainstorming tasks, or quick questions from the user, respond in a friendly, conversational tone. You should ask questions, suggest ideas, and adapt to the user’s style. If you've finished a large amount of work, when describing what you've done to the user, you should follow the final answer formatting guidelines to communicate substantive changes. You don't need to add structured formatting for one-word answers, greetings, or purely conversational exchanges.
+
+You can skip heavy formatting for single, simple actions or confirmations. In these cases, respond in plain sentences with any relevant next step or quick option. Reserve multisection structured responses for results that need grouping or explanation.
+
+The user is working on the same computer as you, and has access to your work. As such there's no need to show the full contents of large files you have already written unless the user explicitly asks for them. Similarly, if you've created or modified files using `edit`, there's no need to tell users to "save the file" or "copy the code into a file"—just reference the file path.
+
+If there's something that you think you could help with as a logical next step, concisely ask the user if they want you to do so. Good examples of this are running tests, committing changes, or building out the next logical component. If there’s something that you couldn't do (even with approval) but that the user might want to do (such as verifying changes by running the app), include those instructions succinctly.
+
+Brevity is very important as a default. You should be very concise (i.e. no more than 10 lines), but can relax this requirement for tasks where additional detail and comprehensiveness is important for the user's understanding.
+
+### Final answer structure and style guidelines
+
+You are producing plain text that will later be styled by the CLI. Follow these rules exactly. Formatting should make results easy to scan, but not feel mechanical. Use judgment to decide how much structure adds value.
+
+**Section Headers**
+
+- Use only when they improve clarity — they are not mandatory for every answer.
+- Choose descriptive names that fit the content
+- Keep headers short (1–3 words) and in `**Title Case**`. Always start headers with `**` and end with `**`
+- Leave no blank line before the first bullet under a header.
+- Section headers should only be used where they genuinely improve scannability; avoid fragmenting the answer.
+
+**Bullets**
+
+- Use `-` followed by a space for every bullet.
+- Merge related points when possible; avoid a bullet for every trivial detail.
+- Keep bullets to one line unless breaking for clarity is unavoidable.
+- Group into short lists (4–6 bullets) ordered by importance.
+- Use consistent keyword phrasing and formatting across sections.
+
+**Monospace**
+
+- Wrap all commands, file paths, env vars, and code identifiers in backticks (`` `...` ``).
+- Apply to inline examples and to bullet keywords if the keyword itself is a literal file/command.
+- Never mix monospace and bold markers; choose one based on whether it’s a keyword (`**`) or inline code/path (`` ` ``).
+
+**File References**
+When referencing files in your response, make sure to include the relevant start line and always follow the below rules:
+  * Use inline code to make file paths clickable.
+  * Each reference should have a standalone path. Even if it's the same file.
+  * Accepted: absolute, workspace‑relative, a/ or b/ diff prefixes, or bare filename/suffix.
+  * Line/column (1‑based, optional): :line[:column] or #Lline[Ccolumn] (column defaults to 1).
+  * Do not use URIs like file://, vscode://, or https://.
+  * Do not provide range of lines
+  * Examples: src/app.ts, src/app.ts:42, b/server/index.js#L10, C:\repo\project\main.rs:12:5
+
+**Structure**
+
+- Place related bullets together; don’t mix unrelated concepts in the same section.
+- Order sections from general → specific → supporting info.
+- For subsections (e.g., “Binaries” under “Rust Workspace”), introduce with a bolded keyword bullet, then list items under it.
+- Match structure to complexity:
+  - Multi-part or detailed results → use clear headers and grouped bullets.
+  - Simple results → minimal headers, possibly just a short list or paragraph.
+
+**Tone**
+
+- Keep the voice collaborative and natural, like a coding partner handing off work.
+- Be concise and factual — no filler or conversational commentary and avoid unnecessary repetition
+- Use present tense and active voice (e.g., “Runs tests” not “This will run tests”).
+- Keep descriptions self-contained; don’t refer to “above” or “below”.
+- Use parallel structure in lists for consistency.
+
+**Don’t**
+
+- Don’t use literal words “bold” or “monospace” in the content.
+- Don’t nest bullets or create deep hierarchies.
+- Don’t output ANSI escape codes directly — the CLI renderer applies them.
+- Don’t cram unrelated keywords into a single bullet; split for clarity.
+- Don’t let keyword lists run long — wrap or reformat for scannability.
+
+Generally, ensure your final answers adapt their shape and depth to the request. For example, answers to code explanations should have a precise, structured explanation with code references that answer the question directly. For tasks with a simple implementation, lead with the outcome and supplement only with what’s needed for clarity. Larger changes can be presented as a logical walkthrough of your approach, grouping related steps, explaining rationale where it adds value, and highlighting next actions to accelerate the user. Your answers should provide the right level of detail while being easily scannable.
+
+For casual greetings, acknowledgements, or other one-off conversational messages that are not delivering substantive information or structured results, respond naturally without section headers or bullet formatting.
+
+# Tool Guidelines
+
+## Shell commands
+
+When using the shell, you must adhere to the following guidelines:
+
+- When searching for text or files, prefer using `rg` or `rg --files` respectively because `rg` is much faster than alternatives like `grep`. (If the `rg` command is not found, then use alternatives.)
+- Read files in chunks with a max chunk size of 250 lines. Do not use python scripts to attempt to output larger chunks of a file. Command line output will be truncated after 10 kilobytes or 256 lines of output, regardless of the command used.
+
+## `todowrite`
+
+A tool named `todowrite` is available to you. You can use it to keep an up‑to‑date, step‑by‑step plan for the task.
+
+To create a new plan, call `todowrite` with a short list of 1‑sentence steps (no more than 5-7 words each) with a `status` for each step (`pending`, `in_progress`, or `completed`).
+
+When steps have been completed, use `todowrite` to mark each finished step as
+`completed` and the next step you are working on as `in_progress`. There should
+always be exactly one `in_progress` step until everything is done. You can mark
+multiple items as complete in a single `todowrite` call.
+
+If all steps are complete, ensure you call `todowrite` to mark all steps as `completed`.

