@kenkaiiii/gg-ai 4.3.237 → 4.3.238

package/dist/index.js

@@ -1,4 +1,18 @@
 // src/errors.ts
+function readHeader(headers, ...names) {
+  if (!headers) return void 0;
+  const getter = typeof headers.get === "function" ? (name) => headers.get(name) ?? void 0 : typeof headers === "object" ? (name) => {
+    const rec = headers;
+    const value = rec[name] ?? rec[name.toLowerCase()];
+    return typeof value === "string" ? value : void 0;
+  } : void 0;
+  if (!getter) return void 0;
+  for (const name of names) {
+    const value = getter(name);
+    if (value != null) return value;
+  }
+  return void 0;
+}
 var GGAIError = class extends Error {
   source;
   requestId;
@@ -51,6 +65,10 @@ function isUsageLimitError(err) {
   if (!(err instanceof Error)) return false;
   return /usage limit reached/i.test(err.message);
 }
+function isHardBillingMessage(message) {
+  const lower = message.toLowerCase();
+  return lower.includes("insufficient balance") || lower.includes("insufficient credits") || lower.includes("more credits") || lower.includes("insufficient_quota") || lower.includes("exceeded your current quota") || lower.includes("quota exceeded") || lower.includes("no resource package") || lower.includes("recharge") || lower.includes("balance is too low") || lower.includes("out of credits") || lower.includes("arrears") || lower.includes("arrearage") || lower.includes("token quota") || lower.includes("exceeded_current_quota_error") || lower.includes("check your account balance") || lower.includes("does not yet include access") || lower.includes("subscription plan") || lower.includes("billing");
+}
 function formatResetTime(resetsAt) {
   const when = new Date(resetsAt * 1e3);
   const sameDay = when.toDateString() === (/* @__PURE__ */ new Date()).toDateString();
@@ -300,6 +318,9 @@ function zodToJsonSchema(schema) {
   const { $schema: _schema, ...rest } = jsonSchema;
   return normalizeRootForAnthropic(rest);
 }
+function resolveToolSchema(tool) {
+  return tool.rawInputSchema ?? zodToJsonSchema(tool.parameters);
+}
 function normalizeRootForAnthropic(schema) {
   const branches = schema.oneOf ?? schema.anyOf;
   if (!branches || branches.length === 0) {
@@ -413,6 +434,7 @@ function toAnthropicAssistantContent(content, isLatest, idMap) {
 }
 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)";
+var NON_VIDEO_USER_PLACEHOLDER = "(video omitted: model does not support video)";
 function stripImages(content, placeholder) {
   const out = [];
   let lastWasPlaceholder = false;
@@ -423,10 +445,33 @@ function stripImages(content, placeholder) {
       continue;
     }
     out.push(block);
-    lastWasPlaceholder = block.text === placeholder;
+    lastWasPlaceholder = block.type === "text" && block.text === placeholder;
+  }
+  return out;
+}
+function stripVideos(content, placeholder) {
+  const out = [];
+  let lastWasPlaceholder = false;
+  for (const block of content) {
+    if (block.type === "video") {
+      if (!lastWasPlaceholder) out.push({ type: "text", text: placeholder });
+      lastWasPlaceholder = true;
+      continue;
+    }
+    out.push(block);
+    lastWasPlaceholder = block.type === "text" && block.text === placeholder;
   }
   return out;
 }
+function downgradeUnsupportedVideos(messages, supportsVideo) {
+  if (supportsVideo === true) return messages;
+  return messages.map((msg) => {
+    if (msg.role === "user" && Array.isArray(msg.content)) {
+      return { ...msg, content: stripVideos(msg.content, NON_VIDEO_USER_PLACEHOLDER) };
+    }
+    return msg;
+  });
+}
 function downgradeUnsupportedImages(messages, supportsImages) {
   if (supportsImages !== false) return messages;
   return messages.map((msg) => {
@@ -503,6 +548,16 @@ function toAnthropicMessages(messages, cacheControl) {
