@dreb/ai 2.61.0 → 2.61.2

package/dist/providers/openai-codex-responses.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"openai-codex-responses.d.ts","sourceRoot":"","sources":["../../src/providers/openai-codex-responses.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAEX,6BAA6B,EAG7B,MAAM,yCAAyC,CAAC;AAkBjD,OAAO,KAAK,EAKX,mBAAmB,EACnB,cAAc,EACd,aAAa,EAEb,MAAM,aAAa,CAAC;AAmCrB,MAAM,WAAW,2BAA4B,SAAQ,aAAa;IACjE,eAAe,CAAC,EAAE,MAAM,GAAG,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,CAAC;IAC3E,gBAAgB,CAAC,EAAE,MAAM,GAAG,SAAS,GAAG,UAAU,GAAG,KAAK,GAAG,IAAI,GAAG,IAAI,CAAC;IACzE,aAAa,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAC;IAC1C,WAAW,CAAC,EAAE,6BAA6B,CAAC,cAAc,CAAC,CAAC;IAC5D,wFAAwF;IACxF,UAAU,CAAC,EAAE,CAAC,QAAQ,EAAE;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;KAAE,KAAK,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CACrG;AAoDD,eAAO,MAAM,0BAA0B,EAAE,cAAc,CAAC,wBAAwB,EAAE,2BAA2B,CA8L5G,CAAC;AAEF,eAAO,MAAM,gCAAgC,EAAE,cAAc,CAAC,wBAAwB,EAAE,mBAAmB,CAiB1G,CAAC","sourcesContent":["import type * as NodeOs from \"node:os\";\nimport type {\n\tTool as OpenAITool,\n\tResponseCreateParamsStreaming,\n\tResponseInput,\n\tResponseStreamEvent,\n} from \"openai/resources/responses/responses.js\";\n\n// NEVER convert to top-level runtime imports - breaks browser/Vite builds (web-ui)\nlet _os: typeof NodeOs | null = null;\n\ntype DynamicImport = (specifier: string) => Promise<unknown>;\n\nconst dynamicImport: DynamicImport = (specifier) => import(specifier);\nconst NODE_OS_SPECIFIER = \"node:\" + \"os\";\n\nif (typeof process !== \"undefined\" && (process.versions?.node || process.versions?.bun)) {\n\tdynamicImport(NODE_OS_SPECIFIER).then((m) => {\n\t\t_os = m as typeof NodeOs;\n\t});\n}\n\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport { supportsXhigh } from \"../models.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tContext,\n\tModel,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tUsage,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { convertResponsesMessages, convertResponsesTools, processResponsesStream } from \"./openai-responses-shared.js\";\nimport { buildBaseOptions, clampReasoning } from \"./simple-options.js\";\n\n// ============================================================================\n// Configuration\n// ============================================================================\n\nconst DEFAULT_CODEX_BASE_URL = \"https://chatgpt.com/backend-api\";\nconst JWT_CLAIM_PATH = \"https://api.openai.com/auth\" as const;\nconst MAX_RETRIES = 3;\nconst BASE_DELAY_MS = 1000;\nconst CODEX_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\"]);\nconst CODEX_COMPATIBILITY_VERSION = \"0.144.0\";\nconst RESPONSES_LITE_MODELS = new Set([\"gpt-5.6-sol\", \"gpt-5.6-terra\", \"gpt-5.6-luna\"]);\nconst RESPONSES_LITE_HEADER = \"x-openai-internal-codex-responses-lite\";\nconst RESPONSES_LITE_CLIENT_METADATA = \"ws_request_header_x_openai_internal_codex_responses_lite\";\nconst CODEX_LITE_SESSION_TTL_MS = 30 * 60 * 1000;\nconst CODEX_LITE_MAX_SESSIONS = 1000;\nconst codexLiteSessionIds = new Map<string, { sessionId: string; lastUsedAt: number }>();\n\nconst CODEX_RESPONSE_STATUSES = new Set<CodexResponseStatus>([\n\t\"completed\",\n\t\"incomplete\",\n\t\"failed\",\n\t\"cancelled\",\n\t\"queued\",\n\t\"in_progress\",\n]);\n\n// ============================================================================\n// Types\n// ============================================================================\n\nexport interface OpenAICodexResponsesOptions extends StreamOptions {\n\treasoningEffort?: \"none\" | \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\";\n\treasoningSummary?: \"auto\" | \"concise\" | \"detailed\" | \"off\" | \"on\" | null;\n\ttextVerbosity?: \"low\" | \"medium\" | \"high\";\n\tserviceTier?: ResponseCreateParamsStreaming[\"service_tier\"];\n\t/** Optional callback invoked when the HTTP response headers are received (SSE only). */\n\tonResponse?: (response: { status: number; headers: Record<string, string> }) => void | Promise<void>;\n}\n\ntype CodexResponseStatus = \"completed\" | \"incomplete\" | \"failed\" | \"cancelled\" | \"queued\" | \"in_progress\";\n\ninterface RequestBody {\n\tmodel: string;\n\tstore?: boolean;\n\tstream?: boolean;\n\tinstructions?: string;\n\tinput?: ResponseInput;\n\ttools?: OpenAITool[];\n\ttool_choice?: \"auto\";\n\tparallel_tool_calls?: boolean;\n\ttemperature?: number;\n\treasoning?: { effort?: string; summary?: string; context?: \"all_turns\" };\n\ttext?: { verbosity?: string };\n\tinclude?: string[];\n\tprompt_cache_key?: string;\n\tservice_tier?: ResponseCreateParamsStreaming[\"service_tier\"];\n\tprevious_response_id?: string;\n\t[key: string]: unknown;\n}\n\n// ============================================================================\n// Retry Helpers\n// ============================================================================\n\nfunction isRetryableError(status: number, errorText: string): boolean {\n\tif (status === 429 || status === 500 || status === 502 || status === 503 || status === 504) {\n\t\treturn true;\n\t}\n\treturn /rate.?limit|overloaded|service.?unavailable|upstream.?connect|connection.?refused/i.test(errorText);\n}\n\nfunction sleep(ms: number, signal?: AbortSignal): Promise<void> {\n\treturn new Promise((resolve, reject) => {\n\t\tif (signal?.aborted) {\n\t\t\treject(new Error(\"Request was aborted\"));\n\t\t\treturn;\n\t\t}\n\t\tconst timeout = setTimeout(resolve, ms);\n\t\tsignal?.addEventListener(\"abort\", () => {\n\t\t\tclearTimeout(timeout);\n\t\t\treject(new Error(\"Request was aborted\"));\n\t\t});\n\t});\n}\n\n// ============================================================================\n// Main Stream Function\n// ============================================================================\n\nexport const streamOpenAICodexResponses: StreamFunction<\"openai-codex-responses\", OpenAICodexResponsesOptions> = (\n\tmodel: Model<\"openai-codex-responses\">,\n\tcontext: Context,\n\toptions?: OpenAICodexResponsesOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t(async () => {\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: \"openai-codex-responses\" as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider) || \"\";\n\t\t\tif (!apiKey) {\n\t\t\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t\t\t}\n\n\t\t\tconst accountId = extractAccountId(apiKey);\n\t\t\tlet body = buildRequestBody(model, context, options);\n\t\t\tconst websocketRequestId = options?.sessionId || createCodexRequestId();\n\t\t\tconst usesResponsesLite = isResponsesLiteModel(model.id);\n\t\t\tconst transport = options?.transport || \"sse\";\n\t\t\tconst liteSessionId = usesResponsesLite\n\t\t\t\t? options?.sessionId\n\t\t\t\t\t? getCodexLiteSessionId(options.sessionId)\n\t\t\t\t\t: createUuidV7()\n\t\t\t\t: undefined;\n\n\t\t\tif (usesResponsesLite) {\n\t\t\t\tbody = prepareResponsesLiteRequest(body, liteSessionId!);\n\t\t\t}\n\t\t\tconst nextBody = await options?.onPayload?.(body, model);\n\t\t\tif (nextBody !== undefined) {\n\t\t\t\tbody = nextBody as RequestBody;\n\t\t\t}\n\t\t\tif (usesResponsesLite) {\n\t\t\t\tvalidateResponsesLiteRequest(body, liteSessionId!);\n\t\t\t}\n\t\t\tconst sseHeaders = buildSSEHeaders(\n\t\t\t\tmodel.headers,\n\t\t\t\toptions?.headers,\n\t\t\t\taccountId,\n\t\t\t\tapiKey,\n\t\t\t\toptions?.sessionId,\n\t\t\t\tliteSessionId,\n\t\t\t);\n\t\t\tconst websocketHeaders = buildWebSocketHeaders(\n\t\t\t\tmodel.headers,\n\t\t\t\toptions?.headers,\n\t\t\t\taccountId,\n\t\t\t\tapiKey,\n\t\t\t\twebsocketRequestId,\n\t\t\t\tliteSessionId,\n\t\t\t);\n\t\t\tconst bodyJson = JSON.stringify(body);\n\n\t\t\tif (transport !== \"sse\") {\n\t\t\t\tlet websocketStarted = false;\n\t\t\t\ttry {\n\t\t\t\t\tawait processWebSocketStream(\n\t\t\t\t\t\tresolveCodexWebSocketUrl(model.baseUrl),\n\t\t\t\t\t\tbody,\n\t\t\t\t\t\twebsocketHeaders,\n\t\t\t\t\t\toutput,\n\t\t\t\t\t\tstream,\n\t\t\t\t\t\tmodel,\n\t\t\t\t\t\t() => {\n\t\t\t\t\t\t\twebsocketStarted = true;\n\t\t\t\t\t\t},\n\t\t\t\t\t\toptions,\n\t\t\t\t\t);\n\n\t\t\t\t\tif (options?.signal?.aborted) {\n\t\t\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t\t\t}\n\t\t\t\t\tstream.push({\n\t\t\t\t\t\ttype: \"done\",\n\t\t\t\t\t\treason: output.stopReason as \"stop\" | \"length\" | \"toolUse\",\n\t\t\t\t\t\tmessage: output,\n\t\t\t\t\t});\n\t\t\t\t\tstream.end();\n\t\t\t\t\treturn;\n\t\t\t\t} catch (error) {\n\t\t\t\t\tif (transport === \"websocket\" || websocketStarted) {\n\t\t\t\t\t\tthrow error;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Fetch with retry logic for rate limits and transient errors\n\t\t\tlet response: Response | undefined;\n\t\t\tlet lastError: Error | undefined;\n\n\t\t\tfor (let attempt = 0; attempt <= MAX_RETRIES; attempt++) {\n\t\t\t\tif (options?.signal?.aborted) {\n\t\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t\t}\n\n\t\t\t\ttry {\n\t\t\t\t\tresponse = await fetch(resolveCodexUrl(model.baseUrl), {\n\t\t\t\t\t\tmethod: \"POST\",\n\t\t\t\t\t\theaders: sseHeaders,\n\t\t\t\t\t\tbody: bodyJson,\n\t\t\t\t\t\tsignal: options?.signal,\n\t\t\t\t\t});\n\n\t\t\t\t\tif (response.ok) {\n\t\t\t\t\t\tbreak;\n\t\t\t\t\t}\n\n\t\t\t\t\tconst errorText = await response.text();\n\t\t\t\t\tif (attempt < MAX_RETRIES && isRetryableError(response.status, errorText)) {\n\t\t\t\t\t\tconst delayMs = BASE_DELAY_MS * 2 ** attempt;\n\t\t\t\t\t\tawait sleep(delayMs, options?.signal);\n\t\t\t\t\t\tcontinue;\n\t\t\t\t\t}\n\n\t\t\t\t\t// Parse error for friendly message on final attempt or non-retryable error\n\t\t\t\t\tconst fakeResponse = new Response(errorText, {\n\t\t\t\t\t\tstatus: response.status,\n\t\t\t\t\t\tstatusText: response.statusText,\n\t\t\t\t\t});\n\t\t\t\t\tconst info = await parseErrorResponse(fakeResponse);\n\t\t\t\t\tthrow new Error(info.friendlyMessage || info.message);\n\t\t\t\t} catch (error) {\n\t\t\t\t\tif (error instanceof Error) {\n\t\t\t\t\t\tif (error.name === \"AbortError\" || error.message === \"Request was aborted\") {\n\t\t\t\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tlastError = error instanceof Error ? error : new Error(String(error));\n\t\t\t\t\t// Network errors are retryable\n\t\t\t\t\tif (attempt < MAX_RETRIES && !lastError.message.includes(\"usage limit\")) {\n\t\t\t\t\t\tconst delayMs = BASE_DELAY_MS * 2 ** attempt;\n\t\t\t\t\t\tawait sleep(delayMs, options?.signal);\n\t\t\t\t\t\tcontinue;\n\t\t\t\t\t}\n\t\t\t\t\tthrow lastError;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (!response?.ok) {\n\t\t\t\tthrow lastError ?? new Error(\"Failed after retries\");\n\t\t\t}\n\n\t\t\t// Fire onResponse exactly once, after the retry loop succeeds.\n\t\t\t// Wrap in its own try/catch so callback errors don't affect stream processing.\n\t\t\ttry {\n\t\t\t\tawait options?.onResponse?.({ status: response.status, headers: headersToRecord(response.headers) });\n\t\t\t} catch {\n\t\t\t\t// Intentionally ignored — callback errors must not disrupt the stream\n\t\t\t}\n\n\t\t\tif (!response.body) {\n\t\t\t\tthrow new Error(\"No response body\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"start\", partial: output });\n\t\t\tawait processStream(response, output, stream, model, options);\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason as \"stop\" | \"length\" | \"toolUse\", message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : String(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimpleOpenAICodexResponses: StreamFunction<\"openai-codex-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"openai-codex-responses\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst base = buildBaseOptions(model, options, apiKey);\n\tconst reasoningEffort = supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning);\n\n\treturn streamOpenAICodexResponses(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies OpenAICodexResponsesOptions);\n};\n\n// ============================================================================\n// Request Building\n// ============================================================================\n\nfunction buildRequestBody(\n\tmodel: Model<\"openai-codex-responses\">,\n\tcontext: Context,\n\toptions?: OpenAICodexResponsesOptions,\n): RequestBody {\n\tconst messages = convertResponsesMessages(model, context, CODEX_TOOL_CALL_PROVIDERS, {\n\t\tincludeSystemPrompt: false,\n\t});\n\n\tconst body: RequestBody = {\n\t\tmodel: model.id,\n\t\tstore: false,\n\t\tstream: true,\n\t\tinstructions: context.systemPrompt,\n\t\tinput: messages,\n\t\ttext: { verbosity: options?.textVerbosity || \"low\" },\n\t\tinclude: [\"reasoning.encrypted_content\"],\n\t\tprompt_cache_key: options?.sessionId,\n\t\ttool_choice: \"auto\",\n\t\tparallel_tool_calls: true,\n\t};\n\n\tif (options?.temperature !