@kenkaiiii/gg-ai 5.93.0 → 5.94.0

package/dist/index.d.cts

@@ -315,12 +315,6 @@ interface StreamOptions {
      *  budget_tokens < max_tokens), no request is sent and the response has
      *  zero usage with stopReason "end_turn". */
     prewarm?: boolean;
-    /** Override the User-Agent sent with OAuth-authenticated Anthropic requests.
-     *  Anthropic's OAuth edge rejects requests whose claude-cli version lags too
-     *  far behind the real Claude Code release; callers that track the live
-     *  version should pass it here. Ignored for non-Anthropic providers and for
-     *  Anthropic requests using a regular API key. */
-    userAgent?: string;
     /** Extra HTTP headers attached to every model request. Used by providers
      *  whose endpoint gates on client identity (e.g. Kimi For Coding requires a
      *  `User-Agent: kimi-code-cli/...` and `X-Msh-*` device headers). Merged
@@ -1067,24 +1061,10 @@ declare function prewarmAnthropicCache(options: {
     tools?: StreamOptions["tools"];
     serverTools?: StreamOptions["serverTools"];
     baseUrl?: string;
-    userAgent?: string;
     cacheRetention?: StreamOptions["cacheRetention"];
     signal?: AbortSignal;
 }): Promise<void>;
 
-declare const CLAUDE_CODE_FALLBACK_VERSION = "2.1.316";
-/** Test hook: the pending background refresh, if any. */
-declare function pendingClaudeCodeVersionRefresh(): Promise<void> | null;
-/**
- * Resolve the current Claude Code release version for spoofing the claude-cli
- * User-Agent on OAuth and inference requests. Cached in-memory for the process
- * lifetime and on disk for 24h. Falls back to a hardcoded constant if the npm
- * registry is unreachable and no cache exists.
- */
-declare function getClaudeCodeVersion(): Promise<string>;
-/** Build the User-Agent string Anthropic's OAuth + inference edges expect. */
-declare function getClaudeCliUserAgent(): Promise<string>;
-
 interface PalsuProviderState {
     /** `stream()` calls served. */
     callCount: number;
@@ -1171,4 +1151,4 @@ declare function palsuEmbedding(text: string, dimensions?: number): number[];
 /** Deterministic fake relevance: the share of distinct query words found in the document (0–1). */
 declare function palsuRerankScore(query: string, document: string): number;
 