         role: "user",
         content: typeof msg.content === "string" ? msg.content : msg.content.map((part) => {
           if (part.type === "text") return { type: "text", text: part.text };
+          if (part.type === "video") {
+            return {
+              type: "video",
+              source: {
+                type: "base64",
+                media_type: part.mediaType,
+                data: part.data
+              }
+            };
+          }
           return {
             type: "image",
             source: {
@@ -602,11 +657,11 @@ function toAnthropicToolChoice(choice) {
   if (choice === "required") return { type: "any" };
   return { type: "tool", name: choice.name };
 }
-function supportsAdaptiveThinking(model) {
-  return /opus-4-8|opus-4-7|opus-4-6|sonnet-4-6/.test(model);
+function isAdaptiveThinkingModel(model) {
+  return /opus-4[-.]8|opus-4[-.]7|opus-4[-.]6|sonnet-4[-.]6/.test(model);
 }
 function toAnthropicThinking(level, maxTokens, model) {
-  if (supportsAdaptiveThinking(model)) {
+  if (isAdaptiveThinkingModel(model)) {
     let effort = level;
     if (effort === "xhigh" && !/opus-4-8|opus-4-7/.test(model)) {
       effort = "high";
@@ -663,6 +718,14 @@ function toOpenAIMessages(messages, options) {
           content: msg.content.map(
             (part) => {
               if (part.type === "text") return { type: "text", text: part.text };
+              if (part.type === "video") {
+                return {
+                  type: "video_url",
+                  video_url: {
+                    url: `data:${part.mediaType};base64,${part.data}`
+                  }
+                };
+              }
               return {
                 type: "image_url",
                 image_url: {
@@ -740,7 +803,7 @@ function toOpenAITools(tools) {
     function: {
       name: tool.name,
       description: tool.description,
-      parameters: tool.rawInputSchema ?? zodToJsonSchema(tool.parameters)
+      parameters: resolveToolSchema(tool)
     }
   }));
 }
@@ -782,10 +845,22 @@ function normalizeOpenAIStopReason(reason) {
   }
 }
 
-// src/providers/anthropic.ts
+// src/utils/json.ts
 function isJsonObject(value) {
   return value != null && typeof value === "object" && !Array.isArray(value);
 }
+function parseToolArguments(argsJson) {
+  if (!argsJson) return {};
+  try {
+    const parsed = JSON.parse(argsJson);
+    const unwrapped = typeof parsed === "string" ? JSON.parse(parsed) : parsed;
+    return isJsonObject(unwrapped) ? unwrapped : {};
+  } catch {
+    return {};
+  }
+}
+
+// src/providers/anthropic.ts
 function createClient(options) {
   const isOAuth = options.apiKey?.startsWith("sk-ant-oat");
   return new Anthropic({
@@ -816,7 +891,8 @@ async function* runStream(options) {
   const useStreaming = options.streaming !== false;
   const cacheControl = toAnthropicCacheControl(options.cacheRetention, options.baseUrl);
   const supportsFirstPartyToolExtras = !options.baseUrl || options.baseUrl.includes("api.anthropic.com");
-  const downgradedMessages = downgradeUnsupportedImages(options.messages, options.supportsImages);
+  const downgradedImages = downgradeUnsupportedImages(options.messages, options.supportsImages);
+  const downgradedMessages = downgradeUnsupportedVideos(downgradedImages, options.supportsVideo);
   const { system: rawSystem, messages } = toAnthropicMessages(downgradedMessages, cacheControl);
   const system = isOAuth ? [
     {
@@ -878,7 +954,7 @@ async function* runStream(options) {
     })(),
     stream: useStreaming
   };
-  const hasAdaptiveThinking = options.model.includes("opus-4-8") || options.model.includes("opus-4.8") || options.model.includes("opus-4-7") || options.model.includes("opus-4.7") || options.model.includes("opus-4-6") || options.model.includes("opus-4.6") || options.model.includes("sonnet-4-6") || options.model.includes("sonnet-4.6");
