@kenkaiiii/gg-ai 4.3.227 → 4.3.228

package/dist/index.d.cts

@@ -335,6 +335,8 @@ interface FormattedError {
     statusCode?: number;
     /** Provider request ID, kept for telemetry / debug — not shown by default. */
     requestId?: string;
+    /** Unix seconds when a usage/rate limit resets, when the provider reports it. */
+    resetsAt?: number;
 }
 declare class GGAIError extends Error {
     readonly source: ErrorSource;
@@ -350,11 +352,14 @@ declare class GGAIError extends Error {
 declare class ProviderError extends GGAIError {
     readonly provider: string;
     readonly statusCode?: number;
+    /** Unix seconds when a usage/rate limit resets, when the provider reports it. */
+    readonly resetsAt?: number;
     constructor(provider: string, message: string, options?: {
         statusCode?: number;
         requestId?: string;
         hint?: string;
         cause?: unknown;
+        resetsAt?: number;
     });
 }
 /**
@@ -362,6 +367,14 @@ declare class ProviderError extends GGAIError {
  * a non-empty `headline` and `guidance` so the UI never has to second-guess
  * what to show the user.
  */
+/**
+ * Is this a subscription/plan usage-window exhaustion error (as opposed to a
+ * transient per-minute throttle)? These don't clear with a quick retry — the
+ * user has to wait for the window to reset — so callers must surface them as a
+ * hard stop, not silently retry for minutes. Detected from the canonical
+ * "usage limit reached" message gg-ai stamps onto the ProviderError.
+ */
+declare function isUsageLimitError(err: unknown): boolean;
 declare function formatError(err: unknown): FormattedError;
 /**
  * Render a FormattedError as a multi-line string for terminal display.
@@ -452,4 +465,4 @@ interface PalsuProviderConfig {
  */
 declare function registerPalsuProvider(config?: PalsuProviderConfig): PalsuProviderHandle;
 
-export { type AssistantMessage, type CacheRetention, type ContentPart, type DoneEvent, type ErrorEvent, type ErrorSource, EventStream, type FormattedError, GGAIError, type ImageContent, type Message, type PalsuModelConfig, type PalsuModelHandle, type PalsuProviderConfig, type PalsuProviderHandle, type PalsuProviderState, type PalsuResponse, type PalsuResponseFactory, type Provider, type ProviderDiagnosticFn, type ProviderEntry, ProviderError, type ProviderStreamFn, type RawContent, type ServerToolCall, type ServerToolCallEvent, type ServerToolDefinition, type ServerToolResult, type ServerToolResultEvent, type StopReason, type StreamEvent, type StreamOptions, type StreamResponse, StreamResult, type SystemMessage, type TextContent, type TextDeltaEvent, type ThinkingContent, type ThinkingDeltaEvent, type ThinkingLevel, type Tool, type ToolCall, type ToolCallDeltaEvent, type ToolCallDoneEvent, type ToolChoice, type ToolResult, type ToolResultContent, type ToolResultMessage, type Usage, type UserMessage, formatError, formatErrorForDisplay, palsuAssistantMessage, palsuText, palsuThinking, palsuToolCall, providerRegistry, registerPalsuProvider, setProviderDiagnostic, stream };
+export { type AssistantMessage, type CacheRetention, type ContentPart, type DoneEvent, type ErrorEvent, type ErrorSource, EventStream, type FormattedError, GGAIError, type ImageContent, type Message, type PalsuModelConfig, type PalsuModelHandle, type PalsuProviderConfig, type PalsuProviderHandle, type PalsuProviderState, type PalsuResponse, type PalsuResponseFactory, type Provider, type ProviderDiagnosticFn, type ProviderEntry, ProviderError, type ProviderStreamFn, type RawContent, type ServerToolCall, type ServerToolCallEvent, type ServerToolDefinition, type ServerToolResult, type ServerToolResultEvent, type StopReason, type StreamEvent, type StreamOptions, type StreamResponse, StreamResult, type SystemMessage, type TextContent, type TextDeltaEvent, type ThinkingContent, type ThinkingDeltaEvent, type ThinkingLevel, type Tool, type ToolCall, type ToolCallDeltaEvent, type ToolCallDoneEvent, type ToolChoice, type ToolResult, type ToolResultContent, type ToolResultMessage, type Usage, type UserMessage, formatError, formatErrorForDisplay, isUsageLimitError, palsuAssistantMessage, palsuText, palsuThinking, palsuToolCall, providerRegistry, registerPalsuProvider, setProviderDiagnostic, stream };