-export { type AssistantMessage, CLAUDE_CODE_FALLBACK_VERSION, CODEX_CLIENT_VERSION, type CacheRetention, type ChatErrorReason, type ContentPart, type DoneEvent, type EmbedInputType, type EmbedOptions, type EmbedResult, type EmbedUsage, type ErrorEvent, type ErrorSource, EventStream, FAST_MODE_SUFFIX, type FastModeRejection, type FastModeResolution, type FormattedChatError, type FormattedError, GGAIError, type ImageContent, LOCAL_NO_API_KEY, type Message, type MessageProvenance, type MessageProvenanceKind, type MessageProvenanceSource, type MessageProvenanceVisibility, type OpenRouterProviderPreferences, PRECONNECT_INTERVAL_MS, type PalsuModelConfig, type PalsuModelHandle, type PalsuProviderConfig, type PalsuProviderHandle, type PalsuProviderState, type PalsuResponse, type PalsuResponseFactory, type PreconnectOptions, type PreparedContext, type Provider, type ProviderCapability, type ProviderDiagnosticFn, type ProviderEmbedFn, type ProviderEmbedRequest, type ProviderEmbedResponse, type ProviderEntry, ProviderError, type ProviderOptions, type ProviderRerankFn, type ProviderRerankRequest, type ProviderRerankResponse, type ProviderStreamFn, REDACTED as REDACTION_MARKER, type RawContent, type RedactionOptions, type RerankOptions, type RerankResponse, type RerankResult, type RerankUsage, ResponseStalledError, type ServerToolCall, type ServerToolCallEvent, type ServerToolDefinition, type ServerToolResult, type ServerToolResultEvent, type StopReason, type StreamEvent, type StreamOptions, type StreamResponse, StreamResult, type SystemMessage, TOOL_CALL_NAME_RULES, type TextContent, type TextDeltaEvent, type ThinkingContent, type ThinkingDeltaEvent, type ThinkingLevel, type Tool, type ToolCall, type ToolCallDeltaEvent, type ToolCallDoneEvent, type ToolCallNameRule, type ToolChoice, type ToolResult, type ToolResultContent, type ToolResultMessage, type Usage, type UserMessage, type VideoContent, clampProviderContextImages, classifyProviderError, continuesTurn, describeFastModeRejection, dropInvalidToolCalls, embed, environmentSecrets, formatChatError, formatError, formatErrorForDisplay, getClaudeCliUserAgent, getClaudeCodeVersion, hasLoneSurrogate, isHardBillingMessage, isUsageLimitError, isValidToolCallName, localBearerKey, localWireModelId, opensTurn, palsuAssistantMessage, palsuEmbedding, palsuRerankScore, palsuText, palsuThinking, palsuToolCall, parseFastModeRejection, pendingClaudeCodeVersionRefresh, preconnectOrigin, preconnectProvider, prewarmAnthropicCache, providerRegistry, redactText, redactValue, registerPalsuProvider, rerank, resolveFastModeModel, resolveToolSchema, sanitizeMessagesForWire, setProviderDiagnostic, sliceHead, sliceTail, stream, toAnthropicMessages, toOpenAIMessages, toWellFormedText, toolCallNameRuleFor, usesResponsesLite };
+export { type AssistantMessage, CODEX_CLIENT_VERSION, type CacheRetention, type ChatErrorReason, type ContentPart, type DoneEvent, type EmbedInputType, type EmbedOptions, type EmbedResult, type EmbedUsage, type ErrorEvent, type ErrorSource, EventStream, FAST_MODE_SUFFIX, type FastModeRejection, type FastModeResolution, type FormattedChatError, type FormattedError, GGAIError, type ImageContent, LOCAL_NO_API_KEY, type Message, type MessageProvenance, type MessageProvenanceKind, type MessageProvenanceSource, type MessageProvenanceVisibility, type OpenRouterProviderPreferences, PRECONNECT_INTERVAL_MS, type PalsuModelConfig, type PalsuModelHandle, type PalsuProviderConfig, type PalsuProviderHandle, type PalsuProviderState, type PalsuResponse, type PalsuResponseFactory, type PreconnectOptions, type PreparedContext, type Provider, type ProviderCapability, type ProviderDiagnosticFn, type ProviderEmbedFn, type ProviderEmbedRequest, type ProviderEmbedResponse, type ProviderEntry, ProviderError, type ProviderOptions, type ProviderRerankFn, type ProviderRerankRequest, type ProviderRerankResponse, type ProviderStreamFn, REDACTED as REDACTION_MARKER, type RawContent, type RedactionOptions, type RerankOptions, type RerankResponse, type RerankResult, type RerankUsage, ResponseStalledError, type ServerToolCall, type ServerToolCallEvent, type ServerToolDefinition, type ServerToolResult, type ServerToolResultEvent, type StopReason, type StreamEvent, type StreamOptions, type StreamResponse, StreamResult, type SystemMessage, TOOL_CALL_NAME_RULES, type TextContent, type TextDeltaEvent, type ThinkingContent, type ThinkingDeltaEvent, type ThinkingLevel, type Tool, type ToolCall, type ToolCallDeltaEvent, type ToolCallDoneEvent, type ToolCallNameRule, type ToolChoice, type ToolResult, type ToolResultContent, type ToolResultMessage, type Usage, type UserMessage, type VideoContent, clampProviderContextImages, classifyProviderError, continuesTurn, describeFastModeRejection, dropInvalidToolCalls, embed, environmentSecrets, formatChatError, formatError, formatErrorForDisplay, hasLoneSurrogate, isHardBillingMessage, isUsageLimitError, isValidToolCallName, localBearerKey, localWireModelId, opensTurn, palsuAssistantMessage, palsuEmbedding, palsuRerankScore, palsuText, palsuThinking, palsuToolCall, parseFastModeRejection, preconnectOrigin, preconnectProvider, prewarmAnthropicCache, providerRegistry, redactText, redactValue, registerPalsuProvider, rerank, resolveFastModeModel, resolveToolSchema, sanitizeMessagesForWire, setProviderDiagnostic, sliceHead, sliceTail, stream, toAnthropicMessages, toOpenAIMessages, toWellFormedText, toolCallNameRuleFor, usesResponsesLite };

