feat(codex): support native Anthropic Messages protocol as upstream (#5071)

* feat(codex): support native Anthropic Messages protocol as upstream Allow gateways that only expose the native Anthropic Messages protocol (/v1/messages) to be used by Codex: the local proxy performs bidirectional request/response/streaming conversion between Responses and Anthropic. Backend: - Add two conversion modules: transform_codex_anthropic / streaming_codex_anthropic - codex.rs: add routing detection and auth: ANTHROPIC_AUTH_TOKEN→Bearer (default), ANTHROPIC_API_KEY→x-api-key, mutually exclusive - handlers.rs: add handle_codex_anthropic_to_responses_transform - forwarder.rs: support optional Claude Code client fingerprint impersonation (User-Agent / anthropic-beta / x-app / system prompt first-line injection) and /responses→/v1/messages rewriting - codex_config: the anthropic format reuses the NativeResponses profile to strip custom tools - ProviderMeta: add impersonateClaudeCode Frontend: - CodexApiFormat: add "anthropic"; the form adds auth field selection and an impersonation toggle - Add en/ja/zh/zh-TW copy Robustness: - Downgrade when tool history / forced tool_choice conflicts with extended thinking, avoiding upstream 400s - Emit cache_creation_input_tokens in usage and use saturating_add to guard against overflow - Append a unique suffix to non-streaming/streaming output-item ids to avoid multi-segment text/thinking overwriting each other * fix(codex): harden Anthropic bridge against empty text blocks and truncated streams - Drop empty/whitespace-only text content blocks when rebuilding Anthropic messages from Responses history; Anthropic 400s on empty text blocks (e.g. an empty assistant text emitted alongside a tool_use), which broke follow-up and tool-result requests. Also drop messages left without content. - Do not report a truncated Anthropic stream as completed: when the SSE connection ends before message_stop with no stop_reason, emit an incomplete response if partial output exists, or a failed (stream_truncated) response otherwise, mirroring the chat converter's EOF handling. * fix(codex): resolve Anthropic-bridge review findings Blocking: 1. Defer stripping the [1m] long-context marker until after catalog matching and model write-back, re-stripping it on the final Codex→Anthropic body and setting a flag to emit the context-1m-2025-08-07 beta header, so the marker is no longer lost or overridden by the default model. 2. Gate Anthropic thinking on the trailing turn only (via trailing_turn_allows_thinking) instead of scanning full history, so a Codex session that resends history each round no longer permanently loses thinking after the first tool call. 3. Inject 5m ephemeral cache_control on the Codex→Anthropic body by reusing cache_injector (handling the system string→array conversion), so system/tools/history are cached instead of re-sent at full price every round. 4. Add a shared base_url_is_full_endpoint helper (normalizing whitespace/query/fragment/trailing slash) used by both the Anthropic and Chat paths, so a base URL already ending in /v1/messages is treated as a full endpoint instead of double-appending to /v1/messages/v1/messages. 5. Align the catalog tool-profile predicate with the routing predicate so resolve_codex_catalog_tool_profile returns the Anthropic profile whenever the request converts to Anthropic, preventing freeform tools like apply_patch from being silently filtered by a ProxyChat catalog. 6. Explicitly disable native web_search for the Anthropic profile (including the no-catalog branch) via set_codex_native_web_search_field, so Codex no longer treats it as available while the transform silently drops it. 7. Only forward tool_choice when tools survive filtering, dropping it otherwise, to avoid a non-retryable 400 from upstream when tool_choice is sent with no tools. 8. Lower the fallback default to max_tokens=8192 (only when max_output_tokens is omitted) and clamp the thinking budget to max_tokens/2, disabling thinking below the 1024 floor, to avoid hard 400s on low-output-ceiling models/gateways. Minor: 9. Centralize the Codex/OpenAI fingerprint-header denylist in is_codex_client_fingerprint_header so impersonating Claude Code uniformly drops originator/session_id/conversation_id/chatgpt-account-id/x-client-request-id/openai-* and the x-stainless-*/x-codex-* prefixes. 10. Retain content_block_start.input as start_input and fall back to it at block close when no input_json_delta arrived, so a gateway that carries the full tool input on the start event no longer yields empty tool arguments. 11. Extract a shared codex_responses_sse module as the single Responses SSE envelope builder that both the chat and anthropic streaming emitters delegate to, with byte-for-byte-unchanged wire output, so future event-format fixes touch one place instead of two. * fix(codex): add per-provider max_output_tokens override for Codex→Anthropic path Codex does not forward model_max_output_tokens in the request body, causing the proxy to fall back to a conservative 8192 default. This truncates long or thinking-heavy responses (stop_reason=max_tokens). - Add maxOutputTokens field to ProviderMeta (Rust + TypeScript) - Inject the value into the Anthropic request body before transform, taking precedence over request-supplied and default values - Add numeric input in Codex form fields (Anthropic format only) - Add i18n entries for label, placeholder, and hint (en/zh/zh-TW/ja) - Include roundtrip and omission unit tests for the new field * fix(codex): harden Anthropic protocol bridge * fix(codex): address Anthropic bridge review * fix(codex): preserve flattened Anthropic inputs * fix(codex): harden Anthropic recovery paths * fix(ci): satisfy Rust clippy --------- Co-authored-by: Jason <farion1231@gmail.com>
2026-07-30 02:14:43 +08:00 · 2026-07-10 23:06:30 +08:00
parent 98ccde0050
commit 99e11e0851
29 changed files with 5733 additions and 371 deletions
@@ -22,6 +22,7 @@ import {
  extractCodexBaseUrl,
  extractCodexExperimentalBearerToken,
  extractCodexWireApi,
+  isCodexAnthropicWireApi,
  isCodexChatWireApi,
 } from "@/utils/providerConfigUtils";
 import { useProviderHealth } from "@/lib/query/failover";
@@ -220,11 +221,16 @@ export function ProviderCard({
    provider.meta?.providerType === PROVIDER_TYPES.CODEX_OAUTH;
  const codexNeedsRouting = useMemo(() => {
    if (appId !== "codex" || provider.category === "official") return false;
-    if (provider.meta?.apiFormat === "openai_chat") return true;
+    if (
+      provider.meta?.apiFormat === "openai_chat" ||
+      provider.meta?.apiFormat === "anthropic"
+    )
+      return true;
    const config = (provider.settingsConfig as Record<string, any>)?.config;
    return (
      typeof config === "string" &&
-      isCodexChatWireApi(extractCodexWireApi(config))
+      (isCodexChatWireApi(extractCodexWireApi(config)) ||
+        isCodexAnthropicWireApi(extractCodexWireApi(config)))
    );
  }, [
    appId,
@@ -36,6 +36,7 @@ import { CustomUserAgentField } from "./CustomUserAgentField";
 import { LocalProxyRequestOverridesField } from "./LocalProxyRequestOverridesField";
 import { cn } from "@/lib/utils";
 import type {
+  ClaudeApiKeyField,
  CodexApiFormat,
  CodexCatalogModel,
  CodexChatReasoning,
@@ -73,6 +74,15 @@ interface CodexFormFieldsProps {
  // Note: wire_api is always "responses" for Codex; apiFormat controls proxy-layer conversion
  apiFormat: CodexApiFormat;
  onApiFormatChange: (format: CodexApiFormat) => void;
+  // Auth field for the Anthropic Messages upstream (only used when apiFormat === "anthropic")
+  anthropicAuthField: ClaudeApiKeyField;
+  onAnthropicAuthFieldChange: (value: ClaudeApiKeyField) => void;
+  // Anthropic path: whether to emulate the Claude Code client
+  impersonateClaudeCode: boolean;
+  onImpersonateClaudeCodeChange: (value: boolean) => void;
+  // Anthropic path: output ceiling override (empty string = use default). Digits only.
+  maxOutputTokens: string;
+  onMaxOutputTokensChange: (value: string) => void;
  codexChatReasoning?: CodexChatReasoning;
  onCodexChatReasoningChange?: (value: CodexChatReasoning) => void;