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

@@ -73,8 +73,37 @@ 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) {

internal/runtime/executor/cache_helpers.go

@@ -1,10 +1,68 @@
 package executor
 
-import "time"
+import (
+	"sync"
+	"time"
+)
 
 type codexCache struct {
 	ID     string
 	Expire time.Time
 }
 
-var codexCacheMap = map[string]codexCache{}
+// codexCacheMap stores prompt cache IDs keyed by model+user_id.
+// Protected by codexCacheMu. Entries expire after 1 hour.
+var (
+	codexCacheMap = make(map[string]codexCache)
+	codexCacheMu  sync.RWMutex
+)
+
+// 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) {
+	codexCacheCleanupOnce.Do(startCodexCacheCleanup)
+	codexCacheMu.RLock()
+	cache, ok := codexCacheMap[key]
+	codexCacheMu.RUnlock()
+	if !ok || cache.Expire.Before(time.Now()) {
+		return codexCache{}, false
+	}
+	return cache, true
+}
+
+// setCodexCache stores a cache entry.
+func setCodexCache(key string, cache codexCache) {
+	codexCacheCleanupOnce.Do(startCodexCacheCleanup)
+	codexCacheMu.Lock()
+	codexCacheMap[key] = cache
+	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)
@@ -56,12 +87,16 @@ func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("codex")
+	userAgent := codexUserAgent(ctx)
 	originalPayload := bytes.Clone(req.Payload)
 	if len(opts.OriginalRequest) > 0 {
 		originalPayload = bytes.Clone(opts.OriginalRequest)
 	}
+	originalPayload = misc.InjectCodexUserAgent(originalPayload, userAgent)
 	originalTranslated := sdktranslator.TranslateRequest(from, to, model, originalPayload, false)
-	body := sdktranslator.TranslateRequest(from, to, model, bytes.Clone(req.Payload), false)
+	body := misc.InjectCodexUserAgent(bytes.Clone(req.Payload), userAgent)
+	body = sdktranslator.TranslateRequest(from, to, model, body, false)
+	body = misc.StripCodexUserAgent(body)
 	body = ApplyReasoningEffortMetadata(body, req.Metadata, model, "reasoning.effort", false)
 	body = NormalizeThinkingConfig(body, model, false)
 	if errValidate := ValidateThinkingConfig(body, model); errValidate != nil {
@@ -71,6 +106,7 @@ func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	body, _ = sjson.SetBytes(body, "model", model)
 	body, _ = sjson.SetBytes(body, "stream", true)
 	body, _ = sjson.DeleteBytes(body, "previous_response_id")
+	body, _ = sjson.DeleteBytes(body, "prompt_cache_retention")
 