package/dist/index.d.ts

@@ -315,12 +315,6 @@ interface StreamOptions {
      *  budget_tokens < max_tokens), no request is sent and the response has
      *  zero usage with stopReason "end_turn". */
     prewarm?: boolean;
-    /** Override the User-Agent sent with OAuth-authenticated Anthropic requests.
-     *  Anthropic's OAuth edge rejects requests whose claude-cli version lags too
-     *  far behind the real Claude Code release; callers that track the live
-     *  version should pass it here. Ignored for non-Anthropic providers and for
-     *  Anthropic requests using a regular API key. */
-    userAgent?: string;
     /** Extra HTTP headers attached to every model request. Used by providers
      *  whose endpoint gates on client identity (e.g. Kimi For Coding requires a
      *  `User-Agent: kimi-code-cli/...` and `X-Msh-*` device headers). Merged
@@ -1067,24 +1061,10 @@ declare function prewarmAnthropicCache(options: {
     tools?: StreamOptions["tools"];
     serverTools?: StreamOptions["serverTools"];
     baseUrl?: string;
-    userAgent?: string;
     cacheRetention?: StreamOptions["cacheRetention"];
     signal?: AbortSignal;
 }): Promise<void>;
 
-declare const CLAUDE_CODE_FALLBACK_VERSION = "2.1.316";
-/** Test hook: the pending background refresh, if any. */
-declare function pendingClaudeCodeVersionRefresh(): Promise<void> | null;
-/**
- * Resolve the current Claude Code release version for spoofing the claude-cli
- * User-Agent on OAuth and inference requests. Cached in-memory for the process
- * lifetime and on disk for 24h. Falls back to a hardcoded constant if the npm
- * registry is unreachable and no cache exists.
- */
-declare function getClaudeCodeVersion(): Promise<string>;
-/** Build the User-Agent string Anthropic's OAuth + inference edges expect. */
-declare function getClaudeCliUserAgent(): Promise<string>;
-
 interface PalsuProviderState {
     /** `stream()` calls served. */
     callCount: number;
@@ -1171,4 +1151,4 @@ declare function palsuEmbedding(text: string, dimensions?: number): number[];
 /** Deterministic fake relevance: the share of distinct query words found in the document (0–1). */
 declare function palsuRerankScore(query: string, document: string): number;
 