+  const hasAdaptiveThinking = isAdaptiveThinkingModel(options.model);
   const betaHeaders = [
     ...isOAuth ? ["claude-code-20250219", "oauth-2025-04-20"] : [],
     ...options.compaction ? ["compact-2026-01-12"] : [],
@@ -1203,19 +1279,13 @@ 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 status = readHeader(headers, "anthropic-ratelimit-unified-status");
+  const resetRaw = readHeader(
+    headers,
+    "anthropic-ratelimit-unified-reset",
+    "anthropic-ratelimit-unified-5h-reset",
+    "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 } : {} };
@@ -1240,6 +1310,14 @@ function toError(err) {
         });
       }
     }
+    if (isHardBillingMessage(message)) {
+      const usageMessage = /usage limit reached/i.test(message) ? message : `usage limit reached: ${message}`;
+      return new ProviderError("anthropic", usageMessage, {
+        statusCode: err.status,
+        ...requestId ? { requestId } : {},
+        cause: err
+      });
+    }
     return new ProviderError("anthropic", message, {
       statusCode: err.status,
       ...requestId ? { requestId } : {},
@@ -1272,19 +1350,30 @@ function fnv1aHash(value) {
   return (hash >>> 0).toString(16).padStart(8, "0");
 }
 
-// src/providers/openai.ts
-function isJsonObject2(value) {
-  return value != null && typeof value === "object" && !Array.isArray(value);
+// src/utils/env.ts
+function getEnvironment() {
+  return globalThis.process?.env;
 }
-function parseToolArguments(argsJson) {
-  if (!argsJson) return {};
-  try {
-    const parsed = JSON.parse(argsJson);
-    const unwrapped = typeof parsed === "string" ? JSON.parse(parsed) : parsed;
-    return isJsonObject2(unwrapped) ? unwrapped : {};
-  } catch {
-    return {};
+
+// src/providers/openai.ts
+function extractOpenAIUsage(usage) {
+  let cacheRead = 0;
+  const details = usage.prompt_tokens_details;
+  if (details?.cached_tokens) {
+    cacheRead = details.cached_tokens;
+  }
+  const usageAny = usage;
+  if (!cacheRead && typeof usageAny.cached_tokens === "number" && usageAny.cached_tokens > 0) {
+    cacheRead = usageAny.cached_tokens;
   }
+  if (!cacheRead && typeof usageAny.prompt_cache_hit_tokens === "number" && usageAny.prompt_cache_hit_tokens > 0) {
+    cacheRead = usageAny.prompt_cache_hit_tokens;
+  }
+  return {
+    inputTokens: usage.prompt_tokens - cacheRead,
+    outputTokens: usage.completion_tokens,
+    cacheRead
+  };
 }
 function createClient2(options) {
   return new OpenAI({
@@ -1301,7 +1390,8 @@ async function* runStream2(options) {
   const useStreaming = options.streaming !== false;
   const client = createClient2(options);
   const usesThinkingParam = options.provider === "glm" || options.provider === "moonshot" || options.provider === "xiaomi";
-  const downgradedMessages = downgradeUnsupportedImages(options.messages, options.supportsImages);
+  const downgradedImages = downgradeUnsupportedImages(options.messages, options.supportsImages);
+  const downgradedMessages = downgradeUnsupportedVideos(downgradedImages, options.supportsVideo);
   const messages = toOpenAIMessages(downgradedMessages, {
     provider: options.provider,
     thinking: !!options.thinking,
@@ -1340,7 +1430,7 @@ async function* runStream2(options) {
       params.thinking = { type: "disabled" };
     }
   }
-  if (globalThis.process && globalThis.process.env?.GGAI_DUMP_REQUEST) {
+  if (getEnvironment()?.GGAI_DUMP_REQUEST) {
     const fs = await import("fs");
     const ts = (/* @__PURE__ */ new Date()).toISOString().replace(/[:.]/g, "-");
     const dumpPath = `/tmp/ggai-request-${ts}.json`;