 	url := strings.TrimSuffix(baseURL, "/") + "/responses"
 	httpReq, err := e.cacheHelper(ctx, from, url, req, body)
@@ -137,7 +173,7 @@ func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 		}
 
 		var param any
-		out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), body, line, &param)
+		out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(originalPayload), body, line, &param)
 		resp = cliproxyexecutor.Response{Payload: []byte(out)}
 		return resp, nil
 	}
@@ -161,12 +197,16 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("codex")
+	userAgent := codexUserAgent(ctx)
 	originalPayload := bytes.Clone(req.Payload)
 	if len(opts.OriginalRequest) > 0 {
 		originalPayload = bytes.Clone(opts.OriginalRequest)
 	}
+	originalPayload = misc.InjectCodexUserAgent(originalPayload, userAgent)
 	originalTranslated := sdktranslator.TranslateRequest(from, to, model, originalPayload, true)
-	body := sdktranslator.TranslateRequest(from, to, model, bytes.Clone(req.Payload), true)
+	body := misc.InjectCodexUserAgent(bytes.Clone(req.Payload), userAgent)
+	body = sdktranslator.TranslateRequest(from, to, model, body, true)
+	body = misc.StripCodexUserAgent(body)
 
 	body = ApplyReasoningEffortMetadata(body, req.Metadata, model, "reasoning.effort", false)
 	body = NormalizeThinkingConfig(body, model, false)
@@ -175,6 +215,7 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 	}
 	body = applyPayloadConfigWithRoot(e.cfg, model, to.String(), "", body, originalTranslated)
 	body, _ = sjson.DeleteBytes(body, "previous_response_id")
+	body, _ = sjson.DeleteBytes(body, "prompt_cache_retention")
 	body, _ = sjson.SetBytes(body, "model", model)
 
 	url := strings.TrimSuffix(baseURL, "/") + "/responses"
@@ -247,7 +288,7 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 				}
 			}
 
-			chunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), body, bytes.Clone(line), &param)
+			chunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(originalPayload), body, bytes.Clone(line), &param)
 			for i := range chunks {
 				out <- cliproxyexecutor.StreamChunk{Payload: []byte(chunks[i])}
 			}
@@ -269,11 +310,15 @@ func (e *CodexExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("codex")
-	body := sdktranslator.TranslateRequest(from, to, model, bytes.Clone(req.Payload), false)
+	userAgent := codexUserAgent(ctx)
+	body := misc.InjectCodexUserAgent(bytes.Clone(req.Payload), userAgent)
+	body = sdktranslator.TranslateRequest(from, to, model, body, false)
+	body = misc.StripCodexUserAgent(body)
 
 	body = ApplyReasoningEffortMetadata(body, req.Metadata, model, "reasoning.effort", false)
 	body, _ = sjson.SetBytes(body, "model", model)
 	body, _ = sjson.DeleteBytes(body, "previous_response_id")
+	body, _ = sjson.DeleteBytes(body, "prompt_cache_retention")
 	body, _ = sjson.SetBytes(body, "stream", false)
 
 	enc, err := tokenizerForCodexModel(model)
@@ -457,14 +502,14 @@ 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 = codexCacheMap[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),
 				}
-				codexCacheMap[key] = cache
+				setCodexCache(key, cache)
 			}
 		}
 	} else if from == "openai-response" {
@@ -522,6 +567,16 @@ func applyCodexHeaders(r *http.Request, auth *cliproxyauth.Auth, token string) {
 	util.ApplyCustomHeadersFromAttrs(r, attrs)
 }
 
+func codexUserAgent(ctx context.Context) string {
+	if ctx == nil {
+		return ""
+	}
+	if ginCtx, ok := ctx.Value("gin").(*gin.Context); ok && ginCtx != nil && ginCtx.Request != nil {
+		return strings.TrimSpace(ginCtx.Request.UserAgent())
+	}
+	return ""
+}
+
 func codexCreds(a *cliproxyauth.Auth) (apiKey, baseURL string) {
 	if a == nil {
 		return "", ""

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/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/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)