-export { type AssistantMessage, CLAUDE_CODE_FALLBACK_VERSION, CODEX_CLIENT_VERSION, type CacheRetention, type ChatErrorReason, type ContentPart, type DoneEvent, type EmbedInputType, type EmbedOptions, type EmbedResult, type EmbedUsage, type ErrorEvent, type ErrorSource, EventStream, FAST_MODE_SUFFIX, type FastModeRejection, type FastModeResolution, type FormattedChatError, type FormattedError, GGAIError, type ImageContent, LOCAL_NO_API_KEY, type Message, type MessageProvenance, type MessageProvenanceKind, type MessageProvenanceSource, type MessageProvenanceVisibility, type OpenRouterProviderPreferences, PRECONNECT_INTERVAL_MS, type PalsuModelConfig, type PalsuModelHandle, type PalsuProviderConfig, type PalsuProviderHandle, type PalsuProviderState, type PalsuResponse, type PalsuResponseFactory, type PreconnectOptions, type PreparedContext, type Provider, type ProviderCapability, type ProviderDiagnosticFn, type ProviderEmbedFn, type ProviderEmbedRequest, type ProviderEmbedResponse, type ProviderEntry, ProviderError, type ProviderOptions, type ProviderRerankFn, type ProviderRerankRequest, type ProviderRerankResponse, type ProviderStreamFn, REDACTED as REDACTION_MARKER, type RawContent, type RedactionOptions, type RerankOptions, type RerankResponse, type RerankResult, type RerankUsage, ResponseStalledError, type ServerToolCall, type ServerToolCallEvent, type ServerToolDefinition, type ServerToolResult, type ServerToolResultEvent, type StopReason, type StreamEvent, type StreamOptions, type StreamResponse, StreamResult, type SystemMessage, TOOL_CALL_NAME_RULES, type TextContent, type TextDeltaEvent, type ThinkingContent, type ThinkingDeltaEvent, type ThinkingLevel, type Tool, type ToolCall, type ToolCallDeltaEvent, type ToolCallDoneEvent, type ToolCallNameRule, type ToolChoice, type ToolResult, type ToolResultContent, type ToolResultMessage, type Usage, type UserMessage, type VideoContent, clampProviderContextImages, classifyProviderError, continuesTurn, describeFastModeRejection, dropInvalidToolCalls, embed, environmentSecrets, formatChatError, formatError, formatErrorForDisplay, getClaudeCliUserAgent, getClaudeCodeVersion, hasLoneSurrogate, isHardBillingMessage, isUsageLimitError, isValidToolCallName, localBearerKey, localWireModelId, opensTurn, palsuAssistantMessage, palsuEmbedding, palsuRerankScore, palsuText, palsuThinking, palsuToolCall, parseFastModeRejection, pendingClaudeCodeVersionRefresh, preconnectOrigin, preconnectProvider, prewarmAnthropicCache, providerRegistry, redactText, redactValue, registerPalsuProvider, rerank, resolveFastModeModel, resolveToolSchema, sanitizeMessagesForWire, setProviderDiagnostic, sliceHead, sliceTail, stream, toAnthropicMessages, toOpenAIMessages, toWellFormedText, toolCallNameRuleFor, usesResponsesLite };
+export { type AssistantMessage, CODEX_CLIENT_VERSION, type CacheRetention, type ChatErrorReason, type ContentPart, type DoneEvent, type EmbedInputType, type EmbedOptions, type EmbedResult, type EmbedUsage, type ErrorEvent, type ErrorSource, EventStream, FAST_MODE_SUFFIX, type FastModeRejection, type FastModeResolution, type FormattedChatError, type FormattedError, GGAIError, type ImageContent, LOCAL_NO_API_KEY, type Message, type MessageProvenance, type MessageProvenanceKind, type MessageProvenanceSource, type MessageProvenanceVisibility, type OpenRouterProviderPreferences, PRECONNECT_INTERVAL_MS, type PalsuModelConfig, type PalsuModelHandle, type PalsuProviderConfig, type PalsuProviderHandle, type PalsuProviderState, type PalsuResponse, type PalsuResponseFactory, type PreconnectOptions, type PreparedContext, type Provider, type ProviderCapability, type ProviderDiagnosticFn, type ProviderEmbedFn, type ProviderEmbedRequest, type ProviderEmbedResponse, type ProviderEntry, ProviderError, type ProviderOptions, type ProviderRerankFn, type ProviderRerankRequest, type ProviderRerankResponse, type ProviderStreamFn, REDACTED as REDACTION_MARKER, type RawContent, type RedactionOptions, type RerankOptions, type RerankResponse, type RerankResult, type RerankUsage, ResponseStalledError, type ServerToolCall, type ServerToolCallEvent, type ServerToolDefinition, type ServerToolResult, type ServerToolResultEvent, type StopReason, type StreamEvent, type StreamOptions, type StreamResponse, StreamResult, type SystemMessage, TOOL_CALL_NAME_RULES, type TextContent, type TextDeltaEvent, type ThinkingContent, type ThinkingDeltaEvent, type ThinkingLevel, type Tool, type ToolCall, type ToolCallDeltaEvent, type ToolCallDoneEvent, type ToolCallNameRule, type ToolChoice, type ToolResult, type ToolResultContent, type ToolResultMessage, type Usage, type UserMessage, type VideoContent, clampProviderContextImages, classifyProviderError, continuesTurn, describeFastModeRejection, dropInvalidToolCalls, embed, environmentSecrets, formatChatError, formatError, formatErrorForDisplay, hasLoneSurrogate, isHardBillingMessage, isUsageLimitError, isValidToolCallName, localBearerKey, localWireModelId, opensTurn, palsuAssistantMessage, palsuEmbedding, palsuRerankScore, palsuText, palsuThinking, palsuToolCall, parseFastModeRejection, preconnectOrigin, preconnectProvider, prewarmAnthropicCache, providerRegistry, redactText, redactValue, registerPalsuProvider, rerank, resolveFastModeModel, resolveToolSchema, sanitizeMessagesForWire, setProviderDiagnostic, sliceHead, sliceTail, stream, toAnthropicMessages, toOpenAIMessages, toWellFormedText, toolCallNameRuleFor, usesResponsesLite };

package/dist/index.js

@@ -2,7 +2,7 @@
 var FAST_MODE_SUFFIX = "-fast";
 var FAST_MODE_BASE_MODELS = {
   anthropic: ["claude-opus-5-5"],
-  openai: ["gpt-6-astra", "gpt-6.1-sol", "gpt-6-luna"]
+  openai: ["gpt-6-astra", "gpt-6.1-sol", "gpt-6.1-terra", "gpt-6-luna"]
 };
 function resolveFastModeModel(provider, model) {
   if (!model.endsWith(FAST_MODE_SUFFIX)) return { model, fast: false };
@@ -608,145 +608,6 @@ var StreamResult = class _StreamResult {
   }
 };
 