@@ -1381,19 +1471,7 @@ async function* runStream2(options) {
   for await (const chunk of stream2) {
     const choice = chunk.choices?.[0];
     if (chunk.usage) {
-      outputTokens = chunk.usage.completion_tokens;
-      const details = chunk.usage.prompt_tokens_details;
-      if (details?.cached_tokens) {
-        cacheRead = details.cached_tokens;
-      }
-      const usageAny = chunk.usage;
-      if (!cacheRead && typeof usageAny.cached_tokens === "number" && usageAny.cached_tokens > 0) {
-        cacheRead = usageAny.cached_tokens;
-      }
-      if (!cacheRead && typeof usageAny.prompt_cache_hit_tokens === "number" && usageAny.prompt_cache_hit_tokens > 0) {
-        cacheRead = usageAny.prompt_cache_hit_tokens;
-      }
-      inputTokens = chunk.usage.prompt_tokens - cacheRead;
+      ({ inputTokens, outputTokens, cacheRead } = extractOpenAIUsage(chunk.usage));
     }
     if (!choice) continue;
     if (choice.finish_reason) {
@@ -1539,17 +1617,7 @@ function completionToResponse(completion) {
   let outputTokens = 0;
   let cacheRead = 0;
   if (completion.usage) {
-    outputTokens = completion.usage.completion_tokens;
-    const details = completion.usage.prompt_tokens_details;
-    if (details?.cached_tokens) cacheRead = details.cached_tokens;
-    const usageAny = completion.usage;
-    if (!cacheRead && typeof usageAny.cached_tokens === "number" && usageAny.cached_tokens > 0) {
-      cacheRead = usageAny.cached_tokens;
-    }
-    if (!cacheRead && typeof usageAny.prompt_cache_hit_tokens === "number" && usageAny.prompt_cache_hit_tokens > 0) {
-      cacheRead = usageAny.prompt_cache_hit_tokens;
-    }
-    inputTokens = completion.usage.prompt_tokens - cacheRead;
+    ({ inputTokens, outputTokens, cacheRead } = extractOpenAIUsage(completion.usage));
   }
   const stopReason = normalizeOpenAIStopReason(choice?.finish_reason ?? null);
   return {
@@ -1561,6 +1629,16 @@ function completionToResponse(completion) {
     usage: { inputTokens, outputTokens, ...cacheRead > 0 && { cacheRead } }
   };
 }
+function classifyOpenAICompatLimit(args) {
+  const { status, code, type, message } = args;
+  const codeType = `${code ?? ""} ${type ?? ""}`.toLowerCase();
+  const isHard = status === 402 || codeType.includes("insufficient_quota") || isHardBillingMessage(message);
+  if (isHard) return "hard";
+  if (status === 429 || codeType.includes("rate_limit_exceeded") || codeType.includes("too_many_requests")) {
+    return "transient";
+  }
+  return null;
+}
 function toError2(err, provider = "openai") {
   if (err instanceof OpenAI.APIError) {
     const body = err.error;
@@ -1572,6 +1650,35 @@ function toError2(err, provider = "openai") {
       hint = "codex-mini-latest requires an OpenAI Pro or Max subscription. Your account currently has access to GPT-5.4 and GPT-5.4 Mini.";
     }
     const requestId = err.request_id ?? (typeof body?.request_id === "string" ? body.request_id : void 0);
+    const code = typeof err.code === "string" ? err.code : void 0;
+    const type = typeof err.type === "string" ? err.type : void 0;
+    const limit = classifyOpenAICompatLimit({
+      status: err.status,
+      code,
+      type,
+      message: cleanMessage
+    });
+    if (limit === "hard") {
+      const message = /usage limit reached/i.test(cleanMessage) ? cleanMessage : `usage limit reached: ${cleanMessage}`;
+      return new ProviderError(provider, message, {
+        statusCode: err.status,
+        ...requestId ? { requestId } : {},
+        ...hint ? { hint } : {},
+        cause: err
+      });
+    }
+    if (limit === "transient") {
+      const retryAfterRaw = readHeader(err.headers, "retry-after");
+      const retryAfterSec = retryAfterRaw != null ? Number(retryAfterRaw) : Number.NaN;