== undefined) {\n\t\tbody.temperature = options.temperature;\n\t}\n\n\tif (context.tools) {\n\t\tbody.tools = convertResponsesTools(context.tools, { strict: null });\n\t}\n\n\tif (options?.reasoningEffort !== undefined) {\n\t\tbody.reasoning = {\n\t\t\teffort: clampReasoningEffort(model.id, options.reasoningEffort),\n\t\t\tsummary: options.reasoningSummary ?? \"auto\",\n\t\t};\n\t}\n\n\tif (options?.serviceTier !== undefined) {\n\t\tbody.service_tier = options.serviceTier;\n\t}\n\n\treturn body;\n}\n\nfunction isResponsesLiteModel(modelId: string): boolean {\n\tconst id = modelId.includes(\"/\") ? modelId.split(\"/\").pop()! : modelId;\n\treturn RESPONSES_LITE_MODELS.has(id);\n}\n\nfunction getCodexLiteSessionId(sourceSessionId: string): string {\n\tconst now = Date.now();\n\tfor (const [key, entry] of codexLiteSessionIds) {\n\t\tif (now - entry.lastUsedAt > CODEX_LITE_SESSION_TTL_MS) {\n\t\t\tcodexLiteSessionIds.delete(key);\n\t\t}\n\t}\n\n\tconst existing = codexLiteSessionIds.get(sourceSessionId);\n\tif (existing) {\n\t\texisting.lastUsedAt = now;\n\t\tcodexLiteSessionIds.delete(sourceSessionId);\n\t\tcodexLiteSessionIds.set(sourceSessionId, existing);\n\t\treturn existing.sessionId;\n\t}\n\n\twhile (codexLiteSessionIds.size >= CODEX_LITE_MAX_SESSIONS) {\n\t\tconst oldestKey = codexLiteSessionIds.keys().next().value;\n\t\tif (oldestKey === undefined) break;\n\t\tcodexLiteSessionIds.delete(oldestKey);\n\t}\n\n\tconst sessionId = createUuidV7();\n\tcodexLiteSessionIds.set(sourceSessionId, { sessionId, lastUsedAt: now });\n\treturn sessionId;\n}\n\nfunction createUuidV7(): string {\n\tconst timestamp = Date.now().toString(16).padStart(12, \"0\").slice(-12);\n\tconst random = new Uint8Array(10);\n\tif (typeof globalThis.crypto?.getRandomValues === \"function\") {\n\t\tglobalThis.crypto.getRandomValues(random);\n\t} else {\n\t\tfor (let index = 0; index < random.length; index++) {\n\t\t\trandom[index] = Math.floor(Math.random() * 256);\n\t\t}\n\t}\n\tconst randomHex = Array.from(random, (byte) => byte.toString(16).padStart(2, \"0\")).join(\"\");\n\tconst variant = (8 + (Number.parseInt(randomHex[3]!, 16) & 0x3)).toString(16);\n\treturn `${timestamp.slice(0, 8)}-${timestamp.slice(8)}-7${randomHex.slice(0, 3)}-${variant}${randomHex.slice(4, 7)}-${randomHex.slice(7, 19)}`;\n}\n\nfunction prepareResponsesLiteRequest(body: RequestBody, sessionId: string): RequestBody {\n\tif (!Array.isArray(body.input)) {\n\t\tthrow new Error(\"Responses Lite requires an input array\");\n\t}\n\tif (body.tools !== undefined && !Array.isArray(body.tools)) {\n\t\tthrow new Error(\"Responses Lite requires a tools array\");\n\t}\n\tif (body.instructions !== undefined && typeof body.instructions !== \"string\") {\n\t\tthrow new Error(\"Responses Lite requires string instructions\");\n\t}\n\n\tconst developerItems: unknown[] = [{ type: \"additional_tools\", role: \"developer\", tools: body.tools ?? [] }];\n\tif (body.instructions) {\n\t\tdeveloperItems.push({\n\t\t\ttype: \"message\",\n\t\t\trole: \"developer\",\n\t\t\tcontent: [{ type: \"input_text\", text: body.instructions }],\n\t\t});\n\t}\n\n\tconst input = [...developerItems, ...(body.input ?? [])] as ResponseInput;\n\tstripInputImageDetails(input);\n\tconst { tools: _tools, instructions: _instructions, reasoning, ...request } = body;\n\treturn {\n\t\t...request,\n\t\tinput,\n\t\ttool_choice: \"auto\",\n\t\tparallel_tool_calls: false,\n\t\tprompt_cache_key: sessionId,\n\t\treasoning: { ...reasoning, context: \"all_turns\" },\n\t};\n}\n\nfunction validateResponsesLiteRequest(body: RequestBody, sessionId: string): void {\n\tif (!body || typeof body !== \"object\" || Array.isArray(body)) {\n\t\tthrow new Error(\"Responses Lite onPayload must return an object payload\");\n\t}\n\n\tconst payload = body as Record<string, unknown>;\n\tif (Object.hasOwn(payload, \"instructions\")) {\n\t\tthrow new Error(\"Responses Lite does not support top-level instructions\");\n\t}\n\tif (Object.hasOwn(payload, \"tools\")) {\n\t\tthrow new Error(\"Responses Lite does not support top-level tools\");\n\t}\n\tif (!Array.isArray(payload.input)) {\n\t\tthrow new Error(\"Responses Lite requires an input array\");\n\t}\n\tif (payload.tool_choice !== \"auto\") {\n\t\tthrow new Error(\"Responses Lite requires tool_choice to be auto\");\n\t}\n\tif (payload.parallel_tool_calls !== false) {\n\t\tthrow new Error(\"Responses Lite requires parallel_tool_calls to be false\");\n\t}\n\tif (payload.prompt_cache_key !== sessionId) {\n\t\tthrow new Error(\"Responses Lite requires prompt_cache_key to match its session ID\");\n\t}\n\tif (\n\t\t!payload.reasoning ||\n\t\ttypeof payload.reasoning !== \"object\" ||\n\t\tArray.isArray(payload.reasoning) ||\n\t\t(payload.reasoning as Record<string, unknown>).context !== \"all_turns\"\n\t) {\n\t\tthrow new Error(\"Responses Lite requires reasoning.context to be all_turns\");\n\t}\n\n\tconst additionalTools = payload.input[0];\n\tif (\n\t\t!additionalTools ||\n\t\ttypeof additionalTools !== \"object\" ||\n\t\tArray.isArray(additionalTools) ||\n\t\t(additionalTools as Record<string, unknown>).type !== \"additional_tools\" ||\n\t\t(additionalTools as Record<string, unknown>).role !== \"developer\" ||\n\t\t!Array.isArray((additionalTools as Record<string, unknown>).tools)\n\t) {\n\t\tthrow new Error(\"Responses Lite requires a developer additional_tools input item\");\n\t}\n\tif (containsInputImageDetail(payload.input)) {\n\t\tthrow new Error(\"Responses Lite does not support input_image detail\");\n\t}\n}\n\nfunction stripInputImageDetails(value: unknown): void {\n\tif (Array.isArray(value)) {\n\t\tfor (const item of value) stripInputImageDetails(item);\n\t\treturn;\n\t}\n\tif (!value || typeof value !== \"object\") return;\n\n\tconst record = value as Record<string, unknown>;\n\tif (record.type === \"input_image\") {\n\t\tdelete record.detail;\n\t}\n\tfor (const nested of Object.values(record)) stripInputImageDetails(nested);\n}\n\nfunction containsInputImageDetail(value: unknown): boolean {\n\tif (Array.isArray(value)) return value.some(containsInputImageDetail);\n\tif (!value || typeof value !== \"object\") return false;\n\n\tconst record = value as Record<string, unknown>;\n\tif (record.type === \"input_image\" && Object.hasOwn(record, \"detail\")) return true;\n\treturn Object.values(record).some(containsInputImageDetail);\n}\n\nfunction clampReasoningEffort(modelId: string, effort: string): string {\n\tconst id = modelId.includes(\"/\") ? modelId.split(\"/\").pop()! : modelId;\n\tif (\n\t\t(id.startsWith(\"gpt-5.2\") ||\n\t\t\tid.startsWith(\"gpt-5.3\") ||\n\t\t\tid.startsWith(\"gpt-5.4\") ||\n\t\t\tid.startsWith(\"gpt-5.5\") ||\n\t\t\tid.startsWith(\"gpt-5.6\")) &&\n\t\teffort === \"minimal\"\n\t)\n\t\treturn \"low\";\n\tif (id === \"gpt-5.1\" && effort === \"xhigh\") return \"high\";\n\tif (id === \"gpt-5.1-codex-mini\") return effort === \"high\" || effort === \"xhigh\" ? \"high\" : \"medium\";\n\treturn effort;\n}\n\nfunction resolveCodexUrl(baseUrl?: string): string {\n\tconst raw = baseUrl && baseUrl.trim().length > 0 ? baseUrl : DEFAULT_CODEX_BASE_URL;\n\tconst normalized = raw.replace(/\\/+$/, \"\");\n\tif (normalized.endsWith(\"/codex/responses\")) return normalized;\n\tif (normalized.endsWith(\"/codex\")) return `${normalized}/responses`;\n\treturn `${normalized}/codex/responses`;\n}\n\nfunction resolveCodexWebSocketUrl(baseUrl?: string): string {\n\tconst url = new URL(resolveCodexUrl(baseUrl));\n\tif (url.protocol === \"https:\") url.protocol = \"wss:\";\n\tif (url.protocol === \"http:\") url.protocol = \"ws:\";\n\treturn url.toString();\n}\n\n// ============================================================================\n// Response Processing\n// ============================================================================\n\nasync function processStream(\n\tresponse: Response,\n\toutput: AssistantMessage,\n\tstream: AssistantMessageEventStream,\n\tmodel: Model<\"openai-codex-responses\">,\n\toptions?: OpenAICodexResponsesOptions,\n): Promise<void> {\n\tawait processResponsesStream(mapCodexEvents(parseSSE(response, options)), output, stream, model, {\n\t\tserviceTier: options?.serviceTier,\n\t\tresolveServiceTier: resolveCodexServiceTier,\n\t\tapplyServiceTierPricing: (usage, serviceTier) => applyServiceTierPricing(usage, serviceTier, model),\n\t\tonWarning: options?.onWarning,\n\t});\n}\n\nasync function* mapCodexEvents(events: AsyncIterable<Record<string, unknown>>): AsyncGenerator<ResponseStreamEvent> {\n\tfor await (const event of events) {\n\t\tconst type = typeof event.type === \"string\" ? event.type : undefined;\n\t\tif (!type) continue;\n\n\t\tif (type === \"error\") {\n\t\t\tconst code = (event as { code?: string }).code || \"\";\n\t\t\tconst message = (event as { message?: string }).message || \"\";\n\t\t\tthrow new Error(`Codex error: ${message || code || JSON.stringify(event)}`);\n\t\t}\n\n\t\tif (type === \"response.failed\") {\n\t\t\tconst msg = (event as { response?: { error?: { message?: string } } }).response?.error?.message;\n\t\t\tthrow new Error(msg || \"Codex response failed\");\n\t\t}\n\n\t\tif (type === \"response.done\" || type === \"response.completed\" || type === \"response.incomplete\") {\n\t\t\tconst response = (event as { response?: { status?: unknown } }).response;\n\t\t\tconst normalizedResponse = response\n\t\t\t\t? { ...response, status: normalizeCodexStatus(response.status) }\n\t\t\t\t: response;\n\t\t\tyield { ...event, type: \"response.completed\", response: normalizedResponse } as ResponseStreamEvent;\n\t\t\treturn;\n\t\t}\n\n\t\tyield event as unknown as ResponseStreamEvent;\n\t}\n}\n\nfunction normalizeCodexStatus(status: unknown): CodexResponseStatus | undefined {\n\tif (typeof status !== \"string\") return undefined;\n\treturn CODEX_RESPONSE_STATUSES.has(status as CodexResponseStatus) ? (status as CodexResponseStatus) : undefined;\n}\n\n// ============================================================================\n// SSE Parsing\n// ============================================================================\n\nasync function* parseSSE(\n\tresponse: Response,\n\toptions?: OpenAICodexResponsesOptions,\n): AsyncGenerator<Record<string, unknown>> {\n\tif (!response.body) return;\n\n\tconst reader = response.body.getReader();\n\tconst decoder = new TextDecoder();\n\tlet sseParseErrors = 0;\n\tlet buffer = \"\";\n\n\ttry {\n\t\twhile (true) {\n\t\t\tconst { done, value } = await reader.read();\n\t\t\tif (done) break;\n\t\t\tbuffer += decoder.decode(value, { stream: true });\n\n\t\t\tlet idx = buffer.indexOf(\"\\n\\n\");\n\t\t\twhile (idx !== -1) {\n\t\t\t\tconst chunk = buffer.slice(0, idx);\n\t\t\t\tbuffer = buffer.slice(idx + 2);\n\n\t\t\t\tconst dataLines = chunk\n\t\t\t\t\t.split(\"\\n\")\n\t\t\t\t\t.filter((l) => l.startsWith(\"data:\"))\n\t\t\t\t\t.map((l) => l.slice(5).trim());\n\t\t\t\tif (dataLines.length > 0) {\n\t\t\t\t\tconst data = dataLines.join(\"\\n\").trim();\n\t\t\t\t\tif (data && data !