-// src/claude-code-version.ts
-import fs from "fs/promises";
-import os from "os";
-import path from "path";
-
-// src/utils/diag.ts
-var _diagFn = null;
-function setProviderDiagnostic(fn) {
-  _diagFn = fn;
-}
-function providerDiag(phase, data) {
-  _diagFn?.(phase, data);
-}
-
-// src/claude-code-version.ts
-var NPM_LATEST_URL = "https://registry.npmjs.org/@anthropic-ai/claude-code/latest";
-var CACHE_TTL_MS = 24 * 60 * 60 * 1e3;
-var FETCH_TIMEOUT_MS = 3e3;
-var CLAUDE_CODE_FALLBACK_VERSION = "2.1.316";
-function isOlder(a, b) {
-  const pa = a.split(".").map((n) => Number.parseInt(n, 10) || 0);
-  const pb = b.split(".").map((n) => Number.parseInt(n, 10) || 0);
-  for (let i = 0; i < Math.max(pa.length, pb.length); i++) {
-    const x = pa[i] ?? 0;
-    const y = pb[i] ?? 0;
-    if (x !== y) return x < y;
-  }
-  return false;
-}
-function atLeastFallback(version) {
-  return isOlder(version, CLAUDE_CODE_FALLBACK_VERSION) ? CLAUDE_CODE_FALLBACK_VERSION : version;
-}
-var memoryCache = null;
-var inflight = null;
-function agentDir() {
-  return path.join(os.homedir(), ".gg");
-}
-function cachePath() {
-  return path.join(agentDir(), "claude-code-version.json");
-}
-async function readDiskCache() {
-  try {
-    const raw = await fs.readFile(cachePath(), "utf-8");
-    const parsed = JSON.parse(raw);
-    if (typeof parsed.version === "string" && typeof parsed.fetchedAt === "number") {
-      return parsed;
-    }
-    return null;
-  } catch {
-    return null;
-  }
-}
-async function writeDiskCache(data) {
-  try {
-    await fs.mkdir(agentDir(), { recursive: true, mode: 448 });
-    await fs.writeFile(cachePath(), JSON.stringify(data), { mode: 384 });
-  } catch (err) {
-    providerDiag("claude_code_version_cache_write_failed", {
-      error: err instanceof Error ? err.message : String(err)
-    });
-  }
-}
-async function fetchLatest() {
-  const controller = new AbortController();
-  const timer = setTimeout(() => controller.abort(), FETCH_TIMEOUT_MS);
-  try {
-    const response = await fetch(NPM_LATEST_URL, { signal: controller.signal });
-    if (!response.ok) return null;
-    const data = await response.json();
-    if (typeof data.version === "string" && /^\d/.test(data.version)) {
-      return data.version;
-    }
-    return null;
-  } catch {
-    return null;
-  } finally {
-    clearTimeout(timer);
-  }
-}
-var refreshing = null;
-function refreshInBackground() {
-  refreshing ??= (async () => {
-    try {
-      const fetched = await fetchLatest();
-      if (!fetched) return;
-      await writeDiskCache({ version: fetched, fetchedAt: Date.now() });
-      memoryCache = { version: fetched, expiresAt: Date.now() + CACHE_TTL_MS };
-    } finally {
-      refreshing = null;
-    }
-  })();
-  return refreshing;
-}
-function pendingClaudeCodeVersionRefresh() {
-  return refreshing;
-}
-async function getClaudeCodeVersion() {
-  if (memoryCache && Date.now() < memoryCache.expiresAt) {
-    return memoryCache.version;
-  }
-  if (inflight) return inflight;
-  inflight = (async () => {
-    const disk = await readDiskCache();
-    const diskFresh = disk && Date.now() - disk.fetchedAt < CACHE_TTL_MS;
-    if (disk && diskFresh) {
-      const version = atLeastFallback(disk.version);
-      memoryCache = { version, expiresAt: Date.now() + CACHE_TTL_MS };
-      return version;
-    }
-    if (disk) {
-      const version = atLeastFallback(disk.version);
-      memoryCache = { version, expiresAt: Date.now() + 5 * 60 * 1e3 };
-      void refreshInBackground();
-      return version;
-    }
-    const fetched = await fetchLatest();
-    if (fetched) {
-      await writeDiskCache({ version: fetched, fetchedAt: Date.now() });
-      memoryCache = { version: fetched, expiresAt: Date.now() + CACHE_TTL_MS };
-      return fetched;
-    }
-    const resolved = CLAUDE_CODE_FALLBACK_VERSION;
-    memoryCache = { version: resolved, expiresAt: Date.now() + 5 * 60 * 1e3 };
-    providerDiag("claude_code_version_fallback", { version: resolved });
-    return resolved;
-  })();
-  try {
-    return await inflight;
-  } finally {
-    inflight = null;
-  }
-}
-function claudeCliUserAgent(version) {
-  return `claude-cli/${version} (external, cli)`;
-}
-async function getClaudeCliUserAgent() {
-  return claudeCliUserAgent(await getClaudeCodeVersion());
-}
-
 // src/utils/provider-options.ts
 var RESERVED_BODY_FIELDS = /* @__PURE__ */ new Set([
   "contents",
@@ -1972,7 +1833,7 @@ function toGlmReasoningEffort(level) {
 }
 function toCodexReasoningEffort(level, model) {
   if (level !== "ultra") return level;
-  return model === "gpt-6-astra" || model === "gpt-6.1-sol" ? "xhigh" : "max";
+  return model === "gpt-6-astra" || model === "gpt-6.1-sol" || model === "gpt-6.1-terra" ? "xhigh" : "max";
 }
 var DECLARED_EFFORT_ORDER = ["low", "medium", "high", "xhigh", "max"];
 function effortForRequest(level, model, declared) {
@@ -2109,8 +1970,7 @@ var anthropicServerFallback = {
   }
 };
 function serverFallbackKey(options) {
-  const auth = options.apiKey?.startsWith("sk-ant-oat") ? "oauth" : "key";
-  return `${options.baseUrl ?? process.env.ANTHROPIC_BASE_URL ?? ""}|${auth}`;
+  return options.baseUrl ?? process.env.ANTHROPIC_BASE_URL ?? "";
 }
 function isDirectAnthropicApi(baseUrl) {
   const effective = baseUrl ?? process.env.ANTHROPIC_BASE_URL;
@@ -2143,30 +2003,18 @@ function sumUsageIterations(usage) {
   }
   return totals;
 }
-async function withClaudeUserAgent(options) {
-  if (options.userAgent || !options.apiKey?.startsWith("sk-ant-oat")) return options;
-  return { ...options, userAgent: await getClaudeCliUserAgent() };
-}
 function createClient(options) {
-  const isOAuth = options.apiKey?.startsWith("sk-ant-oat");