+      const resetsAt = Number.isFinite(retryAfterSec) && retryAfterSec > 0 ? Math.floor(Date.now() / 1e3) + retryAfterSec : void 0;
+      return new ProviderError(provider, cleanMessage, {
+        statusCode: err.status,
+        ...requestId ? { requestId } : {},
+        ...hint ? { hint } : {},
+        ...resetsAt ? { resetsAt } : {},
+        cause: err
+      });
+    }
     return new ProviderError(provider, cleanMessage, {
       statusCode: err.status,
       ...requestId ? { requestId } : {},
@@ -1597,21 +1704,59 @@ function providerDiag(phase, data) {
   _diagFn?.(phase, data);
 }
 
-// src/providers/openai-codex.ts
-var DEFAULT_BASE_URL = "https://chatgpt.com/backend-api";
-function isJsonObject3(value) {
-  return value != null && typeof value === "object" && !Array.isArray(value);
+// src/utils/sse.ts
+function parseSseBuffer(buffer) {
+  const events = [];
+  let cursor = 0;
+  while (true) {
+    const next = buffer.indexOf("\n\n", cursor);
+    if (next === -1) break;
+    const raw = buffer.slice(cursor, next);
+    cursor = next + 2;
+    let eventName;
+    const dataLines = [];
+    for (const line of raw.split("\n")) {
+      if (line.startsWith("event:")) {
+        eventName = line.slice("event:".length).trim();
+      } else if (line.startsWith("data:")) {
+        dataLines.push(line.slice("data:".length).trimStart());
+      }
+    }
+    if (dataLines.length > 0) {
+      events.push({ event: eventName, data: dataLines.join("\n") });
+    }
+  }
+  return { events, remaining: buffer.slice(cursor) };
 }
-function parseToolArguments2(argsJson) {
-  if (!argsJson) return {};
+async function* readSseStream(body) {
+  const reader = body.getReader();
+  const decoder = new TextDecoder();
+  let buffer = "";
   try {
-    const parsed = JSON.parse(argsJson);
-    const unwrapped = typeof parsed === "string" ? JSON.parse(parsed) : parsed;
-    return isJsonObject3(unwrapped) ? unwrapped : {};
-  } catch {
-    return {};
+    while (true) {
+      const { done, value } = await reader.read();
+      if (done) break;
+      buffer += decoder.decode(value, { stream: true }).replace(/\r\n/g, "\n");
+      const parsed2 = parseSseBuffer(buffer);
+      buffer = parsed2.remaining;
+      yield* parsed2.events;
+    }
+    buffer += decoder.decode().replace(/\r\n/g, "\n");
+    const parsed = parseSseBuffer(buffer + "\n\n");
+    yield* parsed.events;
+  } finally {
+    reader.releaseLock();
   }
 }
+
+// src/utils/request-id.ts
+function extractRequestIdFromMessage(message) {
+  const match = message.match(/request ID ([a-z0-9-]{8,})/i);
+  return match?.[1];
+}
+
+// src/providers/openai-codex.ts
+var DEFAULT_BASE_URL = "https://chatgpt.com/backend-api";
 function outputTextKey(itemId, contentIndex) {
   return `${itemId ?? ""}:${contentIndex ?? 0}`;
 }
@@ -1624,7 +1769,8 @@ function streamOpenAICodex(options) {
 async function* runStream3(options) {
   const baseUrl = (options.baseUrl || DEFAULT_BASE_URL).replace(/\/+$/, "");
   const url = `${baseUrl}/codex/responses`;
-  const downgraded = downgradeUnsupportedImages(options.messages, options.supportsImages);
+  const downgradedImages = downgradeUnsupportedImages(options.messages, options.supportsImages);
+  const downgraded = downgradeUnsupportedVideos(downgradedImages, options.supportsVideo);
   const { system, input } = toCodexInput(downgraded, { supportsImages: options.supportsImages });
   const body = {
     model: options.model,
@@ -1676,7 +1822,7 @@ async function* runStream3(options) {
     const text = await response.text().catch(() => "");
     const parsed = parseCodexErrorBody(text);
     const message = parsed.message ?? `Codex API returned HTTP ${response.status}.`;