== \"[DONE]\") {\n\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\tyield JSON.parse(data);\n\t\t\t\t\t\t} catch {\n\t\t\t\t\t\t\t// Malformed SSE event — track for end-of-stream summary\n\t\t\t\t\t\t\tsseParseErrors++;\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tidx = buffer.indexOf(\"\\n\\n\");\n\t\t\t}\n\t\t}\n\t} finally {\n\t\tif (sseParseErrors > 0) {\n\t\t\toptions?.onWarning?.(\n\t\t\t\t\"sse_parse_error\",\n\t\t\t\t`${sseParseErrors} malformed SSE event(s) were dropped during streaming. Response content may be incomplete.`,\n\t\t\t);\n\t\t}\n\t\ttry {\n\t\t\tawait reader.cancel();\n\t\t} catch {\n\t\t\t// Reader may already be closed — cleanup errors are safe to ignore\n\t\t}\n\t\ttry {\n\t\t\treader.releaseLock();\n\t\t} catch {\n\t\t\t// Reader may already be closed — cleanup errors are safe to ignore\n\t\t}\n\t}\n}\n\n// ============================================================================\n// WebSocket Parsing\n// ============================================================================\n\nconst OPENAI_BETA_RESPONSES_WEBSOCKETS = \"responses_websockets=2026-02-06\";\nconst SESSION_WEBSOCKET_CACHE_TTL_MS = 5 * 60 * 1000;\n\ntype WebSocketEventType = \"open\" | \"message\" | \"error\" | \"close\";\ntype WebSocketListener = (event: unknown) => void;\n\ninterface WebSocketLike {\n\tclose(code?: number, reason?: string): void;\n\tsend(data: string): void;\n\taddEventListener(type: WebSocketEventType, listener: WebSocketListener): void;\n\tremoveEventListener(type: WebSocketEventType, listener: WebSocketListener): void;\n}\n\ninterface CachedWebSocketContinuationState {\n\tlastRequestBody: RequestBody;\n\tlastResponseId: string;\n\tlastResponseItems: ResponseInput;\n}\n\ninterface CachedWebSocketConnection {\n\tsocket: WebSocketLike;\n\tbusy: boolean;\n\tidleTimer?: ReturnType<typeof setTimeout>;\n\tcontinuation?: CachedWebSocketContinuationState;\n}\n\nconst websocketSessionCache = new Map<string, CachedWebSocketConnection>();\n\ntype WebSocketConstructor = new (\n\turl: string,\n\tprotocols?: string | string[] | { headers?: Record<string, string> },\n) => WebSocketLike;\n\nfunction getWebSocketConstructor(): WebSocketConstructor | null {\n\tconst ctor = (globalThis as { WebSocket?: unknown }).WebSocket;\n\tif (typeof ctor !== \"function\") return null;\n\treturn ctor as unknown as WebSocketConstructor;\n}\n\nfunction headersToRecord(headers: Headers): Record<string, string> {\n\tconst out: Record<string, string> = {};\n\tfor (const [key, value] of headers.entries()) {\n\t\tout[key] = value;\n\t}\n\treturn out;\n}\n\nfunction getWebSocketReadyState(socket: WebSocketLike): number | undefined {\n\tconst readyState = (socket as { readyState?: unknown }).readyState;\n\treturn typeof readyState === \"number\" ? readyState : undefined;\n}\n\nfunction isWebSocketReusable(socket: WebSocketLike): boolean {\n\tconst readyState = getWebSocketReadyState(socket);\n\t// If readyState is unavailable, assume the runtime keeps it open/reusable.\n\treturn readyState === undefined || readyState === 1;\n}\n\nfunction closeWebSocketSilently(socket: WebSocketLike, code = 1000, reason = \"done\"): void {\n\ttry {\n\t\tsocket.close(code, reason);\n\t} catch {\n\t\t// Intentionally silent — socket may already be closed\n\t}\n}\n\nfunction scheduleSessionWebSocketExpiry(sessionId: string, entry: CachedWebSocketConnection): void {\n\tif (entry.idleTimer) {\n\t\tclearTimeout(entry.idleTimer);\n\t}\n\tentry.idleTimer = setTimeout(() => {\n\t\tif (entry.busy) return;\n\t\tcloseWebSocketSilently(entry.socket, 1000, \"idle_timeout\");\n\t\twebsocketSessionCache.delete(sessionId);\n\t}, SESSION_WEBSOCKET_CACHE_TTL_MS);\n}\n\nasync function connectWebSocket(url: string, headers: Headers, signal?: AbortSignal): Promise<WebSocketLike> {\n\tconst WebSocketCtor = getWebSocketConstructor();\n\tif (!WebSocketCtor) {\n\t\tthrow new Error(\"WebSocket transport is not available in this runtime\");\n\t}\n\n\tconst wsHeaders = headersToRecord(headers);\n\tdelete wsHeaders[\"OpenAI-Beta\"];\n\n\treturn new Promise<WebSocketLike>((resolve, reject) => {\n\t\tlet settled = false;\n\t\tlet socket: WebSocketLike;\n\n\t\ttry {\n\t\t\tsocket = new WebSocketCtor(url, { headers: wsHeaders });\n\t\t} catch (error) {\n\t\t\treject(error instanceof Error ? error : new Error(String(error)));\n\t\t\treturn;\n\t\t}\n\n\t\tconst onOpen: WebSocketListener = () => {\n\t\t\tif (settled) return;\n\t\t\tsettled = true;\n\t\t\tcleanup();\n\t\t\tresolve(socket);\n\t\t};\n\t\tconst onError: WebSocketListener = (event) => {\n\t\t\tconst error = extractWebSocketError(event);\n\t\t\tif (settled) return;\n\t\t\tsettled = true;\n\t\t\tcleanup();\n\t\t\treject(error);\n\t\t};\n\t\tconst onClose: WebSocketListener = (event) => {\n\t\t\tconst error = extractWebSocketCloseError(event);\n\t\t\tif (settled) return;\n\t\t\tsettled = true;\n\t\t\tcleanup();\n\t\t\treject(error);\n\t\t};\n\t\tconst onAbort = () => {\n\t\t\tif (settled) return;\n\t\t\tsettled = true;\n\t\t\tcleanup();\n\t\t\tsocket.close(1000, \"aborted\");\n\t\t\treject(new Error(\"Request was aborted\"));\n\t\t};\n\n\t\tconst cleanup = () => {\n\t\t\tsocket.removeEventListener(\"open\", onOpen);\n\t\t\tsocket.removeEventListener(\"error\", onError);\n\t\t\tsocket.removeEventListener(\"close\", onClose);\n\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t};\n\n\t\tsocket.addEventListener(\"open\", onOpen);\n\t\tsocket.addEventListener(\"error\", onError);\n\t\tsocket.addEventListener(\"close\", onClose);\n\t\tsignal?.addEventListener(\"abort\", onAbort);\n\t});\n}\n\nasync function acquireWebSocket(\n\turl: string,\n\theaders: Headers,\n\tsessionId: string | undefined,\n\tsignal?: AbortSignal,\n): Promise<{\n\tsocket: WebSocketLike;\n\tentry?: CachedWebSocketConnection;\n\trelease: (options?: { keep?: boolean }) => void;\n}> {\n\tif (!sessionId) {\n\t\tconst socket = await connectWebSocket(url, headers, signal);\n\t\treturn {\n\t\t\tsocket,\n\t\t\trelease: () => {\n\t\t\t\tcloseWebSocketSilently(socket);\n\t\t\t},\n\t\t};\n\t}\n\n\tconst cached = websocketSessionCache.get(sessionId);\n\tif (cached) {\n\t\tif (cached.idleTimer) {\n\t\t\tclearTimeout(cached.idleTimer);\n\t\t\tcached.idleTimer = undefined;\n\t\t}\n\t\tif (!cached.busy && isWebSocketReusable(cached.socket)) {\n\t\t\tcached.busy = true;\n\t\t\treturn {\n\t\t\t\tsocket: cached.socket,\n\t\t\t\tentry: cached,\n\t\t\t\trelease: ({ keep } = {}) => {\n\t\t\t\t\tif (!keep || !isWebSocketReusable(cached.socket)) {\n\t\t\t\t\t\tcloseWebSocketSilently(cached.socket);\n\t\t\t\t\t\twebsocketSessionCache.delete(sessionId);\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\tcached.busy = false;\n\t\t\t\t\tscheduleSessionWebSocketExpiry(sessionId, cached);\n\t\t\t\t},\n\t\t\t};\n\t\t}\n\t\tif (cached.busy) {\n\t\t\tconst socket = await connectWebSocket(url, headers, signal);\n\t\t\treturn {\n\t\t\t\tsocket,\n\t\t\t\trelease: () => {\n\t\t\t\t\tcloseWebSocketSilently(socket);\n\t\t\t\t},\n\t\t\t};\n\t\t}\n\t\tif (!isWebSocketReusable(cached.socket)) {\n\t\t\tcloseWebSocketSilently(cached.socket);\n\t\t\twebsocketSessionCache.delete(sessionId);\n\t\t}\n\t}\n\n\tconst socket = await connectWebSocket(url, headers, signal);\n\tconst entry: CachedWebSocketConnection = { socket, busy: true };\n\twebsocketSessionCache.set(sessionId, entry);\n\treturn {\n\t\tsocket,\n\t\tentry,\n\t\trelease: ({ keep } = {}) => {\n\t\t\tif (!keep || !isWebSocketReusable(entry.socket)) {\n\t\t\t\tcloseWebSocketSilently(entry.socket);\n\t\t\t\tif (entry.idleTimer) clearTimeout(entry.idleTimer);\n\t\t\t\tif (websocketSessionCache.get(sessionId) === entry) {\n\t\t\t\t\twebsocketSessionCache.delete(sessionId);\n\t\t\t\t}\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tentry.busy = false;\n\t\t\tscheduleSessionWebSocketExpiry(sessionId, entry);\n\t\t},\n\t};\n}\n\nfunction extractWebSocketError(event: unknown): Error {\n\tif (event && typeof event === \"object\" && \"message\" in event) {\n\t\tconst message = (event as { message?: unknown }).message;\n\t\tif (typeof message === \"string\" && message.length > 0) {\n\t\t\treturn new Error(message);\n\t\t}\n\t}\n\treturn new Error(\"WebSocket error\");\n}\n\nfunction extractWebSocketCloseError(event: unknown): Error {\n\tif (event && typeof event === \"object\") {\n\t\tconst code = \"code\" in event ? (event as { code?: unknown }).code : undefined;\n\t\tconst reason = \"reason\" in event ? (event as { reason?: unknown }).reason : undefined;\n\t\tconst codeText = typeof code === \"number\" ? ` ${code}` : \"\";\n\t\tconst reasonText = typeof reason === \"string\" && reason.length > 0 ? ` ${reason}` : \"\";\n\t\treturn new Error(`WebSocket closed${codeText}${reasonText}`.trim());\n\t}\n\treturn new Error(\"WebSocket closed\");\n}\n\nasync function decodeWebSocketData(data: unknown): Promise<string | null> {\n\tif (typeof data === \"string\") return data;\n\tif (data instanceof ArrayBuffer) {\n\t\treturn new TextDecoder().decode(new Uint8Array(data));\n\t}\n\tif (ArrayBuffer.isView(data)) {\n\t\tconst view = data as ArrayBufferView;\n\t\treturn new TextDecoder().decode(new Uint8Array(view.buffer, view.byteOffset, view.byteLength));\n\t}\n\tif (data && typeof data === \"object\" && \"arrayBuffer\" in data) {\n\t\tconst blobLike = data as { arrayBuffer: () => Promise<ArrayBuffer> };\n\t\tconst arrayBuffer = await blobLike.arrayBuffer();\n\t\treturn new TextDecoder().decode(new Uint8Array(arrayBuffer));\n\t}\n\treturn null;\n}\n\nasync function* parseWebSocket(\n\tsocket: WebSocketLike,\n\tsignal?: AbortSignal,\n\toptions?: OpenAICodexResponsesOptions,\n): AsyncGenerator<Record<string, unknown>> {\n\tconst queue: Record<string, unknown>[] = [];\n\tlet pending: (() => void) | null = null;\n\tlet done = false;\n\tlet failed: Error | null = null;\n\tlet sawCompletion = false;\n\n\tconst wake = () => {\n\t\tif (!pending) return;\n\t\tconst resolve = pending;\n\t\tpending = null;\n\t\tresolve();\n\t};\n\n\tconst onMessage: WebSocketListener = (event) => {\n\t\tvoid (async () => {\n\t\t\tif (!event || typeof event !== \"object\" || !(\"data\" in event)) return;\n\t\t\tlet text: string | null | undefined;\n\t\t\ttry {\n\t\t\t\ttext = await decodeWebSocketData((event as { data?: unknown }).data);\n\t\t\t\tif (!text) {\n\t\t\t\t\twake();\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tconst parsed = JSON.parse(text) as Record<string, unknown>;\n\t\t\t\tconst type = typeof parsed.type === \"string\" ? parsed.type : \"\";\n\t\t\t\tif (type === \"response.completed\" || type === \"response.done\" || type === \"response.incomplete\") {\n\t\t\t\t\tsawCompletion = true;\n\t\t\t\t\tdone = true;\n\t\t\t\t}\n\t\t\t\tqueue.push(parsed);\n\t\t\t} catch (e) {\n\t\t\t\t// Malformed WebSocket message — warn but don't drop wake()\n\t\t\t\tif (text === undefined) {\n\t\t\t\t\toptions?.onWarning?.(\n\t\t\t\t\t\t\"ws_decode_error\",\n\t\t\t\t\t\t`Failed to decode WebSocket message: ${e instanceof Error ? e.message : String(e)}`,\n\t\t\t\t\t);\n\t\t\t\t} else {\n\t\t\t\t\toptions?.onWarning?.