-  const userAgent = isOAuth ? options.userAgent ?? claudeCliUserAgent(CLAUDE_CODE_FALLBACK_VERSION) : "";
-  const cacheKey = `${options.apiKey ?? ""}|${options.baseUrl ?? ""}|${userAgent}`;
+  const cacheKey = `${options.apiKey ?? ""}|${options.baseUrl ?? ""}`;
   if (!options.fetch) {
     const cached = anthropicClientCache.get(cacheKey);
     if (cached) return cached;
   }
   const client = new Anthropic({
-    ...isOAuth ? { apiKey: null, authToken: options.apiKey } : { apiKey: options.apiKey },
+    apiKey: options.apiKey,
     ...options.baseUrl ? { baseURL: options.baseUrl } : {},
     // Default transport lifts undici's 300s header/body caps (see provider-fetch.ts).
     fetch: options.fetch ?? providerFetch,
-    maxRetries: 0,
-    ...isOAuth ? {
-      defaultHeaders: {
-        "user-agent": userAgent,
-        "x-app": "cli"
-      }
-    } : {}
+    maxRetries: 0
   });
   if (!options.fetch) {
     if (anthropicClientCache.size >= 8) {
@@ -2179,13 +2027,10 @@ function createClient(options) {
 }
 async function prewarmAnthropicCache(options) {
   try {
-    const client = createClient(
-      await withClaudeUserAgent({
-        apiKey: options.apiKey,
-        baseUrl: options.baseUrl,
-        userAgent: options.userAgent
-      })
-    );
+    const client = createClient({
+      apiKey: options.apiKey,
+      baseUrl: options.baseUrl
+    });
     const cacheControl = toAnthropicCacheControl(options.cacheRetention ?? "long", options.baseUrl);
     const { system, messages } = toAnthropicMessages(
       [
@@ -2194,21 +2039,13 @@ async function prewarmAnthropicCache(options) {
       ],
       cacheControl
     );
-    const isOAuth = options.apiKey.startsWith("sk-ant-oat");
-    const fullSystem = isOAuth ? [
-      {
-        type: "text",
-        text: "You are Claude Code, Anthropic's official CLI for Claude."
-      },
-      ...system ?? []
-    ] : system;
     const tools = options.tools?.length ? toAnthropicTools(options.tools, {
       cacheControl,
       // Keep the serialized tool bytes identical to runStream so the
       // prewarmed prompt cache actually hits — both are gated by the flag.
       enableFineGrainedToolStreaming: fineGrainedToolStreamingEnabled()
     }) : void 0;
-    if (estimatePrefixTokens(fullSystem, tools) < anthropicMinCacheableTokens(options.model)) {
+    if (estimatePrefixTokens(system, tools) < anthropicMinCacheableTokens(options.model)) {
       return;
     }
     await client.messages.create(
@@ -2216,7 +2053,7 @@ async function prewarmAnthropicCache(options) {
         model: options.model,
         max_tokens: 1,
         messages,
-        ...fullSystem ? { system: fullSystem } : {},
+        ...system ? { system } : {},
         ...tools ? {
           tools: [
             ...tools,
@@ -2227,10 +2064,7 @@ async function prewarmAnthropicCache(options) {
       {
         signal: options.signal ?? void 0,
         ...(() => {
-          const betas = [
-            ...isOAuth ? ["claude-code-20250219", "oauth-2025-04-20"] : [],
-            ...cacheControl?.ttl === "1h" ? ["extended-cache-ttl-2025-04-11"] : []
-          ];
+          const betas = cacheControl?.ttl === "1h" ? ["extended-cache-ttl-2025-04-11"] : [];
           return betas.length ? { headers: { "anthropic-beta": betas.join(",") } } : {};
         })()
       }
@@ -2242,8 +2076,7 @@ function streamAnthropic(options) {
   return new StreamResult(runStream(options), options.signal);
 }
 async function* runStream(options) {
-  const client = createClient(await withClaudeUserAgent(options));
-  const isOAuth = options.apiKey?.startsWith("sk-ant-oat");
+  const client = createClient(options);
   const useStreaming = options.streaming !== false && !options.prewarm;
   const cacheControl = toAnthropicCacheControl(options.cacheRetention, options.baseUrl);
   const supportsFirstPartyToolExtras = !options.baseUrl || options.baseUrl.includes("api.anthropic.com");
@@ -2253,18 +2086,11 @@ async function* runStream(options) {
   const useServerFallback = options.provider !== "minimax" && isDirectAnthropicApi(options.baseUrl) && !anthropicServerFallback.disabled.has(fallbackKey);
   const deferredPlan = options.provider === "anthropic" && isDirectAnthropicApi(options.baseUrl) && supportsAnthropicDeferredTools(options.model) ? planDeferredTools(options.tools, options.deferredTools, downgradedMessages) : void 0;
   const toolReferences = deferredPlan?.loaded;
-  const { system: rawSystem, messages } = toAnthropicMessages(downgradedMessages, cacheControl, {
+  const { system, messages } = toAnthropicMessages(downgradedMessages, cacheControl, {
     fallbackBlocks: useServerFallback,
     toolReferences
   });
   const useFastMode = options.speed === "fast" && options.provider === "anthropic" && isDirectAnthropicApi(options.baseUrl);
-  const system = isOAuth ? [
-    {
-      type: "text",
-      text: "You are Claude Code, Anthropic's official CLI for Claude."
-    },
-    ...rawSystem ?? []
-  ] : rawSystem;
   let maxTokens = options.maxTokens ?? 4096;
   let thinking;
   let outputConfig;
@@ -2340,7 +2166,6 @@ async function* runStream(options) {
   const params = mergeProviderOptions(baseParams, options.providerOptions);
   const hasAdaptiveThinking = isAdaptiveThinkingModel(options.model);
   const betaHeaders = [
-    ...isOAuth ? ["claude-code-20250219", "oauth-2025-04-20"] : [],
     ...options.compaction ? ["compact-2026-01-12"] : [],
     ...options.clearToolUses ? ["context-management-2025-06-27"] : [],
     // Eager tool-input streaming beta — opt-in only (see
@@ -2836,6 +2661,15 @@ function fnv1aHash(value) {
   return (hash >>> 0).toString(16).padStart(8, "0");
 }
 