-    const requestId = parsed.requestId ?? response.headers.get("x-request-id") ?? response.headers.get("openai-request-id") ?? response.headers.get("x-oai-request-id") ?? void 0;
+    const requestId = parsed.requestId ?? readHeader(response.headers, "x-request-id", "openai-request-id", "x-oai-request-id");
     const usageLimit = codexUsageLimitError(parsed.errorObj, response.status, requestId);
     if (usageLimit) throw usageLimit;
     let hint;
@@ -1720,7 +1866,7 @@ async function* runStream3(options) {
       const nested = event.error ?? void 0;
       const message = nested?.message ?? event.message ?? "Codex stream emitted an error chunk without a message.";
       const code = nested?.code ?? nested?.type ?? event.code ?? "server_error";
-      const requestId = extractCodexRequestId(message) ?? event.request_id;
+      const requestId = extractRequestIdFromMessage(message) ?? event.request_id;
       const usageLimit = codexUsageLimitError(
         nested ?? event,
         void 0,
@@ -1735,7 +1881,7 @@ async function* runStream3(options) {
     if (type === "response.failed") {
       const nested = event.error;
       const message = nested?.message ?? "Codex response failed.";
-      const requestId = extractCodexRequestId(message) ?? event.request_id;
+      const requestId = extractRequestIdFromMessage(message) ?? event.request_id;
       throw new ProviderError("openai", message, {
         ...requestId != null ? { requestId } : {}
       });
@@ -1854,7 +2000,7 @@ async function* runStream3(options) {
         const id = `${callId}|${itemId}`;
         const tc = toolCalls.get(id);
         if (tc) {
-          const args = parseToolArguments2(tc.argsJson);
+          const args = parseToolArguments(tc.argsJson);
           yield {
             type: "toolcall_done",
             id: tc.id,
@@ -1878,7 +2024,7 @@ async function* runStream3(options) {
     contentParts.push({ type: "text", text: textAccum });
   }
   for (const [, tc] of toolCalls) {
-    const args = parseToolArguments2(tc.argsJson);
+    const args = parseToolArguments(tc.argsJson);
     const toolCall = {
       type: "tool_call",
       id: tc.id,
@@ -1901,33 +2047,13 @@ async function* runStream3(options) {
   return streamResponse;
 }
 async function* parseSSE(body) {
-  const reader = body.getReader();
-  const decoder = new TextDecoder();
-  let buffer = "";
-  try {
-    while (true) {
-      const { done, value } = await reader.read();
-      if (done) break;
-      buffer += decoder.decode(value, { stream: true });
-      let idx = buffer.indexOf("\n\n");
-      while (idx !== -1) {
-        const chunk = buffer.slice(0, idx);
-        buffer = buffer.slice(idx + 2);
-        const dataLines = chunk.split("\n").filter((l) => l.startsWith("data:")).map((l) => l.slice(5).trim());
-        if (dataLines.length > 0) {
-          const data = dataLines.join("\n").trim();
-          if (data && data !== "[DONE]") {
-            try {
-              yield JSON.parse(data);
-            } catch {
-            }
-          }
-        }
-        idx = buffer.indexOf("\n\n");
-      }
+  for await (const event of readSseStream(body)) {
+    const data = event.data.trim();
+    if (!data || data === "[DONE]") continue;
+    try {
+      yield JSON.parse(data);
+    } catch {
     }
-  } finally {
-    reader.releaseLock();
   }
 }
 function remapCodexId(id, idMap) {
@@ -1938,10 +2064,6 @@ function remapCodexId(id, idMap) {
   idMap.set(id, mapped);
   return mapped;
 }
-function codexToolResultText(content) {
-  if (typeof content === "string") return content;
-  return content.filter((b) => b.type === "text").map((b) => b.text).join("\n");
-}
 function toCodexInput(messages, options) {
   let system;
   const input = [];
@@ -1998,7 +2120,7 @@ function toCodexInput(messages, options) {