(\n\t\t\t\t\t\t\"ws_parse_error\",\n\t\t\t\t\t\t`Malformed WebSocket message dropped: ${text?.slice(0, 200) ?? \"\"}`,\n\t\t\t\t\t);\n\t\t\t\t}\n\t\t\t} finally {\n\t\t\t\t// Always wake the consumer — even on parse failure — to prevent stream hangs\n\t\t\t\twake();\n\t\t\t}\n\t\t})();\n\t};\n\n\tconst onError: WebSocketListener = (event) => {\n\t\tfailed = extractWebSocketError(event);\n\t\tdone = true;\n\t\twake();\n\t};\n\n\tconst onClose: WebSocketListener = (event) => {\n\t\tif (sawCompletion) {\n\t\t\tdone = true;\n\t\t\twake();\n\t\t\treturn;\n\t\t}\n\t\tif (!failed) {\n\t\t\tconst closeError = extractWebSocketCloseError(event);\n\t\t\tfailed = new Error(`WebSocket stream closed before response.completed (${closeError.message})`);\n\t\t}\n\t\tdone = true;\n\t\twake();\n\t};\n\n\tconst onAbort = () => {\n\t\tfailed = new Error(\"Request was aborted\");\n\t\tdone = true;\n\t\twake();\n\t};\n\n\tsocket.addEventListener(\"message\", onMessage);\n\tsocket.addEventListener(\"error\", onError);\n\tsocket.addEventListener(\"close\", onClose);\n\tsignal?.addEventListener(\"abort\", onAbort);\n\n\ttry {\n\t\twhile (true) {\n\t\t\tif (signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\t\t\tif (queue.length > 0) {\n\t\t\t\tyield queue.shift()!;\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (done) break;\n\t\t\tawait new Promise<void>((resolve) => {\n\t\t\t\tpending = resolve;\n\t\t\t});\n\t\t}\n\n\t\tif (failed) {\n\t\t\tthrow failed;\n\t\t}\n\t\tif (!sawCompletion) {\n\t\t\tthrow new Error(\"WebSocket stream closed before response.completed\");\n\t\t}\n\t} finally {\n\t\tsocket.removeEventListener(\"message\", onMessage);\n\t\tsocket.removeEventListener(\"error\", onError);\n\t\tsocket.removeEventListener(\"close\", onClose);\n\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t}\n}\n\nfunction requestBodyWithoutInput(body: RequestBody): RequestBody {\n\tconst { input: _input, previous_response_id: _previousResponseId, ...rest } = body;\n\treturn rest;\n}\n\nfunction responseInputsEqual(a: ResponseInput | undefined, b: ResponseInput | undefined): boolean {\n\treturn JSON.stringify(a ?? []) === JSON.stringify(b ?? []);\n}\n\nfunction requestBodiesMatchExceptInput(a: RequestBody, b: RequestBody): boolean {\n\treturn JSON.stringify(requestBodyWithoutInput(a)) === JSON.stringify(requestBodyWithoutInput(b));\n}\n\nfunction getCachedWebSocketInputDelta(\n\tbody: RequestBody,\n\tcontinuation: CachedWebSocketContinuationState,\n): ResponseInput | undefined {\n\tif (!requestBodiesMatchExceptInput(body, continuation.lastRequestBody)) {\n\t\treturn undefined;\n\t}\n\n\tconst currentInput = body.input ?? [];\n\tconst baseline = [...(continuation.lastRequestBody.input ?? []), ...continuation.lastResponseItems];\n\tif (currentInput.length < baseline.length) {\n\t\treturn undefined;\n\t}\n\n\tconst prefix = currentInput.slice(0, baseline.length);\n\tif (!responseInputsEqual(prefix, baseline)) {\n\t\treturn undefined;\n\t}\n\n\treturn currentInput.slice(baseline.length);\n}\n\nfunction buildCachedWebSocketRequestBody(entry: CachedWebSocketConnection, body: RequestBody): RequestBody {\n\tconst continuation = entry.continuation;\n\tif (!continuation) {\n\t\treturn body;\n\t}\n\n\tconst delta = getCachedWebSocketInputDelta(body, continuation);\n\tif (!delta || !continuation.lastResponseId) {\n\t\tentry.continuation = undefined;\n\t\treturn body;\n\t}\n\n\treturn {\n\t\t...body,\n\t\tprevious_response_id: continuation.lastResponseId,\n\t\tinput: delta,\n\t};\n}\n\nfunction addResponsesLiteWebSocketMetadata(body: RequestBody): RequestBody {\n\tconst clientMetadata = body.client_metadata;\n\treturn {\n\t\t...body,\n\t\tclient_metadata: {\n\t\t\t...(clientMetadata && typeof clientMetadata === \"object\" && !Array.isArray(clientMetadata)\n\t\t\t\t? clientMetadata\n\t\t\t\t: {}),\n\t\t\t[RESPONSES_LITE_CLIENT_METADATA]: \"true\",\n\t\t},\n\t};\n}\n\nasync function processWebSocketStream(\n\turl: string,\n\tbody: RequestBody,\n\theaders: Headers,\n\toutput: AssistantMessage,\n\tstream: AssistantMessageEventStream,\n\tmodel: Model<\"openai-codex-responses\">,\n\tonStart: () => void,\n\toptions?: OpenAICodexResponsesOptions,\n): Promise<void> {\n\tconst cacheSessionId = options?.sessionId\n\t\t? `${isResponsesLiteModel(model.id) ? \"responses-lite\" : \"legacy\"}:${options.sessionId}`\n\t\t: undefined;\n\tconst { socket, entry, release } = await acquireWebSocket(url, headers, cacheSessionId, options?.signal);\n\tlet keepConnection = true;\n\tconst fullBody = body;\n\tconst requestBody = entry ? buildCachedWebSocketRequestBody(entry, fullBody) : fullBody;\n\tconst wireBody = isResponsesLiteModel(model.id) ? addResponsesLiteWebSocketMetadata(requestBody) : requestBody;\n\ttry {\n\t\tsocket.send(JSON.stringify({ type: \"response.create\", ...wireBody }));\n\t\tonStart();\n\t\tstream.push({ type: \"start\", partial: output });\n\t\tawait processResponsesStream(\n\t\t\tmapCodexEvents(parseWebSocket(socket, options?.signal, options)),\n\t\t\toutput,\n\t\t\tstream,\n\t\t\tmodel,\n\t\t\t{\n\t\t\t\tserviceTier: options?.serviceTier,\n\t\t\t\tresolveServiceTier: resolveCodexServiceTier,\n\t\t\t\tapplyServiceTierPricing: (usage, serviceTier) => applyServiceTierPricing(usage, serviceTier, model),\n\t\t\t\tonWarning: options?.onWarning,\n\t\t\t},\n\t\t);\n\t\tif (entry && output.responseId && !options?.signal?.aborted) {\n\t\t\tconst responseItems = convertResponsesMessages(model, { messages: [output] }, CODEX_TOOL_CALL_PROVIDERS, {\n\t\t\t\tincludeSystemPrompt: false,\n\t\t\t}).filter((item) => item.type !== \"function_call_output\");\n\t\t\tentry.continuation = {\n\t\t\t\tlastRequestBody: fullBody,\n\t\t\t\tlastResponseId: output.responseId,\n\t\t\t\tlastResponseItems: responseItems,\n\t\t\t};\n\t\t}\n\t\tif (options?.signal?.aborted) {\n\t\t\tkeepConnection = false;\n\t\t}\n\t} catch (error) {\n\t\tif (entry) {\n\t\t\tentry.continuation = undefined;\n\t\t}\n\t\tkeepConnection = false;\n\t\tthrow error;\n\t} finally {\n\t\trelease({ keep: keepConnection });\n\t}\n}\n\n// ============================================================================\n// Service Tier Pricing\n// ============================================================================\n\nfunction getServiceTierCostMultiplier(\n\tserviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n\tmodelId?: string,\n): number {\n\tswitch (serviceTier) {\n\t\tcase \"flex\":\n\t\t\treturn 0.5;\n\t\tcase \"priority\": {\n\t\t\tconst id = modelId?.includes(\"/\") ? modelId.split(\"/\").pop()! : modelId;\n\t\t\tif (id?.startsWith(\"gpt-5.5\")) return 2.5;\n\t\t\treturn 2;\n\t\t}\n\t\tdefault:\n\t\t\treturn 1;\n\t}\n}\n\nfunction applyServiceTierPricing(\n\tusage: Usage,\n\tserviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n\tmodel: Pick<Model<\"openai-codex-responses\">, \"id\">,\n) {\n\tconst multiplier = getServiceTierCostMultiplier(serviceTier, model.id);\n\tif (multiplier === 1) return;\n\n\tusage.cost.input *= multiplier;\n\tusage.cost.output *= multiplier;\n\tusage.cost.cacheRead *= multiplier;\n\tusage.cost.cacheWrite *= multiplier;\n\tusage.cost.total = usage.cost.input + usage.cost.output + usage.cost.cacheRead + usage.cost.cacheWrite;\n}\n\nfunction resolveCodexServiceTier(\n\tresponseTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n\trequestedTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n): ResponseCreateParamsStreaming[\"service_tier\"] | undefined {\n\tif (responseTier === \"default\" && (requestedTier === \"flex\" || requestedTier === \"priority\")) {\n\t\treturn requestedTier;\n\t}\n\treturn responseTier ?? requestedTier;\n}\n\n// ============================================================================\n// Error Handling\n// ============================================================================\n\nasync function parseErrorResponse(response: Response): Promise<{ message: string; friendlyMessage?: string }> {\n\tconst raw = await response.text();\n\tlet message = raw || response.statusText || \"Request failed\";\n\tlet friendlyMessage: string | undefined;\n\n\ttry {\n\t\tconst parsed = JSON.parse(raw) as {\n\t\t\terror?: { code?: string; type?: string; message?: string; plan_type?: string; resets_at?: number };\n\t\t};\n\t\tconst err = parsed?.error;\n\t\tif (err) {\n\t\t\tconst code = err.code || err.type || \"\";\n\t\t\tif (/usage_limit_reached|usage_not_included|rate_limit_exceeded/i.test(code) || response.status === 429) {\n\t\t\t\tconst plan = err.plan_type ? ` (${err.plan_type.toLowerCase()} plan)` : \"\";\n\t\t\t\tconst mins = err.resets_at\n\t\t\t\t\t? Math.max(0, Math.round((err.resets_at * 1000 - Date.now()) / 60000))\n\t\t\t\t\t: undefined;\n\t\t\t\tconst when = mins !== undefined ? ` Try again in ~${mins} min.` : \"\";\n\t\t\t\tfriendlyMessage = `You have hit your ChatGPT usage limit${plan}.${when}`.trim();\n\t\t\t}\n\t\t\tmessage = err.message || friendlyMessage || message;\n\t\t}\n\t} catch {\n\t\t// Best-effort extraction of friendly error message — raw text is the fallback\n\t}\n\n\treturn { message, friendlyMessage };\n}\n\n// ============================================================================\n// Auth & Headers\n// ============================================================================\n\nfunction extractAccountId(token: string): string {\n\ttry {\n\t\tconst parts = token.split(\".\");\n\t\tif (parts.length !== 3) throw new Error(\"Invalid token\");\n\t\tconst payload = JSON.parse(atob(parts[1]));\n\t\tconst accountId = payload?.[JWT_CLAIM_PATH]?.chatgpt_account_id;\n\t\tif (!accountId) throw new Error(\"No account ID in token\");\n\t\treturn accountId;\n\t} catch {\n\t\t// Re-throw with unified message for any JWT decode failure\n\t\tthrow new Error(\"Failed to extract accountId from token\");\n\t}\n}\n\nfunction createCodexRequestId(): string {\n\tif (typeof globalThis.crypto?.randomUUID === \"function\") {\n\t\treturn globalThis.crypto.randomUUID();\n\t}\n\treturn `codex_${Date.now()}_${Math.random().toString(36).slice(2, 10)}`;\n}\n\nfunction buildBaseCodexHeaders(\n\tinitHeaders: Record<string, string> | undefined,\n\tadditionalHeaders: Record<string, string> | undefined,\n\taccountId: string,\n\ttoken: string,\n): Headers {\n\tconst headers = new Headers(initHeaders);\n\tfor (const [key, value] of Object.entries(additionalHeaders || {})) {\n\t\theaders.set(key, value);\n\t}\n\theaders.set(\"Authorization\", `Bearer ${token}`);\n\theaders.set(\"chatgpt-account-id\", accountId);\n\theaders.set(\"originator\", \"dreb\");\n\tconst userAgent = _os ? `dreb (${_os.platform()} ${_os.release()}; ${_os.arch()})` : \"dreb (browser)\";\n\theaders.set(\"User-Agent\", userAgent);\n\treturn headers;\n}\n\nfunction buildSSEHeaders(\n\tinitHeaders: Record<string, string> | undefined,\n\tadditionalHeaders: Record<string, string> | undefined,\n\taccountId: string,\n\ttoken: string,\n\tsessionId?: string,\n\tliteSessionId?: string,\n): Headers {\n\tconst headers = buildBaseCodexHeaders(initHeaders, additionalHeaders, accountId, token);\n\theaders.set(\"OpenAI-Beta\", \"responses=experimental\");\n\theaders.set(\"accept\", \"text/event-stream\");\n\theaders.set(\"content-type\", \"application/json\");\n\n\tif (liteSessionId) {\n\t\theaders.set(\"session-id\", liteSessionId);\n\t\theaders.set(\"x-session-affinity\", liteSessionId);\n\t\theaders.set(\"version\", CODEX_COMPATIBILITY_VERSION);\n\t\theaders.set(RESPONSES_LITE_HEADER, \"true\");\n\t} else if (sessionId) {\n\t\theaders.set(\"session_id\", sessionId);\n\t}\n\n\treturn headers;\n}\n\nfunction buildWebSocketHeaders(\n\tinitHeaders: Record<string, string> | undefined,\n\tadditionalHeaders: Record<string, string> | undefined,\n\taccountId: string,\n\ttoken: string,\n\trequestId: string,\n\tliteSessionId?: string,\n): Headers {\n\tconst headers = buildBaseCodexHeaders(initHeaders, additionalHeaders, accountId, token);\n\theaders.delete(\"accept\");\n\theaders.delete(\"content-type\");\n\theaders.delete(\"OpenAI-Beta\");\n\theaders.delete(\"openai-beta\");\n\theaders.set(\"OpenAI-Beta\", OPENAI_BETA_RESPONSES_WEBSOCKETS);\n\theaders.set(\"x-client-request-id\", requestId);\n\tif (liteSessionId) {\n\t\theaders.set(\"session-id\", liteSessionId);\n\t\theaders.set(\"x-session-affinity\", liteSessionId);\n\t\theaders.set(\"version\", CODEX_COMPATIBILITY_VERSION);\n\t\theaders.set(RESPONSES_LITE_HEADER, \"true\");\n\t} else {\n\t\theaders.set(\"session_id\", requestId);\n\t}\n\treturn headers;\n}\n"]}