+// src/utils/diag.ts
+var _diagFn = null;
+function setProviderDiagnostic(fn) {
+  _diagFn = fn;
+}
+function providerDiag(phase, data) {
+  _diagFn?.(phase, data);
+}
+
 // src/providers/moonshot-video.ts
 async function uploadMoonshotVideos(client, messages, signal) {
   for (const msg of messages) {
@@ -3079,11 +2913,11 @@ async function* runStream2(options) {
   }
   const requestParams = mergeProviderOptions(params, options.providerOptions);
   if (getEnvironment()?.GGAI_DUMP_REQUEST) {
-    const fs2 = await import("fs");
+    const fs = await import("fs");
     const ts = (/* @__PURE__ */ new Date()).toISOString().replace(/[:.]/g, "-");
     const dumpPath = `/tmp/ggai-request-${ts}.json`;
-    fs2.writeFileSync(dumpPath, JSON.stringify(requestParams, null, 2));
-    fs2.appendFileSync(
+    fs.writeFileSync(dumpPath, JSON.stringify(requestParams, null, 2));
+    fs.appendFileSync(
       "/tmp/ggai-requests.log",
       `[${ts}] ${dumpPath} messages=${requestParams.messages.length}
 `
@@ -3413,7 +3247,7 @@ function toError2(err, provider = "openai") {
 }
 