package/dist/index.d.ts

@@ -335,6 +335,8 @@ interface FormattedError {
     statusCode?: number;
     /** Provider request ID, kept for telemetry / debug — not shown by default. */
     requestId?: string;
+    /** Unix seconds when a usage/rate limit resets, when the provider reports it. */
+    resetsAt?: number;
 }
 declare class GGAIError extends Error {
     readonly source: ErrorSource;
@@ -350,11 +352,14 @@ declare class GGAIError extends Error {
 declare class ProviderError extends GGAIError {
     readonly provider: string;
     readonly statusCode?: number;
+    /** Unix seconds when a usage/rate limit resets, when the provider reports it. */
+    readonly resetsAt?: number;
     constructor(provider: string, message: string, options?: {
         statusCode?: number;
         requestId?: string;
         hint?: string;
         cause?: unknown;
+        resetsAt?: number;
     });
 }
 /**
@@ -362,6 +367,14 @@ declare class ProviderError extends GGAIError {
  * a non-empty `headline` and `guidance` so the UI never has to second-guess
  * what to show the user.
  */
+/**
+ * Is this a subscription/plan usage-window exhaustion error (as opposed to a
+ * transient per-minute throttle)? These don't clear with a quick retry — the
+ * user has to wait for the window to reset — so callers must surface them as a
+ * hard stop, not silently retry for minutes. Detected from the canonical
+ * "usage limit reached" message gg-ai stamps onto the ProviderError.
+ */
+declare function isUsageLimitError(err: unknown): boolean;
 declare function formatError(err: unknown): FormattedError;
 /**
  * Render a FormattedError as a multi-line string for terminal display.
@@ -452,4 +465,4 @@ interface PalsuProviderConfig {
  */
 declare function registerPalsuProvider(config?: PalsuProviderConfig): PalsuProviderHandle;
 
-export { type AssistantMessage, type CacheRetention, type ContentPart, type DoneEvent, type ErrorEvent, type ErrorSource, EventStream, type FormattedError, GGAIError, type ImageContent, type Message, type PalsuModelConfig, type PalsuModelHandle, type PalsuProviderConfig, type PalsuProviderHandle, type PalsuProviderState, type PalsuResponse, type PalsuResponseFactory, type Provider, type ProviderDiagnosticFn, type ProviderEntry, ProviderError, type ProviderStreamFn, type RawContent, type ServerToolCall, type ServerToolCallEvent, type ServerToolDefinition, type ServerToolResult, type ServerToolResultEvent, type StopReason, type StreamEvent, type StreamOptions, type StreamResponse, StreamResult, type SystemMessage, type TextContent, type TextDeltaEvent, type ThinkingContent, type ThinkingDeltaEvent, type ThinkingLevel, type Tool, type ToolCall, type ToolCallDeltaEvent, type ToolCallDoneEvent, type ToolChoice, type ToolResult, type ToolResultContent, type ToolResultMessage, type Usage, type UserMessage, formatError, formatErrorForDisplay, palsuAssistantMessage, palsuText, palsuThinking, palsuToolCall, providerRegistry, registerPalsuProvider, setProviderDiagnostic, stream };
+export { type AssistantMessage, type CacheRetention, type ContentPart, type DoneEvent, type ErrorEvent, type ErrorSource, EventStream, type FormattedError, GGAIError, type ImageContent, type Message, type PalsuModelConfig, type PalsuModelHandle, type PalsuProviderConfig, type PalsuProviderHandle, type PalsuProviderState, type PalsuResponse, type PalsuResponseFactory, type Provider, type ProviderDiagnosticFn, type ProviderEntry, ProviderError, type ProviderStreamFn, type RawContent, type ServerToolCall, type ServerToolCallEvent, type ServerToolDefinition, type ServerToolResult, type ServerToolResultEvent, type StopReason, type StreamEvent, type StreamOptions, type StreamResponse, StreamResult, type SystemMessage, type TextContent, type TextDeltaEvent, type ThinkingContent, type ThinkingDeltaEvent, type ThinkingLevel, type Tool, type ToolCall, type ToolCallDeltaEvent, type ToolCallDoneEvent, type ToolChoice, type ToolResult, type ToolResultContent, type ToolResultMessage, type Usage, type UserMessage, formatError, formatErrorForDisplay, isUsageLimitError, palsuAssistantMessage, palsuText, palsuThinking, palsuToolCall, providerRegistry, registerPalsuProvider, setProviderDiagnostic, stream };