+{"version":3,"file":"openai-codex-responses.d.ts","sourceRoot":"","sources":["../../src/providers/openai-codex-responses.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAEX,6BAA6B,EAG7B,MAAM,yCAAyC,CAAC;AAiBjD,OAAO,KAAK,EAKX,mBAAmB,EACnB,cAAc,EACd,aAAa,EAEb,MAAM,aAAa,CAAC;AAmCrB,MAAM,WAAW,2BAA4B,SAAQ,aAAa;IACjE,eAAe,CAAC,EAAE,MAAM,GAAG,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,GAAG,KAAK,CAAC;IACnF,gBAAgB,CAAC,EAAE,MAAM,GAAG,SAAS,GAAG,UAAU,GAAG,KAAK,GAAG,IAAI,GAAG,IAAI,CAAC;IACzE,aAAa,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAC;IAC1C,WAAW,CAAC,EAAE,6BAA6B,CAAC,cAAc,CAAC,CAAC;IAC5D,wFAAwF;IACxF,UAAU,CAAC,EAAE,CAAC,QAAQ,EAAE;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;KAAE,KAAK,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CACrG;AAoDD,eAAO,MAAM,0BAA0B,EAAE,cAAc,CAAC,wBAAwB,EAAE,2BAA2B,CA8L5G,CAAC;AAEF,eAAO,MAAM,gCAAgC,EAAE,cAAc,CAAC,wBAAwB,EAAE,mBAAmB,CAiB1G,CAAC","sourcesContent":["import type * as NodeOs from \"node:os\";\nimport type {\n\tTool as OpenAITool,\n\tResponseCreateParamsStreaming,\n\tResponseInput,\n\tResponseStreamEvent,\n} from \"openai/resources/responses/responses.js\";\n\n// NEVER convert to top-level runtime imports - breaks browser/Vite builds (web-ui)\nlet _os: typeof NodeOs | null = null;\n\ntype DynamicImport = (specifier: string) => Promise<unknown>;\n\nconst dynamicImport: DynamicImport = (specifier) => import(specifier);\nconst NODE_OS_SPECIFIER = \"node:\" + \"os\";\n\nif (typeof process !== \"undefined\" && (process.versions?.node || process.versions?.bun)) {\n\tdynamicImport(NODE_OS_SPECIFIER).then((m) => {\n\t\t_os = m as typeof NodeOs;\n\t});\n}\n\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tContext,\n\tModel,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tUsage,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { convertResponsesMessages, convertResponsesTools, processResponsesStream } from \"./openai-responses-shared.js\";\nimport { buildBaseOptions, resolveReasoningEffort } from \"./simple-options.js\";\n\n// ============================================================================\n// Configuration\n// ============================================================================\n\nconst DEFAULT_CODEX_BASE_URL = \"https://chatgpt.com/backend-api\";\nconst JWT_CLAIM_PATH = \"https://api.openai.com/auth\" as const;\nconst MAX_RETRIES = 3;\nconst BASE_DELAY_MS = 1000;\nconst CODEX_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\"]);\nconst CODEX_COMPATIBILITY_VERSION = \"0.144.0\";\nconst RESPONSES_LITE_MODELS = new Set([\"gpt-5.6-sol\", \"gpt-5.6-terra\", \"gpt-5.6-luna\"]);\nconst RESPONSES_LITE_HEADER = \"x-openai-internal-codex-responses-lite\";\nconst RESPONSES_LITE_CLIENT_METADATA = \"ws_request_header_x_openai_internal_codex_responses_lite\";\nconst CODEX_LITE_SESSION_TTL_MS = 30 * 60 * 1000;\nconst CODEX_LITE_MAX_SESSIONS = 1000;\nconst codexLiteSessionIds = new Map<string, { sessionId: string; lastUsedAt: number }>();\n\nconst CODEX_RESPONSE_STATUSES = new Set<CodexResponseStatus>([\n\t\"completed\",\n\t\"incomplete\",\n\t\"failed\",\n\t\"cancelled\",\n\t\"queued\",\n\t\"in_progress\",\n]);\n\n// ============================================================================\n// Types\n// ============================================================================\n\nexport interface OpenAICodexResponsesOptions extends StreamOptions {\n\treasoningEffort?: \"none\" | \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\" | \"max\";\n\treasoningSummary?: \"auto\" | \"concise\" | \"detailed\" | \"off\" | \"on\" | null;\n\ttextVerbosity?: \"low\" | \"medium\" | \"high\";\n\tserviceTier?: ResponseCreateParamsStreaming[\"service_tier\"];\n\t/** Optional callback invoked when the HTTP response headers are received (SSE only). */\n\tonResponse?: (response: { status: number; headers: Record<string, string> }) => void | Promise<void>;\n}\n\ntype CodexResponseStatus = \"completed\" | \"incomplete\" | \"failed\" | \"cancelled\" | \"queued\" | \"in_progress\";\n\ninterface RequestBody {\n\tmodel: string;\n\tstore?: boolean;\n\tstream?: boolean;\n\tinstructions?: string;\n\tinput?: ResponseInput;\n\ttools?: OpenAITool[];\n\ttool_choice?: \"auto\";\n\tparallel_tool_calls?: boolean;\n\ttemperature?: number;\n\treasoning?: { effort?: string; summary?: string; context?: \"all_turns\" };\n\ttext?: { verbosity?: string };\n\tinclude?: string[];\n\tprompt_cache_key?: string;\n\tservice_tier?: ResponseCreateParamsStreaming[\"service_tier\"];\n\tprevious_response_id?: string;\n\t[key: string]: unknown;\n}\n\n// ============================================================================\n// Retry Helpers\n// ============================================================================\n\nfunction isRetryableError(status: number, errorText: string): boolean {\n\tif (status === 429 || status === 500 || status === 502 || status === 503 || status === 504) {\n\t\treturn true;\n\t}\n\treturn /rate.?limit|overloaded|service.?unavailable|upstream.?connect|connection.?refused/i.test(errorText);\n}\n\nfunction sleep(ms: number, signal?: AbortSignal): Promise<void> {\n\treturn new Promise((resolve, reject) => {\n\t\tif (signal?.aborted) {\n\t\t\treject(new Error(\"Request was aborted\"));\n\t\t\treturn;\n\t\t}\n\t\tconst timeout = setTimeout(resolve, ms);\n\t\tsignal?.addEventListener(\"abort\", () => {\n\t\t\tclearTimeout(timeout);\n\t\t\treject(new Error(\"Request was aborted\"));\n\t\t});\n\t});\n}\n\n// ============================================================================\n// Main Stream Function\n// ============================================================================\n\nexport const streamOpenAICodexResponses: StreamFunction<\"openai-codex-responses\", OpenAICodexResponsesOptions> = (\n\tmodel: Model<\"openai-codex-responses\">,\n\tcontext: Context,\n\toptions?: OpenAICodexResponsesOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t(async () => {\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: \"openai-codex-responses\" as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider) || \"\";\n\t\t\tif (!apiKey) {\n\t\t\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t\t\t}\n\n\t\t\tconst accountId = extractAccountId(apiKey);\n\t\t\tlet body = buildRequestBody(model, context, options);\n\t\t\tconst websocketRequestId = options?.sessionId || createCodexRequestId();\n\t\t\tconst usesResponsesLite = isResponsesLiteModel(model.id);\n\t\t\tconst transport = options?.transport || \"sse\";\n\t\t\tconst liteSessionId = usesResponsesLite\n\t\t\t\t? options?.sessionId\n\t\t\t\t\t? getCodexLiteSessionId(options.sessionId)\n\t\t\t\t\t: createUuidV7()\n\t\t\t\t: undefined;\n\n\t\t\tif (usesResponsesLite) {\n\t\t\t\tbody = prepareResponsesLiteRequest(body, liteSessionId!);\n\t\t\t}\n\t\t\tconst nextBody = await options?.onPayload?.(body, model);\n\t\t\tif (nextBody !== undefined) {\n\t\t\t\tbody = nextBody as RequestBody;\n\t\t\t}\n\t\t\tif (usesResponsesLite) {\n\t\t\t\tvalidateResponsesLiteRequest(body, liteSessionId!);\n\t\t\t}\n\t\t\tconst sseHeaders = buildSSEHeaders(\n\t\t\t\tmodel.headers,\n\t\t\t\toptions?.headers,\n\t\t\t\taccountId,\n\t\t\t\tapiKey,\n\t\t\t\toptions?.sessionId,\n\t\t\t\tliteSessionId,\n\t\t\t);\n\t\t\tconst websocketHeaders = buildWebSocketHeaders(\n\t\t\t\tmodel.headers,\n\t\t\t\toptions?.headers,\n\t\t\t\taccountId,\n\t\t\t\tapiKey,\n\t\t\t\twebsocketRequestId,\n\t\t\t\tliteSessionId,\n\t\t\t);\n\t\t\tconst bodyJson = JSON.stringify(body);\n\n\t\t\tif (transport !== \"sse\") {\n\t\t\t\tlet websocketStarted = false;\n\t\t\t\ttry {\n\t\t\t\t\tawait processWebSocketStream(\n\t\t\t\t\t\tresolveCodexWebSocketUrl(model.baseUrl),\n\t\t\t\t\t\tbody,\n\t\t\t\t\t\twebsocketHeaders,\n\t\t\t\t\t\toutput,\n\t\t\t\t\t\tstream,\n\t\t\t\t\t\tmodel,\n\t\t\t\t\t\t() => {\n\t\t\t\t\t\t\twebsocketStarted = true;\n\t\t\t\t\t\t},\n\t\t\t\t\t\toptions,\n\t\t\t\t\t);\n\n\t\t\t\t\tif (options?.signal?.aborted) {\n\t\t\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t\t\t}\n\t\t\t\t\tstream.push({\n\t\t\t\t\t\ttype: \"done\",\n\t\t\t\t\t\treason: output.stopReason as \"stop\" | \"length\" | \"toolUse\",\n\t\t\t\t\t\tmessage: output,\n\t\t\t\t\t});\n\t\t\t\t\tstream.end();\n\t\t\t\t\treturn;\n\t\t\t\t} catch (error) {\n\t\t\t\t\tif (transport === \"websocket\" || websocketStarted) {\n\t\t\t\t\t\tthrow error;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Fetch with retry logic for rate limits and transient errors\n\t\t\tlet response: Response | undefined;\n\t\t\tlet lastError: Error | undefined;\n\n\t\t\tfor (let attempt = 0; attempt <= MAX_RETRIES; attempt++) {\n\t\t\t\tif (options?.signal?.aborted) {\n\t\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t\t}\n\n\t\t\t\ttry {\n\t\t\t\t\tresponse = await fetch(resolveCodexUrl(model.baseUrl), {\n\t\t\t\t\t\tmethod: \"POST\",\n\t\t\t\t\t\theaders: sseHeaders,\n\t\t\t\t\t\tbody: bodyJson,\n\t\t\t\t\t\tsignal: options?.signal,\n\t\t\t\t\t});\n\n\t\t\t\t\tif (response.ok) {\n\t\t\t\t\t\tbreak;\n\t\t\t\t\t}\n\n\t\t\t\t\tconst errorText = await response.text();\n\t\t\t\t\tif (attempt < MAX_RETRIES && isRetryableError(response.status, errorText)) {\n\t\t\t\t\t\tconst delayMs = BASE_DELAY_MS * 2 ** attempt;\n\t\t\t\t\t\tawait sleep(delayMs, options?.signal);\n\t\t\t\t\t\tcontinue;\n\t\t\t\t\t}\n\n\t\t\t\t\t// Parse error for friendly message on final attempt or non-retryable error\n\t\t\t\t\tconst fakeResponse = new Response(errorText, {\n\t\t\t\t\t\tstatus: response.status,\n\t\t\t\t\t\tstatusText: response.statusText,\n\t\t\t\t\t});\n\t\t\t\t\tconst info = await parseErrorResponse(fakeResponse);\n\t\t\t\t\tthrow new Error(info.friendlyMessage || info.message);\n\t\t\t\t} catch (error) {\n\t\t\t\t\tif (error instanceof Error) {\n\t\t\t\t\t\tif (error.name === \"AbortError\" || error.message === \"Request was aborted\") {\n\t\t\t\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tlastError = error instanceof Error ? error : new Error(String(error));\n\t\t\t\t\t// Network errors are retryable\n\t\t\t\t\tif (attempt < MAX_RETRIES && !lastError.message.includes(\"usage limit\")) {\n\t\t\t\t\t\tconst delayMs = BASE_DELAY_MS * 2 ** attempt;\n\t\t\t\t\t\tawait sleep(delayMs, options?.signal);\n\t\t\t\t\t\tcontinue;\n\t\t\t\t\t}\n\t\t\t\t\tthrow lastError;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (!response?.ok) {\n\t\t\t\tthrow lastError ?? new Error(\"Failed after retries\");\n\t\t\t}\n\n\t\t\t// Fire onResponse exactly once, after the retry loop succeeds.\n\t\t\t// Wrap in its own try/catch so callback errors don't affect stream processing.\n\t\t\ttry {\n\t\t\t\tawait options?.onResponse?.({ status: response.status, headers: headersToRecord(response.headers) });\n\t\t\t} catch {\n\t\t\t\t// Intentionally ignored — callback errors must not disrupt the stream\n\t\t\t}\n\n\t\t\tif (!response.body) {\n\t\t\t\tthrow new Error(\"No response body\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"start\", partial: output });\n\t\t\tawait processStream(response, output, stream, model, options);\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason as \"stop\" | \"length\" | \"toolUse\", message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : String(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimpleOpenAICodexResponses: StreamFunction<\"openai-codex-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"openai-codex-responses\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst base = buildBaseOptions(model, options, apiKey);\n\tconst reasoningEffort = resolveReasoningEffort(model, options?.reasoning);\n\n\treturn streamOpenAICodexResponses(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies OpenAICodexResponsesOptions);\n};\n\n// ============================================================================\n// Request Building\n// ============================================================================\n\nfunction buildRequestBody(\n\tmodel: Model<\"openai-codex-responses\">,\n\tcontext: Context,\n\toptions?: OpenAICodexResponsesOptions,\n): RequestBody {\n\tconst messages = convertResponsesMessages(model, context, CODEX_TOOL_CALL_PROVIDERS, {\n\t\tincludeSystemPrompt: false,\n\t});\n\n\tconst body: RequestBody = {\n\t\tmodel: model.id,\n\t\tstore: false,\n\t\tstream: true,\n\t\tinstructions: context.systemPrompt,\n\t\tinput: messages,\n\t\ttext: { verbosity: options?.textVerbosity || \"low\" },\n\t\tinclude: [\"reasoning.encrypted_content\"],\n\t\tprompt_cache_key: options?.sessionId,\n\t\ttool_choice: \"auto\",\n\t\tparallel_tool_calls: true,\n\t};\n\n\tif (options?.temperature !