@@ -158,6 +168,12 @@ export function CodexFormFields({
  onAutoSelectChange,
  apiFormat,
  onApiFormatChange,
+  anthropicAuthField,
+  onAnthropicAuthFieldChange,
+  impersonateClaudeCode,
+  onImpersonateClaudeCodeChange,
+  maxOutputTokens,
+  onMaxOutputTokensChange,
  codexChatReasoning = {},
  onCodexChatReasoningChange,
  catalogModels = [],
@@ -177,6 +193,7 @@ export function CodexFormFields({
  // 思考能力随 Chat 格式显示（仅 Chat Completions 转换路径用得上）；模型映射常驻
  //（填了才生成 catalog）。两者都已与「路由接管」概念解耦。
  const isChatFormat = apiFormat === "openai_chat";
+  const isAnthropicFormat = apiFormat === "anthropic";
  const canEditCatalog = Boolean(onCatalogModelsChange);
  const canEditReasoning = Boolean(onCodexChatReasoningChange);
  const supportsThinking =
@@ -194,8 +211,10 @@ export function CodexFormFields({
    hasRequestOverrides ||
    catalogModels.length > 0 ||
    apiFormat === "openai_responses" ||
+    isAnthropicFormat ||
    supportsThinking ||
-    supportsEffort;
+    supportsEffort ||
+    !!maxOutputTokens;
  const [advancedExpanded, setAdvancedExpanded] = useState(hasAnyAdvancedValue);