package/dist/index.js

@@ -14,6 +14,8 @@ var GGAIError = class extends Error {
 var ProviderError = class extends GGAIError {
   provider;
   statusCode;
+  /** Unix seconds when a usage/rate limit resets, when the provider reports it. */
+  resetsAt;
   constructor(provider, message, options) {
     super(message, {
       source: "provider",
@@ -24,6 +26,7 @@ var ProviderError = class extends GGAIError {
     this.name = "ProviderError";
     this.provider = provider;
     this.statusCode = options?.statusCode;
+    this.resetsAt = options?.resetsAt;
   }
 };
 var PROVIDER_DISPLAY = {
@@ -44,10 +47,36 @@ var PROVIDER_STATUS_URL = {
 function providerDisplayName(provider) {
   return PROVIDER_DISPLAY[provider] ?? provider;
 }
+function isUsageLimitError(err) {
+  if (!(err instanceof Error)) return false;
+  return /usage limit reached/i.test(err.message);
+}
+function formatResetTime(resetsAt) {
+  const when = new Date(resetsAt * 1e3);
+  const sameDay = when.toDateString() === (/* @__PURE__ */ new Date()).toDateString();
+  return sameDay ? when.toLocaleTimeString(void 0, { hour: "numeric", minute: "2-digit" }) : when.toLocaleString(void 0, {
+    weekday: "short",
+    hour: "numeric",
+    minute: "2-digit"
+  });
+}
 function formatError(err) {
   if (err instanceof ProviderError) {
     const name = providerDisplayName(err.provider);
     const cleanMessage = cleanProviderMessage(err.message);
+    if (isUsageLimitError(err)) {
+      const resetClause = err.resetsAt ? ` It resets at ${formatResetTime(err.resetsAt)}.` : "";
+      return {
+        headline: `${name} usage limit reached.`,
+        source: "provider",
+        message: `Your ${name} usage is finished.${resetClause}`,
+        provider: err.provider,
+        statusCode: err.statusCode,
+        ...err.requestId ? { requestId: err.requestId } : {},
+        ...err.resetsAt ? { resetsAt: err.resetsAt } : {},
+        guidance: "Try again once it's back. Your conversation is preserved."
+      };
+    }
     return {
       headline: `${name} returned an error.`,
       source: "provider",
@@ -327,6 +356,14 @@ function normalizeRootForAnthropic(schema) {
 }
 