== undefined) {\n\t\tbody.temperature = options.temperature;\n\t}\n\n\tif (context.tools) {\n\t\tbody.tools = convertResponsesTools(context.tools, { strict: null });\n\t}\n\n\tif (options?.reasoningEffort !== undefined) {\n\t\tbody.reasoning = {\n\t\t\teffort: clampReasoningEffort(model.id, options.reasoningEffort),\n\t\t\tsummary: options.reasoningSummary ?? \"auto\",\n\t\t};\n\t}\n\n\tif (options?.serviceTier !== undefined) {\n\t\tbody.service_tier = options.serviceTier;\n\t}\n\n\treturn body;\n}\n\nfunction isResponsesLiteModel(modelId: string): boolean {\n\tconst id = modelId.includes(\"/\") ? modelId.split(\"/\").pop()! : modelId;\n\treturn RESPONSES_LITE_MODELS.has(id);\n}\n\nfunction getCodexLiteSessionId(sourceSessionId: string): string {\n\tconst now = Date.now();\n\tfor (const [key, entry] of codexLiteSessionIds) {\n\t\tif (now - entry.lastUsedAt > CODEX_LITE_SESSION_TTL_MS) {\n\t\t\tcodexLiteSessionIds.delete(key);\n\t\t}\n\t}\n\n\tconst existing = codexLiteSessionIds.get(sourceSessionId);\n\tif (existing) {\n\t\texisting.lastUsedAt = now;\n\t\tcodexLiteSessionIds.delete(sourceSessionId);\n\t\tcodexLiteSessionIds.set(sourceSessionId, existing);\n\t\treturn existing.sessionId;\n\t}\n\n\twhile (codexLiteSessionIds.size >= CODEX_LITE_MAX_SESSIONS) {\n\t\tconst oldestKey = codexLiteSessionIds.keys().next().value;\n\t\tif (oldestKey === undefined) break;\n\t\tcodexLiteSessionIds.delete(oldestKey);\n\t}\n\n\tconst sessionId = createUuidV7();\n\tcodexLiteSessionIds.set(sourceSessionId, { sessionId, lastUsedAt: now });\n\treturn sessionId;\n}\n\nfunction createUuidV7(): string {\n\tconst timestamp = Date.now().toString(16).padStart(12, \"0\").slice(-12);\n\tconst random = new Uint8Array(10);\n\tif (typeof globalThis.crypto?.getRandomValues === \"function\") {\n\t\tglobalThis.crypto.getRandomValues(random);\n\t} else {\n\t\tfor (let index = 0; index < random.length; index++) {\n\t\t\trandom[index] = Math.floor(Math.random() * 256);\n\t\t}\n\t}\n\tconst randomHex = Array.from(random, (byte) => byte.toString(16).padStart(2, \"0\")).join(\"\");\n\tconst variant = (8 + (Number.parseInt(randomHex[3]!, 16) & 0x3)).toString(16);\n\treturn `${timestamp.slice(0, 8)}-${timestamp.slice(8)}-7${randomHex.slice(0, 3)}-${variant}${randomHex.slice(4, 7)}-${randomHex.slice(7, 19)}`;\n}\n\nfunction prepareResponsesLiteRequest(body: RequestBody, sessionId: string): RequestBody {\n\tif (!Array.isArray(body.input)) {\n\t\tthrow new Error(\"Responses Lite requires an input array\");\n\t}\n\tif (body.tools !== undefined && !Array.isArray(body.tools)) {\n\t\tthrow new Error(\"Responses Lite requires a tools array\");\n\t}\n\tif (body.instructions !== undefined && typeof body.instructions !== \"string\") {\n\t\tthrow new Error(\"Responses Lite requires string instructions\");\n\t}\n\n\tconst developerItems: unknown[] = [{ type: \"additional_tools\", role: \"developer\", tools: body.tools ?? [] }];\n\tif (body.instructions) {\n\t\tdeveloperItems.push({\n\t\t\ttype: \"message\",\n\t\t\trole: \"developer\",\n\t\t\tcontent: [{ type: \"input_text\", text: body.instructions }],\n\t\t});\n\t}\n\n\tconst input = [...developerItems, ...(body.input ?? [])] as ResponseInput;\n\tstripInputImageDetails(input);\n\tconst { tools: _tools, instructions: _instructions, reasoning, ...request } = body;\n\treturn {\n\t\t...request,\n\t\tinput,\n\t\ttool_choice: \"auto\",\n\t\tparallel_tool_calls: false,\n\t\tprompt_cache_key: sessionId,\n\t\treasoning: { ...reasoning, context: \"all_turns\" },\n\t};\n}\n\nfunction validateResponsesLiteRequest(body: RequestBody, sessionId: string): void {\n\tif (!body || typeof body !== \"object\" || Array.isArray(body)) {\n\t\tthrow new Error(\"Responses Lite onPayload must return an object payload\");\n\t}\n\n\tconst payload = body as Record<string, unknown>;\n\tif (Object.hasOwn(payload, \"instructions\")) {\n\t\tthrow new Error(\"Responses Lite does not support top-level instructions\");\n\t}\n\tif (Object.hasOwn(payload, \"tools\")) {\n\t\tthrow new Error(\"Responses Lite does not support top-level tools\");\n\t}\n\tif (!Array.isArray(payload.input)) {\n\t\tthrow new Error(\"Responses Lite requires an input array\");\n\t}\n\tif (payload.tool_choice !== \"auto\") {\n\t\tthrow new Error(\"Responses Lite requires tool_choice to be auto\");\n\t}\n\tif (payload.parallel_tool_calls !== false) {\n\t\tthrow new Error(\"Responses Lite requires parallel_tool_calls to be false\");\n\t}\n\tif (payload.prompt_cache_key !== sessionId) {\n\t\tthrow new Error(\"Responses Lite requires prompt_cache_key to match its session ID\");\n\t}\n\tif (\n\t\t!payload.reasoning ||\n\t\ttypeof payload.reasoning !== \"object\" ||\n\t\tArray.isArray(payload.reasoning) ||\n\t\t(payload.reasoning as Record<string, unknown>).context !== \"all_turns\"\n\t) {\n\t\tthrow new Error(\"Responses Lite requires reasoning.context to be all_turns\");\n\t}\n\n\tconst additionalTools = payload.input[0];\n\tif (\n\t\t!additionalTools ||\n\t\ttypeof additionalTools !== \"object\" ||\n\t\tArray.isArray(additionalTools) ||\n\t\t(additionalTools as Record<string, unknown>).type !== \"additional_tools\" ||\n\t\t(additionalTools as Record<string, unknown>).role !== \"developer\" ||\n\t\t!Array.isArray((additionalTools as Record<string, unknown>).tools)\n\t) {\n\t\tthrow new Error(\"Responses Lite requires a developer additional_tools input item\");\n\t}\n\tif (containsInputImageDetail(payload.input)) {\n\t\tthrow new Error(\"Responses Lite does not support input_image detail\");\n\t}\n}\n\nfunction stripInputImageDetails(value: unknown): void {\n\tif (Array.isArray(value)) {\n\t\tfor (const item of value) stripInputImageDetails(item);\n\t\treturn;\n\t}\n\tif (!value || typeof value !== \"object\") return;\n\n\tconst record = value as Record<string, unknown>;\n\tif (record.type === \"input_image\") {\n\t\tdelete record.detail;\n\t}\n\tfor (const nested of Object.values(record)) stripInputImageDetails(nested);\n}\n\nfunction containsInputImageDetail(value: unknown): boolean {\n\tif (Array.isArray(value)) return value.some(containsInputImageDetail);\n\tif (!value || typeof value !== \"object\") return false;\n\n\tconst record = value as Record<string, unknown>;\n\tif (record.type === \"input_image\" && Object.hasOwn(record, \"detail\")) return true;\n\treturn Object.values(record).some(containsInputImageDetail);\n}\n\nfunction clampReasoningEffort(modelId: string, effort: string): string {\n\tconst id = modelId.includes(\"/\") ? modelId.split(\"/\").pop()! : modelId;\n\tif (\n\t\t(id.startsWith(\"gpt-5.2\") ||\n\t\t\tid.startsWith(\"gpt-5.3\") ||\n\t\t\tid.startsWith(\"gpt-5.4\") ||\n\t\t\tid.startsWith(\"gpt-5.5\") ||\n\t\t\tid.startsWith(\"gpt-5.6\")) &&\n\t\teffort === \"minimal\"\n\t)\n\t\treturn \"low\";\n\tif (id === \"gpt-5.1\" && effort === \"xhigh\") return \"high\";\n\tif (id === \"gpt-5.1-codex-mini\") return effort === \"high\" || effort === \"xhigh\" ? \"high\" : \"medium\";\n\treturn effort;\n}\n\nfunction resolveCodexUrl(baseUrl?: string): string {\n\tconst raw = baseUrl && baseUrl.trim().length > 0 ? baseUrl : DEFAULT_CODEX_BASE_URL;\n\tconst normalized = raw.replace(/\\/+$/, \"\");\n\tif (normalized.endsWith(\"/codex/responses\")) return normalized;\n\tif (normalized.endsWith(\"/codex\")) return `${normalized}/responses`;\n\treturn `${normalized}/codex/responses`;\n}\n\nfunction resolveCodexWebSocketUrl(baseUrl?: string): string {\n\tconst url = new URL(resolveCodexUrl(baseUrl));\n\tif (url.protocol === \"https:\") url.protocol = \"wss:\";\n\tif (url.protocol === \"http:\") url.protocol = \"ws:\";\n\treturn url.toString();\n}\n\n// ============================================================================\n// Response Processing\n// ============================================================================\n\nasync function processStream(\n\tresponse: Response,\n\toutput: AssistantMessage,\n\tstream: AssistantMessageEventStream,\n\tmodel: Model<\"openai-codex-responses\">,\n\toptions?: OpenAICodexResponsesOptions,\n): Promise<void> {\n\tawait processResponsesStream(mapCodexEvents(parseSSE(response, options)), output, stream, model, {\n\t\tserviceTier: options?.serviceTier,\n\t\tresolveServiceTier: resolveCodexServiceTier,\n\t\tapplyServiceTierPricing: (usage, serviceTier) => applyServiceTierPricing(usage, serviceTier, model),\n\t\tonWarning: options?.onWarning,\n\t});\n}\n\nasync function* mapCodexEvents(events: AsyncIterable<Record<string, unknown>>): AsyncGenerator<ResponseStreamEvent> {\n\tfor await (const event of events) {\n\t\tconst type = typeof event.type === \"string\" ? event.type : undefined;\n\t\tif (!type) continue;\n\n\t\tif (type === \"error\") {\n\t\t\tconst code = (event as { code?: string }).code || \"\";\n\t\t\tconst message = (event as { message?: string }).message || \"\";\n\t\t\tthrow new Error(`Codex error: ${message || code || JSON.stringify(event)}`);\n\t\t}\n\n\t\tif (type === \"response.failed\") {\n\t\t\tconst msg = (event as { response?: { error?: { message?: string } } }).response?.error?.message;\n\t\t\tthrow new Error(msg || \"Codex response failed\");\n\t\t}\n\n\t\tif (type === \"response.done\" || type === \"response.completed\" || type === \"response.incomplete\") {\n\t\t\tconst response = (event as { response?: { status?: unknown } }).response;\n\t\t\tconst normalizedResponse = response\n\t\t\t\t? { ...response, status: normalizeCodexStatus(response.status) }\n\t\t\t\t: response;\n\t\t\tyield { ...event, type: \"response.completed\", response: normalizedResponse } as ResponseStreamEvent;\n\t\t\treturn;\n\t\t}\n\n\t\tyield event as unknown as ResponseStreamEvent;\n\t}\n}\n\nfunction normalizeCodexStatus(status: unknown): CodexResponseStatus | undefined {\n\tif (typeof status !== \"string\") return undefined;\n\treturn CODEX_RESPONSE_STATUSES.has(status as CodexResponseStatus) ? (status as CodexResponseStatus) : undefined;\n}\n\n// ============================================================================\n// SSE Parsing\n// ============================================================================\n\nasync function* parseSSE(\n\tresponse: Response,\n\toptions?: OpenAICodexResponsesOptions,\n): AsyncGenerator<Record<string, unknown>> {\n\tif (!response.body) return;\n\n\tconst reader = response.body.getReader();\n\tconst decoder = new TextDecoder();\n\tlet sseParseErrors = 0;\n\tlet buffer = \"\";\n\n\ttry {\n\t\twhile (true) {\n\t\t\tconst { done, value } = await reader.read();\n\t\t\tif (done) break;\n\t\t\tbuffer += decoder.decode(value, { stream: true });\n\n\t\t\tlet idx = buffer.indexOf(\"\\n\\n\");\n\t\t\twhile (idx !== -1) {\n\t\t\t\tconst chunk = buffer.slice(0, idx);\n\t\t\t\tbuffer = buffer.slice(idx + 2);\n\n\t\t\t\tconst dataLines = chunk\n\t\t\t\t\t.split(\"\\n\")\n\t\t\t\t\t.filter((l) => l.startsWith(\"data:\"))\n\t\t\t\t\t.map((l) => l.slice(5).trim());\n\t\t\t\tif (dataLines.length > 0) {\n\t\t\t\t\tconst data = dataLines.join(\"\\n\").trim();\n\t\t\t\t\tif (data && data !== \"[DONE]\") {\n\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\tyield JSON.parse(data);\n\t\t\t\t\t\t} catch {\n\t\t\t\t\t\t\t// Malformed SSE event — track for end-of-stream summary\n\t\t\t\t\t\t\tsseParseErrors++;\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tidx = buffer.indexOf(\"\\n\\n\");\n\t\t\t}\n\t\t}\n\t} finally {\n\t\tif (sseParseErrors > 0) {\n\t\t\toptions?.onWarning?.(\n\t\t\t\t\"sse_parse_error\",\n\t\t\t\t`${sseParseErrors} malformed SSE event(s) were dropped during streaming. Response content may be incomplete.