  // 预设/编辑加载填充高级值后自动展开（仅从折叠→展开，不会自动折叠）
@@ -449,15 +468,118 @@ export function CodexFormFields({
                          defaultValue: "Responses（原生）",
                        })}
                      </SelectItem>
+                      <SelectItem value="anthropic">
+                        {t("codexConfig.upstreamFormatAnthropic", {
+                          defaultValue: "Anthropic Messages（需开启路由）",
+                        })}
+                      </SelectItem>
                    </SelectContent>
                  </Select>
                  <p className="text-xs leading-relaxed text-muted-foreground">
                    {t("codexConfig.upstreamFormatHint", {
                      defaultValue:
-                        "供应商原生是 Responses API 就选 Responses（直连，不转换格式）；使用 Chat Completions 协议就选 Chat（需开启路由接管才能转换为 Chat Completions）。",
+                        "供应商原生是 Responses API 就选 Responses（直连，不转换格式）；使用 Chat Completions 协议就选 Chat；供应商只提供原生 Anthropic Messages 协议就选 Anthropic Messages。Chat 与 Anthropic Messages 均需开启路由接管才能转换为 Responses。",
                    })}
                  </p>
                </div>
+
+                {isAnthropicFormat && (
+                  <div className="space-y-1.5">
+                    <FormLabel htmlFor="codex-anthropic-auth-field">
+                      {t("codexConfig.anthropicAuthFieldLabel", {
+                        defaultValue: "认证字段",
+                      })}
+                    </FormLabel>
+                    <Select
+                      value={anthropicAuthField}
+                      onValueChange={(value) =>
+                        onAnthropicAuthFieldChange(value as ClaudeApiKeyField)
+                      }
+                    >
+                      <SelectTrigger
+                        id="codex-anthropic-auth-field"
+                        className="w-full"
+                      >
+                        <SelectValue />
+                      </SelectTrigger>
+                      <SelectContent>
+                        <SelectItem value="ANTHROPIC_AUTH_TOKEN">
+                          {t("codexConfig.anthropicAuthFieldAuthToken", {
+                            defaultValue:
+                              "ANTHROPIC_AUTH_TOKEN（Authorization）",
+                          })}
+                        </SelectItem>
+                        <SelectItem value="ANTHROPIC_API_KEY">
+                          {t("codexConfig.anthropicAuthFieldApiKey", {
+                            defaultValue: "ANTHROPIC_API_KEY（x-api-key）",
+                          })}
+                        </SelectItem>
+                      </SelectContent>
+                    </Select>
+                    <p className="text-xs leading-relaxed text-muted-foreground">
+                      {t("codexConfig.anthropicAuthFieldHint", {
+                        defaultValue:
+                          "选择网关接收 API Key 的请求头：ANTHROPIC_AUTH_TOKEN 发送 Authorization: Bearer；ANTHROPIC_API_KEY 发送 x-api-key。两者只发其一。",
+                      })}
+                    </p>
+                  </div>
+                )}
+
+                {isAnthropicFormat && (
+                  <div className="flex items-center justify-between gap-4 border-t border-border-default pt-3">
+                    <div className="space-y-1">
+                      <FormLabel>
+                        {t("codexConfig.impersonateClaudeCodeLabel", {
+                          defaultValue: "模拟 Claude Code 客户端",
+                        })}
+                      </FormLabel>
+                      <p className="text-xs leading-relaxed text-muted-foreground">
+                        {t("codexConfig.impersonateClaudeCodeHint", {
+                          defaultValue:
+                            "网关或其上游限制只能通过 Claude Code 使用时开启：伪装 User-Agent、anthropic-beta、x-app 请求头，并在系统提示首行注入 Claude Code 身份。",
+                        })}
+                      </p>
+                    </div>
+                    <Switch
+                      checked={impersonateClaudeCode}
+                      onCheckedChange={onImpersonateClaudeCodeChange}
+                      aria-label={t("codexConfig.impersonateClaudeCodeLabel", {
+                        defaultValue: "模拟 Claude Code 客户端",
+                      })}
+                    />
+                  </div>
+                )}
+
+                {isAnthropicFormat && (
+                  <div className="space-y-1.5 border-t border-border-default pt-3">
+                    <FormLabel htmlFor="codex-anthropic-max-output-tokens">
+                      {t("codexConfig.maxOutputTokensLabel", {
+                        defaultValue: "最大输出 tokens",
+                      })}
+                    </FormLabel>
+                    <Input
+                      id="codex-anthropic-max-output-tokens"
+                      type="number"
+                      min={1}
+                      inputMode="numeric"
+                      value={maxOutputTokens}
+                      onChange={(event) =>
+                        onMaxOutputTokensChange(
+                          event.target.value.replace(/[^\d]/g, ""),
+                        )
+                      }
+                      placeholder={t("codexConfig.maxOutputTokensPlaceholder", {
+                        defaultValue: "留空则使用默认 8192",
+                      })}
+                    />
+                    <p className="text-xs leading-relaxed text-muted-foreground">
+                      {t("codexConfig.maxOutputTokensHint", {
+                        defaultValue:
+                          "Codex 不会把 model_max_output_tokens 写进请求体，默认上限 8192 容易在长回答或深度思考时被截断（stop_reason=max_tokens）。此处设置会作为 Anthropic 的 max_tokens 覆盖请求值。请勿超过该模型/网关的真实输出上限，否则可能 400。留空使用默认 8192。",
+                      })}
+                    </p>
+                  </div>
+                )}
              </div>
            )}