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_request.go

@@ -37,10 +37,11 @@ import (
 //   - []byte: The transformed request data in internal client format
 func ConvertClaudeRequestToCodex(modelName string, inputRawJSON []byte, _ bool) []byte {
 	rawJSON := bytes.Clone(inputRawJSON)
+	userAgent := misc.ExtractCodexUserAgent(rawJSON)
 
 	template := `{"model":"","instructions":"","input":[]}`
 
-	_, instructions := misc.CodexInstructionsForModel(modelName, "")
+	_, instructions := misc.CodexInstructionsForModel(modelName, "", userAgent)
 	template, _ = sjson.Set(template, "instructions", instructions)
 
 	rootResult := gjson.ParseBytes(rawJSON)

internal/translator/codex/gemini/codex_gemini_request.go

@@ -38,11 +38,12 @@ import (
 //   - []byte: The transformed request data in Codex API format
 func ConvertGeminiRequestToCodex(modelName string, inputRawJSON []byte, _ bool) []byte {
 	rawJSON := bytes.Clone(inputRawJSON)
+	userAgent := misc.ExtractCodexUserAgent(rawJSON)
 	// Base template
 	out := `{"model":"","instructions":"","input":[]}`
 
 	// Inject standard Codex instructions
-	_, instructions := misc.CodexInstructionsForModel(modelName, "")
+	_, instructions := misc.CodexInstructionsForModel(modelName, "", userAgent)
 	out, _ = sjson.Set(out, "instructions", instructions)
 
 	root := gjson.ParseBytes(rawJSON)

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

@@ -31,6 +31,7 @@ import (
 //   - []byte: The transformed request data in OpenAI Responses API format
 func ConvertOpenAIRequestToCodex(modelName string, inputRawJSON []byte, stream bool) []byte {
 	rawJSON := bytes.Clone(inputRawJSON)
+	userAgent := misc.ExtractCodexUserAgent(rawJSON)
 	// Start with empty JSON object
 	out := `{}`
 
@@ -96,7 +97,7 @@ func ConvertOpenAIRequestToCodex(modelName string, inputRawJSON []byte, stream b
 
 	// Extract system instructions from first system message (string or text object)
 	messages := gjson.GetBytes(rawJSON, "messages")
-	_, instructions := misc.CodexInstructionsForModel(modelName, "")
+	_, instructions := misc.CodexInstructionsForModel(modelName, "", userAgent)
 	out, _ = sjson.Set(out, "instructions", instructions)
 	// if messages.IsArray() {
 	// 	arr := messages.Array()

internal/translator/codex/openai/responses/codex_openai-responses_request.go

@@ -12,6 +12,8 @@ import (
 
 func ConvertOpenAIResponsesRequestToCodex(modelName string, inputRawJSON []byte, _ bool) []byte {
 	rawJSON := bytes.Clone(inputRawJSON)
+	userAgent := misc.ExtractCodexUserAgent(rawJSON)
+	rawJSON = misc.StripCodexUserAgent(rawJSON)
 
 	rawJSON, _ = sjson.SetBytes(rawJSON, "stream", true)
 	rawJSON, _ = sjson.SetBytes(rawJSON, "store", false)
@@ -32,7 +34,7 @@ func ConvertOpenAIResponsesRequestToCodex(modelName string, inputRawJSON []byte,
 		originalInstructionsText = originalInstructionsResult.String()
 	}
 
-	hasOfficialInstructions, instructions := misc.CodexInstructionsForModel(modelName, originalInstructionsResult.String())
+	hasOfficialInstructions, instructions := misc.CodexInstructionsForModel(modelName, originalInstructionsResult.String(), userAgent)
 
 	inputResult := gjson.GetBytes(rawJSON, "input")
 	var inputResults []gjson.Result

internal/translator/codex/openai/responses/codex_openai-responses_response.go

@@ -5,6 +5,7 @@ import (
 	"context"
 	"fmt"
 