`,\n\t\t\t);\n\t\t}\n\t\ttry {\n\t\t\tawait reader.cancel();\n\t\t} catch {\n\t\t\t// Reader may already be closed — cleanup errors are safe to ignore\n\t\t}\n\t\ttry {\n\t\t\treader.releaseLock();\n\t\t} catch {\n\t\t\t// Reader may already be closed — cleanup errors are safe to ignore\n\t\t}\n\t}\n}\n\n// ============================================================================\n// WebSocket Parsing\n// ============================================================================\n\nconst OPENAI_BETA_RESPONSES_WEBSOCKETS = \"responses_websockets=2026-02-06\";\nconst SESSION_WEBSOCKET_CACHE_TTL_MS = 5 * 60 * 1000;\n\ntype WebSocketEventType = \"open\" | \"message\" | \"error\" | \"close\";\ntype WebSocketListener = (event: unknown) => void;\n\ninterface WebSocketLike {\n\tclose(code?: number, reason?: string): void;\n\tsend(data: string): void;\n\taddEventListener(type: WebSocketEventType, listener: WebSocketListener): void;\n\tremoveEventListener(type: WebSocketEventType, listener: WebSocketListener): void;\n}\n\ninterface CachedWebSocketContinuationState {\n\tlastRequestBody: RequestBody;\n\tlastResponseId: string;\n\tlastResponseItems: ResponseInput;\n}\n\ninterface CachedWebSocketConnection {\n\tsocket: WebSocketLike;\n\tbusy: boolean;\n\tidleTimer?: ReturnType<typeof setTimeout>;\n\tcontinuation?: CachedWebSocketContinuationState;\n}\n\nconst websocketSessionCache = new Map<string, CachedWebSocketConnection>();\n\ntype WebSocketConstructor = new (\n\turl: string,\n\tprotocols?: string | string[] | { headers?: Record<string, string> },\n) => WebSocketLike;\n\nfunction getWebSocketConstructor(): WebSocketConstructor | null {\n\tconst ctor = (globalThis as { WebSocket?: unknown }).WebSocket;\n\tif (typeof ctor !== \"function\") return null;\n\treturn ctor as unknown as WebSocketConstructor;\n}\n\nfunction headersToRecord(headers: Headers): Record<string, string> {\n\tconst out: Record<string, string> = {};\n\tfor (const [key, value] of headers.entries()) {\n\t\tout[key] = value;\n\t}\n\treturn out;\n}\n\nfunction getWebSocketReadyState(socket: WebSocketLike): number | undefined {\n\tconst readyState = (socket as { readyState?: unknown }).readyState;\n\treturn typeof readyState === \"number\" ? readyState : undefined;\n}\n\nfunction isWebSocketReusable(socket: WebSocketLike): boolean {\n\tconst readyState = getWebSocketReadyState(socket);\n\t// If readyState is unavailable, assume the runtime keeps it open/reusable.\n\treturn readyState === undefined || readyState === 1;\n}\n\nfunction closeWebSocketSilently(socket: WebSocketLike, code = 1000, reason = \"done\"): void {\n\ttry {\n\t\tsocket.close(code, reason);\n\t} catch {\n\t\t// Intentionally silent — socket may already be closed\n\t}\n}\n\nfunction scheduleSessionWebSocketExpiry(sessionId: string, entry: CachedWebSocketConnection): void {\n\tif (entry.idleTimer) {\n\t\tclearTimeout(entry.idleTimer);\n\t}\n\tentry.idleTimer = setTimeout(() => {\n\t\tif (entry.busy) return;\n\t\tcloseWebSocketSilently(entry.socket, 1000, \"idle_timeout\");\n\t\twebsocketSessionCache.delete(sessionId);\n\t}, SESSION_WEBSOCKET_CACHE_TTL_MS);\n}\n\nasync function connectWebSocket(url: string, headers: Headers, signal?: AbortSignal): Promise<WebSocketLike> {\n\tconst WebSocketCtor = getWebSocketConstructor();\n\tif (!WebSocketCtor) {\n\t\tthrow new Error(\"WebSocket transport is not available in this runtime\");\n\t}\n\n\tconst wsHeaders = headersToRecord(headers);\n\tdelete wsHeaders[\"OpenAI-Beta\"];\n\n\treturn new Promise<WebSocketLike>((resolve, reject) => {\n\t\tlet settled = false;\n\t\tlet socket: WebSocketLike;\n\n\t\ttry {\n\t\t\tsocket = new WebSocketCtor(url, { headers: wsHeaders });\n\t\t} catch (error) {\n\t\t\treject(error instanceof Error ? error : new Error(String(error)));\n\t\t\treturn;\n\t\t}\n\n\t\tconst onOpen: WebSocketListener = () => {\n\t\t\tif (settled) return;\n\t\t\tsettled = true;\n\t\t\tcleanup();\n\t\t\tresolve(socket);\n\t\t};\n\t\tconst onError: WebSocketListener = (event) => {\n\t\t\tconst error = extractWebSocketError(event);\n\t\t\tif (settled) return;\n\t\t\tsettled = true;\n\t\t\tcleanup();\n\t\t\treject(error);\n\t\t};\n\t\tconst onClose: WebSocketListener = (event) => {\n\t\t\tconst error = extractWebSocketCloseError(event);\n\t\t\tif (settled) return;\n\t\t\tsettled = true;\n\t\t\tcleanup();\n\t\t\treject(error);\n\t\t};\n\t\tconst onAbort = () => {\n\t\t\tif (settled) return;\n\t\t\tsettled = true;\n\t\t\tcleanup();\n\t\t\tsocket.close(1000, \"aborted\");\n\t\t\treject(new Error(\"Request was aborted\"));\n\t\t};\n\n\t\tconst cleanup = () => {\n\t\t\tsocket.removeEventListener(\"open\", onOpen);\n\t\t\tsocket.removeEventListener(\"error\", onError);\n\t\t\tsocket.removeEventListener(\"close\", onClose);\n\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t};\n\n\t\tsocket.addEventListener(\"open\", onOpen);\n\t\tsocket.addEventListener(\"error\", onError);\n\t\tsocket.addEventListener(\"close\", onClose);\n\t\tsignal?.addEventListener(\"abort\", onAbort);\n\t});\n}\n\nasync function acquireWebSocket(\n\turl: string,\n\theaders: Headers,\n\tsessionId: string | undefined,\n\tsignal?: AbortSignal,\n): Promise<{\n\tsocket: WebSocketLike;\n\tentry?: CachedWebSocketConnection;\n\trelease: (options?: { keep?: boolean }) => void;\n}> {\n\tif (!sessionId) {\n\t\tconst socket = await connectWebSocket(url, headers, signal);\n\t\treturn {\n\t\t\tsocket,\n\t\t\trelease: () => {\n\t\t\t\tcloseWebSocketSilently(socket);\n\t\t\t},\n\t\t};\n\t}\n\n\tconst cached = websocketSessionCache.get(sessionId);\n\tif (cached) {\n\t\tif (cached.idleTimer) {\n\t\t\tclearTimeout(cached.idleTimer);\n\t\t\tcached.idleTimer = undefined;\n\t\t}\n\t\tif (!cached.busy && isWebSocketReusable(cached.socket)) {\n\t\t\tcached.busy = true;\n\t\t\treturn {\n\t\t\t\tsocket: cached.socket,\n\t\t\t\tentry: cached,\n\t\t\t\trelease: ({ keep } = {}) => {\n\t\t\t\t\tif (!keep || !isWebSocketReusable(cached.socket)) {\n\t\t\t\t\t\tcloseWebSocketSilently(cached.socket);\n\t\t\t\t\t\twebsocketSessionCache.delete(sessionId);\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\tcached.busy = false;\n\t\t\t\t\tscheduleSessionWebSocketExpiry(sessionId, cached);\n\t\t\t\t},\n\t\t\t};\n\t\t}\n\t\tif (cached.busy) {\n\t\t\tconst socket = await connectWebSocket(url, headers, signal);\n\t\t\treturn {\n\t\t\t\tsocket,\n\t\t\t\trelease: () => {\n\t\t\t\t\tcloseWebSocketSilently(socket);\n\t\t\t\t},\n\t\t\t};\n\t\t}\n\t\tif (!isWebSocketReusable(cached.socket)) {\n\t\t\tcloseWebSocketSilently(cached.socket);\n\t\t\twebsocketSessionCache.delete(sessionId);\n\t\t}\n\t}\n\n\tconst socket = await connectWebSocket(url, headers, signal);\n\tconst entry: CachedWebSocketConnection = { socket, busy: true };\n\twebsocketSessionCache.set(sessionId, entry);\n\treturn {\n\t\tsocket,\n\t\tentry,\n\t\trelease: ({ keep } = {}) => {\n\t\t\tif (!keep || !isWebSocketReusable(entry.socket)) {\n\t\t\t\tcloseWebSocketSilently(entry.socket);\n\t\t\t\tif (entry.idleTimer) clearTimeout(entry.idleTimer);\n\t\t\t\tif (websocketSessionCache.get(sessionId) === entry) {\n\t\t\t\t\twebsocketSessionCache.delete(sessionId);\n\t\t\t\t}\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tentry.busy = false;\n\t\t\tscheduleSessionWebSocketExpiry(sessionId, entry);\n\t\t},\n\t};\n}\n\nfunction extractWebSocketError(event: unknown): Error {\n\tif (event && typeof event === \"object\" && \"message\" in event) {\n\t\tconst message = (event as { message?: unknown }).message;\n\t\tif (typeof message === \"string\" && message.length > 0) {\n\t\t\treturn new Error(message);\n\t\t}\n\t}\n\treturn new Error(\"WebSocket error\");\n}\n\nfunction extractWebSocketCloseError(event: unknown): Error {\n\tif (event && typeof event === \"object\") {\n\t\tconst code = \"code\" in event ? (event as { code?: unknown }).code : undefined;\n\t\tconst reason = \"reason\" in event ? (event as { reason?: unknown }).reason : undefined;\n\t\tconst codeText = typeof code === \"number\" ? ` ${code}` : \"\";\n\t\tconst reasonText = typeof reason === \"string\" && reason.length > 0 ? ` ${reason}` : \"\";\n\t\treturn new Error(`WebSocket closed${codeText}${reasonText}`.trim());\n\t}\n\treturn new Error(\"WebSocket closed\");\n}\n\nasync function decodeWebSocketData(data: unknown): Promise<string | null> {\n\tif (typeof data === \"string\") return data;\n\tif (data instanceof ArrayBuffer) {\n\t\treturn new TextDecoder().decode(new Uint8Array(data));\n\t}\n\tif (ArrayBuffer.isView(data)) {\n\t\tconst view = data as ArrayBufferView;\n\t\treturn new TextDecoder().decode(new Uint8Array(view.buffer, view.byteOffset, view.byteLength));\n\t}\n\tif (data && typeof data === \"object\" && \"arrayBuffer\" in data) {\n\t\tconst blobLike = data as { arrayBuffer: () => Promise<ArrayBuffer> };\n\t\tconst arrayBuffer = await blobLike.arrayBuffer();\n\t\treturn new TextDecoder().decode(new Uint8Array(arrayBuffer));\n\t}\n\treturn null;\n}\n\nasync function* parseWebSocket(\n\tsocket: WebSocketLike,\n\tsignal?: AbortSignal,\n\toptions?: OpenAICodexResponsesOptions,\n): AsyncGenerator<Record<string, unknown>> {\n\tconst queue: Record<string, unknown>[] = [];\n\tlet pending: (() => void) | null = null;\n\tlet done = false;\n\tlet failed: Error | null = null;\n\tlet sawCompletion = false;\n\n\tconst wake = () => {\n\t\tif (!pending) return;\n\t\tconst resolve = pending;\n\t\tpending = null;\n\t\tresolve();\n\t};\n\n\tconst onMessage: WebSocketListener = (event) => {\n\t\tvoid (async () => {\n\t\t\tif (!event || typeof event !== \"object\" || !(\"data\" in event)) return;\n\t\t\tlet text: string | null | undefined;\n\t\t\ttry {\n\t\t\t\ttext = await decodeWebSocketData((event as { data?: unknown }).data);\n\t\t\t\tif (!text) {\n\t\t\t\t\twake();\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tconst parsed = JSON.parse(text) as Record<string, unknown>;\n\t\t\t\tconst type = typeof parsed.type === \"string\" ? parsed.type : \"\";\n\t\t\t\tif (type === \"response.completed\" || type === \"response.done\" || type === \"response.incomplete\") {\n\t\t\t\t\tsawCompletion = true;\n\t\t\t\t\tdone = true;\n\t\t\t\t}\n\t\t\t\tqueue.push(parsed);\n\t\t\t} catch (e) {\n\t\t\t\t// Malformed WebSocket message — warn but don't drop wake()\n\t\t\t\tif (text === undefined) {\n\t\t\t\t\toptions?.onWarning?.(\n\t\t\t\t\t\t\"ws_decode_error\",\n\t\t\t\t\t\t`Failed to decode WebSocket message: ${e instanceof Error ? e.message : String(e)}`,\n\t\t\t\t\t);\n\t\t\t\t} else {\n\t\t\t\t\toptions?.onWarning?.(\n\t\t\t\t\t\t\"ws_parse_error\",\n\t\t\t\t\t\t`Malformed WebSocket message dropped: ${text?.slice(0, 200) ?? \"\"}`,\n\t\t\t\t\t);\n\t\t\t\t}\n\t\t\t} finally {\n\t\t\t\t// Always wake the consumer — even on parse failure — to prevent stream hangs\n\t\t\t\twake();\n\t\t\t}\n\t\t})();\n\t};\n\n\tconst onError: WebSocketListener = (event) => {\n\t\tfailed = extractWebSocketError(event);\n\t\tdone = true;\n\t\twake();\n\t};\n\n\tconst onClose: WebSocketListener = (event) => {\n\t\tif (sawCompletion) {\n\t\t\tdone = true;\n\t\t\twake();\n\t\t\treturn;\n\t\t}\n\t\tif (!failed) {\n\t\t\tconst closeError = extractWebSocketCloseError(event);\n\t\t\tfailed = new Error(`WebSocket stream closed before response.completed (${closeError.message})`);\n\t\t}\n\t\tdone = true;\n\t\twake();\n\t};\n\n\tconst onAbort = () => {\n\t\tfailed = new Error(\"Request was aborted\");\n\t\tdone = true;\n\t\twake();\n\t};\n\n\tsocket.addEventListener(\"message\", onMessage);\n\tsocket.addEventListener(\"error\", onError);\n\tsocket.addEventListener(\"close\", onClose);\n\tsignal?.addEventListener(\"abort\", onAbort);\n\n\ttry {\n\t\twhile (true) {\n\t\t\tif (signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\t\t\tif (queue.length > 0) {\n\t\t\t\tyield queue.shift()!;\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (done) break;\n\t\t\tawait new Promise<void>((resolve) => {\n\t\t\t\tpending = resolve;\n\t\t\t});\n\t\t}\n\n\t\tif (failed) {\n\t\t\tthrow failed;\n\t\t}\n\t\tif (!sawCompletion) {\n\t\t\tthrow new Error(\"WebSocket stream closed before response.completed\");\n\t\t}\n\t} finally {\n\t\tsocket.removeEventListener(\"message\", onMessage);\n\t\tsocket.removeEventListener(\"error\", onError);\n\t\tsocket.removeEventListener(\"close\", onClose);\n\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t}\n}\n\nfunction requestBodyWithoutInput(body: RequestBody): RequestBody {\n\tconst { input: _input, previous_response_id: _previousResponseId, ...rest } = body;\n\treturn rest;\n}\n\nfunction responseInputsEqual(a: ResponseInput | undefined, b: ResponseInput | undefined): boolean {\n\treturn JSON.stringify(a ?? []) === JSON.stringify(b ?? []);\n}\n\nfunction requestBodiesMatchExceptInput(a: RequestBody, b: RequestBody): boolean {\n\treturn JSON.stringify(requestBodyWithoutInput(a)) === JSON.stringify(requestBodyWithoutInput(b));\n}\n\nfunction getCachedWebSocketInputDelta(\n\tbody: RequestBody,\n\tcontinuation: CachedWebSocketContinuationState,\n): ResponseInput | undefined {\n\tif (!requestBodiesMatchExceptInput(body, continuation.lastRequestBody)) {\n\t\treturn undefined;\n\t}\n\n\tconst currentInput = body.input ?? [];\n\tconst baseline = [...