       const toolImages = [];
       for (const result of msg.content) {
         const [callId] = result.toolCallId.includes("|") ? result.toolCallId.split("|", 2) : [result.toolCallId];
-        const text = codexToolResultText(result.content);
+        const text = toolResultText(result.content);
         input.push({
           type: "function_call_output",
           call_id: remapCodexId(callId, idMap),
@@ -2033,14 +2155,10 @@ function toCodexTools(tools) {
     type: "function",
     name: tool.name,
     description: tool.description,
-    parameters: tool.rawInputSchema ?? zodToJsonSchema(tool.parameters),
+    parameters: resolveToolSchema(tool),
     strict: null
   }));
 }
-function extractCodexRequestId(message) {
-  const match = message.match(/request ID ([a-z0-9-]{8,})/i);
-  return match?.[1];
-}
 function parseCodexErrorBody(text) {
   if (!text) return {};
   try {
@@ -2048,7 +2166,7 @@ function parseCodexErrorBody(text) {
     const error = parsed.error;
     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);
+    const requestId = parsed.request_id ?? error?.request_id ?? (message ? extractRequestIdFromMessage(message) : void 0);
     const errorObj = error ?? parsed;
     return {
       ...message ? { message } : {},
@@ -2098,12 +2216,6 @@ var CODE_ASSIST_SUPPORTED_MODELS = /* @__PURE__ */ new Set([
   "gemma-4-31b-it",
   "gemma-4-26b-a4b-it"
 ]);
-function isJsonObject4(value) {
-  return value != null && typeof value === "object" && !Array.isArray(value);
-}
-function getEnvironment() {
-  return globalThis.process?.env;
-}
 function getGoogleProject(options) {
   const env = getEnvironment();
   return options.projectId ?? env?.GOOGLE_CLOUD_PROJECT ?? env?.GOOGLE_CLOUD_PROJECT_ID;
@@ -2125,6 +2237,20 @@ ${formatUnsupportedModelMessage(model)}`;
   }
   return `Gemini API error (${status}): ${body}`;
 }
+function parseRetryDelaySeconds(body) {
+  const match = body.match(/"retryDelay"\s*:\s*"(\d+(?:\.\d+)?)s"/);
+  if (!match) return void 0;
+  const seconds = Number(match[1]);
+  return Number.isFinite(seconds) ? seconds : void 0;
+}
+function parseGeminiQuota(status, body) {
+  if (status !== 429) return null;
+  const lower = body.toLowerCase();
+  if (!lower.includes("resource_exhausted") && !lower.includes("quota")) return null;
+  const retryDelaySeconds = parseRetryDelaySeconds(body);
+  const exhausted = retryDelaySeconds === void 0;
+  return { exhausted, retryDelaySeconds };
+}
 function toSystemAndContents(messages) {
   let systemText = "";
   const contents = [];
@@ -2203,7 +2329,7 @@ function toGeminiTools(tools) {
       functionDeclarations: tools.map((tool) => ({
         name: tool.name,
         description: tool.description,
-        parameters: sanitizeSchema(tool.rawInputSchema ?? zodToJsonSchema(tool.parameters))
+        parameters: sanitizeSchema(resolveToolSchema(tool))
       }))
     }
   ];
@@ -2214,7 +2340,7 @@ function sanitizeSchema(schema) {
   return clone;
 }
 function stripUnsupportedSchemaFields(value) {
-  if (!isJsonObject4(value)) {
+  if (!isJsonObject(value)) {
     if (Array.isArray(value)) {
       for (const item of value) stripUnsupportedSchemaFields(item);
     }
@@ -2223,7 +2349,7 @@ function stripUnsupportedSchemaFields(value) {
   delete value.$schema;
   delete value.additionalProperties;
   for (const item of Object.values(value)) {
-    if (isJsonObject4(item) || Array.isArray(item)) {
+    if (isJsonObject(item) || Array.isArray(item)) {
       stripUnsupportedSchemaFields(item);
     }
   }
@@ -2276,7 +2402,8 @@ function toThinkingConfig(model, level) {
   };
 }
 function buildGenerateRequest(options) {
-  const downgradedMessages = downgradeUnsupportedImages(options.messages, options.supportsImages);