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/misc"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
 )
@@ -18,7 +19,10 @@ func ConvertCodexResponseToOpenAIResponses(ctx context.Context, modelName string
 		if typeResult := gjson.GetBytes(rawJSON, "type"); typeResult.Exists() {
 			typeStr := typeResult.String()
 			if typeStr == "response.created" || typeStr == "response.in_progress" || typeStr == "response.completed" {
-				rawJSON, _ = sjson.SetBytes(rawJSON, "response.instructions", gjson.GetBytes(originalRequestRawJSON, "instructions").String())
+				if gjson.GetBytes(rawJSON, "response.instructions").Exists() {
+					instructions := selectInstructions(originalRequestRawJSON, requestRawJSON)
+					rawJSON, _ = sjson.SetBytes(rawJSON, "response.instructions", instructions)
+				}
 			}
 		}
 		out := fmt.Sprintf("data: %s", string(rawJSON))
@@ -37,6 +41,16 @@ func ConvertCodexResponseToOpenAIResponsesNonStream(_ context.Context, modelName
 	}
 	responseResult := rootResult.Get("response")
 	template := responseResult.Raw
-	template, _ = sjson.Set(template, "instructions", gjson.GetBytes(originalRequestRawJSON, "instructions").String())
+	if responseResult.Get("instructions").Exists() {
+		template, _ = sjson.Set(template, "instructions", selectInstructions(originalRequestRawJSON, requestRawJSON))
+	}
 	return template
 }
+
+func selectInstructions(originalRequestRawJSON, requestRawJSON []byte) string {
+	userAgent := misc.ExtractCodexUserAgent(originalRequestRawJSON)
+	if misc.IsOpenCodeUserAgent(userAgent) {
+		return gjson.GetBytes(requestRawJSON, "instructions").String()
+	}
+	return gjson.GetBytes(originalRequestRawJSON, "instructions").String()
+}

sdk/api/handlers/handlers_stream_bootstrap_test.go

@@ -56,6 +56,14 @@ func (e *failOnceStreamExecutor) CountTokens(context.Context, *coreauth.Auth, co
 	return coreexecutor.Response{}, &coreauth.Error{Code: "not_implemented", Message: "CountTokens not implemented"}
 }
 
+func (e *failOnceStreamExecutor) HttpRequest(ctx context.Context, auth *coreauth.Auth, req *http.Request) (*http.Response, error) {
+	return nil, &coreauth.Error{
+		Code:       "not_implemented",
+		Message:    "HttpRequest not implemented",
+		HTTPStatus: http.StatusNotImplemented,
+	}
+}
+
 func (e *failOnceStreamExecutor) Calls() int {
 	e.mu.Lock()
 	defer e.mu.Unlock()

sdk/cliproxy/auth/conductor.go

@@ -1,9 +1,11 @@
 package auth
 
 import (
+	"bytes"
 	"context"
 	"encoding/json"
 	"errors"
+	"io"
 	"net/http"
 	"path/filepath"
 	"strconv"
@@ -32,6 +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.
@@ -1572,6 +1577,23 @@ 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 {
@@ -1609,7 +1631,6 @@ func formatOauthIdentity(auth *Auth, provider string, accountInfo string) string
 	if auth == nil {
 		return ""
 	}
-	authIndex := auth.EnsureIndex()
 	// Prefer the auth's provider when available.
 	providerName := strings.TrimSpace(auth.Provider)
 	if providerName == "" {
@@ -1631,16 +1652,10 @@ func formatOauthIdentity(auth *Auth, provider string, accountInfo string) string
 	if authFile != "" {
 		parts = append(parts, "auth_file="+authFile)
 	}
-	if authIndex != "" {
-		parts = append(parts, "auth_index="+authIndex)
-	}
 	if len(parts) == 0 {
 		return accountInfo
 	}
-	if accountInfo == "" {
-		return strings.Join(parts, " ")
-	}
-	return strings.Join(parts, " ") + " account=" + strconv.Quote(accountInfo)
+	return strings.Join(parts, " ")
 }
 
 // InjectCredentials delegates per-provider HTTP request preparation when supported.
@@ -1654,7 +1669,7 @@ 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 {
@@ -1665,3 +1680,80 @@ 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/service.go

@@ -1237,7 +1237,7 @@ func applyOAuthModelMappings(cfg *config.Config, provider, authKind string, mode
 		fork  bool
 	}
 