 // src/providers/transform.ts
+function isPositionSensitiveThinking(part) {
+  if (part.type === "thinking") return !!part.signature;
+  if (part.type === "raw") {
+    const t = part.data.type;
+    return t === "thinking" || t === "redacted_thinking";
+  }
+  return false;
+}
 var NON_VISION_USER_IMAGE_PLACEHOLDER = "(image omitted: model does not support images)";
 var NON_VISION_TOOL_IMAGE_PLACEHOLDER = "(tool image omitted: model does not support images)";
 function stripImages(content, placeholder) {
@@ -426,14 +463,19 @@ function toAnthropicMessages(messages, cacheControl) {
       continue;
     }
     if (msg.role === "assistant") {
-      const content = typeof msg.content === "string" ? msg.content : msg.content.filter((part) => {
-        if (part.type === "thinking" && !part.signature) return false;
-        if (part.type === "text" && !part.text) return false;
+      const lastThinkingIdx = typeof msg.content === "string" ? -1 : msg.content.reduce(
+        (last, part, idx) => isPositionSensitiveThinking(part) ? idx : last,
+        -1
+      );
+      const content = typeof msg.content === "string" ? msg.content : msg.content.filter((part, idx) => {
+        if (part.type === "thinking" && !part.signature && !part.text) return false;
+        if (part.type === "text" && !part.text && idx > lastThinkingIdx) return false;
         return true;
       }).map((part) => {
         if (part.type === "text") return { type: "text", text: part.text };
-        if (part.type === "thinking")
-          return { type: "thinking", thinking: part.text, signature: part.signature };
+        if (part.type === "thinking") {
+          return part.signature ? { type: "thinking", thinking: part.text, signature: part.signature } : { type: "text", text: part.text };
+        }
         if (part.type === "tool_call")
           return {
             type: "tool_use",
@@ -1138,6 +1180,24 @@ function messageToResponse(message) {
     }
   };
 }
+function readUnifiedRateLimit(headers) {
+  const get = (name) => {
+    if (headers && typeof headers.get === "function") {
+      return headers.get(name);
+    }
+    if (headers && typeof headers === "object") {
+      const rec = headers;
+      const value = rec[name] ?? rec[name.toLowerCase()];
+      return typeof value === "string" ? value : null;
+    }
+    return null;
+  };
+  const status = get("anthropic-ratelimit-unified-status");
+  const resetRaw = get("anthropic-ratelimit-unified-reset") ?? get("anthropic-ratelimit-unified-5h-reset") ?? get("anthropic-ratelimit-unified-7d-reset");
+  const resetNum = resetRaw != null ? Number(resetRaw) : Number.NaN;
+  const resetsAt = Number.isFinite(resetNum) && resetNum > 0 ? resetNum : void 0;
+  return { rejected: status === "rejected", ...resetsAt ? { resetsAt } : {} };
+}
 function toError(err) {
   if (err instanceof Anthropic.APIError) {
     const errorBody = err.error;
@@ -1146,6 +1206,18 @@ function toError(err) {
     const bodyMessage = typeof nestedError?.message === "string" ? nestedError.message : typeof errorBody?.message === "string" ? errorBody.message : void 0;
     const bodyType = typeof nestedError?.type === "string" ? nestedError.type : typeof errorBody?.type === "string" ? errorBody.type : typeof err.type === "string" ? err.type : void 0;
     const message = bodyType && bodyMessage ? `${bodyType}: ${bodyMessage}` : bodyMessage ?? err.message;
+    if (err.status === 429) {
+      const limit = readUnifiedRateLimit(err.headers);
+      const farOff = limit.resetsAt != null && limit.resetsAt * 1e3 - Date.now() > 6e4;
+      if (limit.rejected || farOff) {
+        return new ProviderError("anthropic", "Claude usage limit reached", {
+          statusCode: 429,