+  const downgradedImages = downgradeUnsupportedImages(options.messages, options.supportsImages);
+  const downgradedMessages = downgradeUnsupportedVideos(downgradedImages, options.supportsVideo);
   const { systemInstruction, contents } = toSystemAndContents(downgradedMessages);
   const tools = toGeminiTools(options.tools);
   const toolConfig = toGeminiToolConfig(options.toolChoice, options.tools);
@@ -2338,54 +2465,11 @@ function normalizeGeminiStopReason(reason) {
       return "end_turn";
   }
 }
-function parseSseEvents(buffer) {
-  const events = [];
-  let cursor = 0;
-  while (true) {
-    const next = buffer.indexOf("\n\n", cursor);
-    if (next === -1) break;
-    const raw = buffer.slice(cursor, next);
-    cursor = next + 2;
-    let eventName;
-    const dataLines = [];
-    for (const line of raw.split("\n")) {
-      if (line.startsWith("event:")) {
-        eventName = line.slice("event:".length).trim();
-      } else if (line.startsWith("data:")) {
-        dataLines.push(line.slice("data:".length).trimStart());
-      }
-    }
-    if (dataLines.length > 0) {
-      events.push({ event: eventName, data: dataLines.join("\n") });
-    }
-  }
-  return { events, remaining: buffer.slice(cursor) };
-}
 async function* streamSse(response) {
   if (!response.body) return;
-  const reader = response.body.getReader();
-  const decoder = new TextDecoder();
-  let buffer = "";
-  try {
-    while (true) {
-      const { done, value } = await reader.read();
-      if (done) break;
-      buffer += decoder.decode(value, { stream: true }).replace(/\r\n/g, "\n");
-      const parsed2 = parseSseEvents(buffer);
-      buffer = parsed2.remaining;
-      for (const event of parsed2.events) {
-        if (event.data === "[DONE]") continue;
-        yield JSON.parse(event.data);
-      }
-    }
-    buffer += decoder.decode().replace(/\r\n/g, "\n");
-    const parsed = parseSseEvents(buffer + "\n\n");
-    for (const event of parsed.events) {
-      if (event.data === "[DONE]") continue;
-      yield JSON.parse(event.data);
-    }
-  } finally {
-    reader.releaseLock();
+  for await (const event of readSseStream(response.body)) {
+    if (event.data === "[DONE]") continue;
+    yield JSON.parse(event.data);
   }
 }
 function candidatesFromResponse(response) {
@@ -2408,7 +2492,7 @@ function readFunctionCallPart(part) {
   return {
     ...part.functionCall.id ? { id: part.functionCall.id } : {},
     name: part.functionCall.name,
-    args: isJsonObject4(part.functionCall.args) ? part.functionCall.args : {}
+    args: isJsonObject(part.functionCall.args) ? part.functionCall.args : {}
   };
 }
 function makeToolCallId(index, providerId) {
@@ -2447,8 +2531,17 @@ async function fetchCodeAssist(plan, options) {
     });
     if (!response.ok) {
       const text = await response.text().catch(() => "");
-      throw new ProviderError("gemini", formatErrorMessage(response.status, text, options.model), {
-        statusCode: response.status
+      const quota = parseGeminiQuota(response.status, text);
+      let message = formatErrorMessage(response.status, text, options.model);
+      let resetsAt;
+      if (quota?.exhausted) {
+        message = `Gemini quota exhausted \u2014 usage limit reached. ${message}`;
+      } else if (quota?.retryDelaySeconds !== void 0) {
+        resetsAt = Math.floor(Date.now() / 1e3) + Math.ceil(quota.retryDelaySeconds);
+      }
+      throw new ProviderError("gemini", message, {
+        statusCode: response.status,
+        ...resetsAt !== void 0 ? { resetsAt } : {}
       });
     }
     return response;
@@ -2839,6 +2932,7 @@ export {
   StreamResult,
   formatError,
   formatErrorForDisplay,
+  isHardBillingMessage,
   isUsageLimitError,
   palsuAssistantMessage,
   palsuText,