-	forward := make(map[string]mappingEntry, len(mappings))
+	forward := make(map[string][]mappingEntry, len(mappings))
 	for i := range mappings {
 		name := strings.TrimSpace(mappings[i].Name)
 		alias := strings.TrimSpace(mappings[i].Alias)
@@ -1248,14 +1248,12 @@ func applyOAuthModelMappings(cfg *config.Config, provider, authKind string, mode
 			continue
 		}
 		key := strings.ToLower(name)
-		if _, exists := forward[key]; exists {
-			continue
-		}
-		forward[key] = mappingEntry{alias: alias, fork: mappings[i].Fork}
+		forward[key] = append(forward[key], mappingEntry{alias: alias, fork: mappings[i].Fork})
 	}
 	if len(forward) == 0 {
 		return models
 	}
+
 	out := make([]*ModelInfo, 0, len(models))
 	seen := make(map[string]struct{}, len(models))
 	for _, model := range models {
@@ -1267,17 +1265,8 @@ func applyOAuthModelMappings(cfg *config.Config, provider, authKind string, mode
 			continue
 		}
 		key := strings.ToLower(id)
-		entry, ok := forward[key]
-		if !ok {
-			if _, exists := seen[key]; exists {
-				continue
-			}
-			seen[key] = struct{}{}
-			out = append(out, model)
-			continue
-		}
-		mappedID := strings.TrimSpace(entry.alias)
-		if mappedID == "" {
+		entries := forward[key]
+		if len(entries) == 0 {
 			if _, exists := seen[key]; exists {
 				continue
 			}
@@ -1286,11 +1275,29 @@ func applyOAuthModelMappings(cfg *config.Config, provider, authKind string, mode
 			continue
 		}
 
-		if entry.fork {
+		keepOriginal := false
+		for _, entry := range entries {
+			if entry.fork {
+				keepOriginal = true
+				break
+			}
+		}
+		if keepOriginal {
 			if _, exists := seen[key]; !exists {
 				seen[key] = struct{}{}
 				out = append(out, model)
 			}
+		}
+
+		addedAlias := false
+		for _, entry := range entries {
+			mappedID := strings.TrimSpace(entry.alias)
+			if mappedID == "" {
+				continue
+			}
+			if strings.EqualFold(mappedID, id) {
+				continue
+			}
 			aliasKey := strings.ToLower(mappedID)
 			if _, exists := seen[aliasKey]; exists {
 				continue
@@ -1302,24 +1309,16 @@ func applyOAuthModelMappings(cfg *config.Config, provider, authKind string, mode
 				clone.Name = rewriteModelInfoName(clone.Name, id, mappedID)
 			}
 			out = append(out, &clone)
-			continue
+			addedAlias = true
 		}
 
-		uniqueKey := strings.ToLower(mappedID)
-		if _, exists := seen[uniqueKey]; exists {
-			continue
-		}
-		seen[uniqueKey] = struct{}{}
-		if mappedID == id {
+		if !keepOriginal && !addedAlias {
+			if _, exists := seen[key]; exists {
+				continue
+			}
+			seen[key] = struct{}{}
 			out = append(out, model)
-			continue
 		}
-		clone := *model
-		clone.ID = mappedID
-		if clone.Name != "" {
-			clone.Name = rewriteModelInfoName(clone.Name, id, mappedID)
-		}
-		out = append(out, &clone)
 	}
 	return out
 }

sdk/cliproxy/service_oauth_model_mappings_test.go

@@ -56,3 +56,37 @@ func TestApplyOAuthModelMappings_ForkAddsAlias(t *testing.T) {
 		t.Fatalf("expected forked model name %q, got %q", "models/g5", out[1].Name)
 	}
 }
+
+func TestApplyOAuthModelMappings_ForkAddsMultipleAliases(t *testing.T) {
+	cfg := &config.Config{
+		OAuthModelMappings: map[string][]config.ModelNameMapping{
+			"codex": {
+				{Name: "gpt-5", Alias: "g5", Fork: true},
+				{Name: "gpt-5", Alias: "g5-2", Fork: true},
+			},
+		},
+	}
+	models := []*ModelInfo{
+		{ID: "gpt-5", Name: "models/gpt-5"},
+	}
+
+	out := applyOAuthModelMappings(cfg, "codex", "oauth", models)
+	if len(out) != 3 {
+		t.Fatalf("expected 3 models, got %d", len(out))
+	}
+	if out[0].ID != "gpt-5" {
+		t.Fatalf("expected first model id %q, got %q", "gpt-5", out[0].ID)
+	}
+	if out[1].ID != "g5" {
+		t.Fatalf("expected second model id %q, got %q", "g5", out[1].ID)
+	}
+	if out[1].Name != "models/g5" {
+		t.Fatalf("expected forked model name %q, got %q", "models/g5", out[1].Name)
+	}
+	if out[2].ID != "g5-2" {
+		t.Fatalf("expected third model id %q, got %q", "g5-2", out[2].ID)
+	}
+	if out[2].Name != "models/g5-2" {
+		t.Fatalf("expected forked model name %q, got %q", "models/g5-2", out[2].Name)
+	}
+}