+          ...requestId ? { requestId } : {},
+          ...limit.resetsAt ? { resetsAt: limit.resetsAt } : {},
+          cause: err
+        });
+      }
+    }
     return new ProviderError("anthropic", message, {
       statusCode: err.status,
       ...requestId ? { requestId } : {},
@@ -1583,6 +1655,8 @@ async function* runStream3(options) {
     const parsed = parseCodexErrorBody(text);
     const message = parsed.message ?? `Codex API returned HTTP ${response.status}.`;
     const requestId = parsed.requestId ?? response.headers.get("x-request-id") ?? response.headers.get("openai-request-id") ?? response.headers.get("x-oai-request-id") ?? void 0;
+    const usageLimit = codexUsageLimitError(parsed.errorObj, response.status, requestId);
+    if (usageLimit) throw usageLimit;
     let hint;
     if (response.status === 400 && text.includes("not supported")) {
       if (options.model === "gpt-5.5-pro") {
@@ -1625,6 +1699,12 @@ async function* runStream3(options) {
       const message = nested?.message ?? event.message ?? "Codex stream emitted an error chunk without a message.";
       const code = nested?.code ?? nested?.type ?? event.code ?? "server_error";
       const requestId = extractCodexRequestId(message) ?? event.request_id;
+      const usageLimit = codexUsageLimitError(
+        nested ?? event,
+        void 0,
+        requestId
+      );
+      if (usageLimit) throw usageLimit;
       throw new ProviderError("openai", message, {
         ...requestId != null ? { requestId } : {},
         ...code === "server_error" ? { statusCode: 500 } : {}
@@ -1947,12 +2027,34 @@ function parseCodexErrorBody(text) {
     const detail = parsed.detail;
     const message = error?.message ?? parsed.message ?? (typeof detail === "string" ? detail : void 0);
     const requestId = parsed.request_id ?? error?.request_id ?? (message ? extractCodexRequestId(message) : void 0);
-    return { ...message ? { message } : {}, ...requestId ? { requestId } : {} };
+    const errorObj = error ?? parsed;
+    return {
+      ...message ? { message } : {},
+      ...requestId ? { requestId } : {},
+      ...errorObj ? { errorObj } : {}
+    };
   } catch {
     const trimmed = text.trim().slice(0, 240);
     return trimmed ? { message: trimmed } : {};
   }
 }
+var CODEX_USAGE_LIMIT_CODE = /usage_limit_reached|usage_not_included/i;
+var CODEX_RATE_LIMIT_CODE = /rate_limit_exceeded/i;
+function codexUsageLimitError(errorObj, statusCode, requestId) {
+  const code = String(errorObj?.code ?? errorObj?.type ?? "");
+  const rateLimits = errorObj?.rate_limits;
+  const resetsAtRaw = (typeof errorObj?.resets_at === "number" ? errorObj.resets_at : void 0) ?? rateLimits?.primary?.resets_at ?? rateLimits?.secondary?.resets_at;
+  const resetsInSeconds = typeof errorObj?.resets_in_seconds === "number" ? errorObj.resets_in_seconds : void 0;
+  const resetsAt = typeof resetsAtRaw === "number" && resetsAtRaw > 0 ? resetsAtRaw : resetsInSeconds != null && resetsInSeconds > 0 ? Math.floor(Date.now() / 1e3) + resetsInSeconds : void 0;
+  const isHardUsage = CODEX_USAGE_LIMIT_CODE.test(code);
+  const isRateOr429 = CODEX_RATE_LIMIT_CODE.test(code) || statusCode === 429;
+  if (!isHardUsage && !(isRateOr429 && resetsAt != null)) return null;
+  return new ProviderError("openai", "ChatGPT usage limit reached", {
+    statusCode: statusCode ?? 429,
+    ...requestId ? { requestId } : {},
+    ...resetsAt ? { resetsAt } : {}
+  });
+}
 
 // src/providers/gemini.ts
 var DEFAULT_CODE_ASSIST_BASE_URL = "https://cloudcode-pa.googleapis.com";
@@ -2715,6 +2817,7 @@ export {
   StreamResult,
   formatError,
   formatErrorForDisplay,
+  isUsageLimitError,
   palsuAssistantMessage,
   palsuText,
   palsuThinking,