 // src/providers/openai-codex.ts
-import os2 from "os";
+import os from "os";
 import * as zstd from "@bokuweb/zstd-wasm";
 
 // src/utils/sse.ts
@@ -4059,7 +3893,7 @@ async function* runStream3(options, retriedWithoutReasoning = false) {
     Authorization: `Bearer ${options.apiKey}`,
     "OpenAI-Beta": "responses=experimental",
     originator: responsesLite ? "codex_cli_rs" : "ggcoder",
-    "User-Agent": responsesLite ? `codex_cli_rs/${CODEX_CLIENT_VERSION}` : `ggcoder (${os2.platform()} ${os2.release()}; ${os2.arch()})`,
+    "User-Agent": responsesLite ? `codex_cli_rs/${CODEX_CLIENT_VERSION}` : `ggcoder (${os.platform()} ${os.release()}; ${os.arch()})`,
     ...responsesLite ? { version: CODEX_CLIENT_VERSION } : {},
     ...liteShape ? { "X-OpenAI-Internal-Codex-Responses-Lite": "true" } : {}
   };
@@ -5199,8 +5033,8 @@ function bound(text) {
   const trimmed = text.trim();
   return trimmed.length > MAX_ERROR_MESSAGE_CHARS ? `${trimmed.slice(0, MAX_ERROR_MESSAGE_CHARS)}\u2026` : trimmed;
 }
-function joinUrl(baseUrl, path2) {
-  return `${baseUrl.replace(/\/+$/, "")}/${path2.replace(/^\/+/, "")}`;
+function joinUrl(baseUrl, path) {
+  return `${baseUrl.replace(/\/+$/, "")}/${path.replace(/^\/+/, "")}`;
 }
 
 // src/providers/gemini-embeddings.ts
@@ -6507,7 +6341,6 @@ function palsuRerankScore(query, document) {
   return hits / queryTokens.size;
 }
 export {
-  CLAUDE_CODE_FALLBACK_VERSION,
   CODEX_CLIENT_VERSION,
   EventStream,
   FAST_MODE_SUFFIX,
@@ -6529,8 +6362,6 @@ export {
   formatChatError,
   formatError,
   formatErrorForDisplay,
-  getClaudeCliUserAgent,
-  getClaudeCodeVersion,
   hasLoneSurrogate,
   isHardBillingMessage,
   isUsageLimitError,
@@ -6545,7 +6376,6 @@ export {
   palsuThinking,
   palsuToolCall,
   parseFastModeRejection,
-  pendingClaudeCodeVersionRefresh,
   preconnectOrigin,
   preconnectProvider,
   prewarmAnthropicCache,