(continuation.lastRequestBody.input ?? []), ...continuation.lastResponseItems];\n\tif (currentInput.length < baseline.length) {\n\t\treturn undefined;\n\t}\n\n\tconst prefix = currentInput.slice(0, baseline.length);\n\tif (!responseInputsEqual(prefix, baseline)) {\n\t\treturn undefined;\n\t}\n\n\treturn currentInput.slice(baseline.length);\n}\n\nfunction buildCachedWebSocketRequestBody(entry: CachedWebSocketConnection, body: RequestBody): RequestBody {\n\tconst continuation = entry.continuation;\n\tif (!continuation) {\n\t\treturn body;\n\t}\n\n\tconst delta = getCachedWebSocketInputDelta(body, continuation);\n\tif (!delta || !continuation.lastResponseId) {\n\t\tentry.continuation = undefined;\n\t\treturn body;\n\t}\n\n\treturn {\n\t\t...body,\n\t\tprevious_response_id: continuation.lastResponseId,\n\t\tinput: delta,\n\t};\n}\n\nfunction addResponsesLiteWebSocketMetadata(body: RequestBody): RequestBody {\n\tconst clientMetadata = body.client_metadata;\n\treturn {\n\t\t...body,\n\t\tclient_metadata: {\n\t\t\t...(clientMetadata && typeof clientMetadata === \"object\" && !Array.isArray(clientMetadata)\n\t\t\t\t? clientMetadata\n\t\t\t\t: {}),\n\t\t\t[RESPONSES_LITE_CLIENT_METADATA]: \"true\",\n\t\t},\n\t};\n}\n\nasync function processWebSocketStream(\n\turl: string,\n\tbody: RequestBody,\n\theaders: Headers,\n\toutput: AssistantMessage,\n\tstream: AssistantMessageEventStream,\n\tmodel: Model<\"openai-codex-responses\">,\n\tonStart: () => void,\n\toptions?: OpenAICodexResponsesOptions,\n): Promise<void> {\n\tconst cacheSessionId = options?.sessionId\n\t\t? `${isResponsesLiteModel(model.id) ? \"responses-lite\" : \"legacy\"}:${options.sessionId}`\n\t\t: undefined;\n\tconst { socket, entry, release } = await acquireWebSocket(url, headers, cacheSessionId, options?.signal);\n\tlet keepConnection = true;\n\tconst fullBody = body;\n\tconst requestBody = entry ? buildCachedWebSocketRequestBody(entry, fullBody) : fullBody;\n\tconst wireBody = isResponsesLiteModel(model.id) ? addResponsesLiteWebSocketMetadata(requestBody) : requestBody;\n\ttry {\n\t\tsocket.send(JSON.stringify({ type: \"response.create\", ...wireBody }));\n\t\tonStart();\n\t\tstream.push({ type: \"start\", partial: output });\n\t\tawait processResponsesStream(\n\t\t\tmapCodexEvents(parseWebSocket(socket, options?.signal, options)),\n\t\t\toutput,\n\t\t\tstream,\n\t\t\tmodel,\n\t\t\t{\n\t\t\t\tserviceTier: options?.serviceTier,\n\t\t\t\tresolveServiceTier: resolveCodexServiceTier,\n\t\t\t\tapplyServiceTierPricing: (usage, serviceTier) => applyServiceTierPricing(usage, serviceTier, model),\n\t\t\t\tonWarning: options?.onWarning,\n\t\t\t},\n\t\t);\n\t\tif (entry && output.responseId && !options?.signal?.aborted) {\n\t\t\tconst responseItems = convertResponsesMessages(model, { messages: [output] }, CODEX_TOOL_CALL_PROVIDERS, {\n\t\t\t\tincludeSystemPrompt: false,\n\t\t\t}).filter((item) => item.type !== \"function_call_output\");\n\t\t\tentry.continuation = {\n\t\t\t\tlastRequestBody: fullBody,\n\t\t\t\tlastResponseId: output.responseId,\n\t\t\t\tlastResponseItems: responseItems,\n\t\t\t};\n\t\t}\n\t\tif (options?.signal?.aborted) {\n\t\t\tkeepConnection = false;\n\t\t}\n\t} catch (error) {\n\t\tif (entry) {\n\t\t\tentry.continuation = undefined;\n\t\t}\n\t\tkeepConnection = false;\n\t\tthrow error;\n\t} finally {\n\t\trelease({ keep: keepConnection });\n\t}\n}\n\n// ============================================================================\n// Service Tier Pricing\n// ============================================================================\n\nfunction getServiceTierCostMultiplier(\n\tserviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n\tmodelId?: string,\n): number {\n\tswitch (serviceTier) {\n\t\tcase \"flex\":\n\t\t\treturn 0.5;\n\t\tcase \"priority\": {\n\t\t\tconst id = modelId?.includes(\"/\") ? modelId.split(\"/\").pop()! : modelId;\n\t\t\tif (id?.startsWith(\"gpt-5.5\")) return 2.5;\n\t\t\treturn 2;\n\t\t}\n\t\tdefault:\n\t\t\treturn 1;\n\t}\n}\n\nfunction applyServiceTierPricing(\n\tusage: Usage,\n\tserviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n\tmodel: Pick<Model<\"openai-codex-responses\">, \"id\">,\n) {\n\tconst multiplier = getServiceTierCostMultiplier(serviceTier, model.id);\n\tif (multiplier === 1) return;\n\n\tusage.cost.input *= multiplier;\n\tusage.cost.output *= multiplier;\n\tusage.cost.cacheRead *= multiplier;\n\tusage.cost.cacheWrite *= multiplier;\n\tusage.cost.total = usage.cost.input + usage.cost.output + usage.cost.cacheRead + usage.cost.cacheWrite;\n}\n\nfunction resolveCodexServiceTier(\n\tresponseTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n\trequestedTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n): ResponseCreateParamsStreaming[\"service_tier\"] | undefined {\n\tif (responseTier === \"default\" && (requestedTier === \"flex\" || requestedTier === \"priority\")) {\n\t\treturn requestedTier;\n\t}\n\treturn responseTier ?? requestedTier;\n}\n\n// ============================================================================\n// Error Handling\n// ============================================================================\n\nasync function parseErrorResponse(response: Response): Promise<{ message: string; friendlyMessage?: string }> {\n\tconst raw = await response.text();\n\tlet message = raw || response.statusText || \"Request failed\";\n\tlet friendlyMessage: string | undefined;\n\n\ttry {\n\t\tconst parsed = JSON.parse(raw) as {\n\t\t\terror?: { code?: string; type?: string; message?: string; plan_type?: string; resets_at?: number };\n\t\t};\n\t\tconst err = parsed?.error;\n\t\tif (err) {\n\t\t\tconst code = err.code || err.type || \"\";\n\t\t\tif (/usage_limit_reached|usage_not_included|rate_limit_exceeded/i.test(code) || response.status === 429) {\n\t\t\t\tconst plan = err.plan_type ? ` (${err.plan_type.toLowerCase()} plan)` : \"\";\n\t\t\t\tconst mins = err.resets_at\n\t\t\t\t\t? Math.max(0, Math.round((err.resets_at * 1000 - Date.now()) / 60000))\n\t\t\t\t\t: undefined;\n\t\t\t\tconst when = mins !== undefined ? ` Try again in ~${mins} min.` : \"\";\n\t\t\t\tfriendlyMessage = `You have hit your ChatGPT usage limit${plan}.${when}`.trim();\n\t\t\t}\n\t\t\tmessage = err.message || friendlyMessage || message;\n\t\t}\n\t} catch {\n\t\t// Best-effort extraction of friendly error message — raw text is the fallback\n\t}\n\n\treturn { message, friendlyMessage };\n}\n\n// ============================================================================\n// Auth & Headers\n// ============================================================================\n\nfunction extractAccountId(token: string): string {\n\ttry {\n\t\tconst parts = token.split(\".\");\n\t\tif (parts.length !== 3) throw new Error(\"Invalid token\");\n\t\tconst payload = JSON.parse(atob(parts[1]));\n\t\tconst accountId = payload?.[JWT_CLAIM_PATH]?.chatgpt_account_id;\n\t\tif (!accountId) throw new Error(\"No account ID in token\");\n\t\treturn accountId;\n\t} catch {\n\t\t// Re-throw with unified message for any JWT decode failure\n\t\tthrow new Error(\"Failed to extract accountId from token\");\n\t}\n}\n\nfunction createCodexRequestId(): string {\n\tif (typeof globalThis.crypto?.randomUUID === \"function\") {\n\t\treturn globalThis.crypto.randomUUID();\n\t}\n\treturn `codex_${Date.now()}_${Math.random().toString(36).slice(2, 10)}`;\n}\n\nfunction buildBaseCodexHeaders(\n\tinitHeaders: Record<string, string> | undefined,\n\tadditionalHeaders: Record<string, string> | undefined,\n\taccountId: string,\n\ttoken: string,\n): Headers {\n\tconst headers = new Headers(initHeaders);\n\tfor (const [key, value] of Object.entries(additionalHeaders || {})) {\n\t\theaders.set(key, value);\n\t}\n\theaders.set(\"Authorization\", `Bearer ${token}`);\n\theaders.set(\"chatgpt-account-id\", accountId);\n\theaders.set(\"originator\", \"dreb\");\n\tconst userAgent = _os ? `dreb (${_os.platform()} ${_os.release()}; ${_os.arch()})` : \"dreb (browser)\";\n\theaders.set(\"User-Agent\", userAgent);\n\treturn headers;\n}\n\nfunction buildSSEHeaders(\n\tinitHeaders: Record<string, string> | undefined,\n\tadditionalHeaders: Record<string, string> | undefined,\n\taccountId: string,\n\ttoken: string,\n\tsessionId?: string,\n\tliteSessionId?: string,\n): Headers {\n\tconst headers = buildBaseCodexHeaders(initHeaders, additionalHeaders, accountId, token);\n\theaders.set(\"OpenAI-Beta\", \"responses=experimental\");\n\theaders.set(\"accept\", \"text/event-stream\");\n\theaders.set(\"content-type\", \"application/json\");\n\n\tif (liteSessionId) {\n\t\theaders.set(\"session-id\", liteSessionId);\n\t\theaders.set(\"x-session-affinity\", liteSessionId);\n\t\theaders.set(\"version\", CODEX_COMPATIBILITY_VERSION);\n\t\theaders.set(RESPONSES_LITE_HEADER, \"true\");\n\t} else if (sessionId) {\n\t\theaders.set(\"session_id\", sessionId);\n\t}\n\n\treturn headers;\n}\n\nfunction buildWebSocketHeaders(\n\tinitHeaders: Record<string, string> | undefined,\n\tadditionalHeaders: Record<string, string> | undefined,\n\taccountId: string,\n\ttoken: string,\n\trequestId: string,\n\tliteSessionId?: string,\n): Headers {\n\tconst headers = buildBaseCodexHeaders(initHeaders, additionalHeaders, accountId, token);\n\theaders.delete(\"accept\");\n\theaders.delete(\"content-type\");\n\theaders.delete(\"OpenAI-Beta\");\n\theaders.delete(\"openai-beta\");\n\theaders.set(\"OpenAI-Beta\", OPENAI_BETA_RESPONSES_WEBSOCKETS);\n\theaders.set(\"x-client-request-id\", requestId);\n\tif (liteSessionId) {\n\t\theaders.set(\"session-id\", liteSessionId);\n\t\theaders.set(\"x-session-affinity\", liteSessionId);\n\t\theaders.set(\"version\", CODEX_COMPATIBILITY_VERSION);\n\t\theaders.set(RESPONSES_LITE_HEADER, \"true\");\n\t} else {\n\t\theaders.set(\"session_id\", requestId);\n\t}\n\treturn headers;\n}\n"]}

package/dist/providers/openai-codex-responses.js

@@ -8,10 +8,9 @@ if (typeof process !== "undefined" && (process.versions?.node || process.version
     });
 }
 import { getEnvApiKey } from "../env-api-keys.js";
-import { supportsXhigh } from "../models.js";
 import { AssistantMessageEventStream } from "../utils/event-stream.js";
 import { convertResponsesMessages, convertResponsesTools, processResponsesStream } from "./openai-responses-shared.js";
-import { buildBaseOptions, clampReasoning } from "./simple-options.js";
+import { buildBaseOptions, resolveReasoningEffort } from "./simple-options.js";
 // ============================================================================
 // Configuration
 // ============================================================================
@@ -215,7 +214,7 @@ export const streamSimpleOpenAICodexResponses = (model, context, options) => {
         throw new Error(`No API key for provider: ${model.provider}`);
     }
     const base = buildBaseOptions(model, options, apiKey);
-    const reasoningEffort = supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning);
+    const reasoningEffort = resolveReasoningEffort(model, options?.reasoning);
     return streamOpenAICodexResponses(model, context, {
         ...base,
         reasoningEffort,