@@ -60,6 +60,7 @@ import {
 } from "@/utils/providerConfigUtils";
 import { mergeProviderMeta } from "@/utils/providerMetaUtils";
 import {
+  codexApiFormatFromWireApi,
  extractCodexWireApi,
  setCodexWireApi,
  setCodexModelName as setCodexModelNameInConfig,
@@ -131,25 +132,6 @@ type PresetEntry = {
    | HermesProviderPreset;
 };

-const codexApiFormatFromWireApi = (
-  wireApi: string | undefined,
-): CodexApiFormat | undefined => {
-  switch (wireApi?.trim().toLowerCase()) {
-    case "chat":
-    case "chat_completions":
-    case "chat-completions":
-    case "openai_chat":
-    case "openai-chat":
-      return "openai_chat";
-    case "responses":
-    case "openai_responses":
-    case "openai-responses":
-      return "openai_responses";
-    default:
-      return undefined;
-  }
-};
-
 export const normalizeCodexCatalogModelsForSave = (
  models: CodexCatalogModel[],
 ): CodexCatalogModel[] => {
@@ -586,19 +568,43 @@ function ProviderFormFull({
  const initialCodexApiFormat: CodexApiFormat =
    initialData?.meta?.apiFormat === "openai_chat"
      ? "openai_chat"
-      : initialData?.meta?.apiFormat === "openai_responses"
-        ? "openai_responses"
-        : (codexApiFormatFromWireApi(
-            extractCodexWireApi(
-              typeof initialData?.settingsConfig?.config === "string"
-                ? initialData.settingsConfig.config
-                : "",
-            ),
-          ) ?? "openai_responses");
+      : initialData?.meta?.apiFormat === "anthropic"
+        ? "anthropic"
+        : initialData?.meta?.apiFormat === "openai_responses"
+          ? "openai_responses"
+          : (codexApiFormatFromWireApi(
+              extractCodexWireApi(
+                typeof initialData?.settingsConfig?.config === "string"
+                  ? initialData.settingsConfig.config
+                  : "",
+              ),
+            ) ?? "openai_responses");

  const [localCodexApiFormat, setLocalCodexApiFormat] =
    useState<CodexApiFormat>(initialCodexApiFormat);

+  // Auth-field choice for the Anthropic Messages upstream (defaults to the Bearer form)
+  const initialCodexAnthropicAuthField: ClaudeApiKeyField =
+    initialData?.meta?.apiKeyField === "ANTHROPIC_API_KEY"
+      ? "ANTHROPIC_API_KEY"
+      : "ANTHROPIC_AUTH_TOKEN";
+  const [localCodexAnthropicAuthField, setLocalCodexAnthropicAuthField] =
+    useState<ClaudeApiKeyField>(initialCodexAnthropicAuthField);
+
+  // Emulate the Claude Code client: off by default, enabled only when the user explicitly turns it on (true)
+  const [localCodexImpersonateClaudeCode, setLocalCodexImpersonateClaudeCode] =
+    useState<boolean>(initialData?.meta?.impersonateClaudeCode === true);
+
+  // Codex → Anthropic output ceiling override (empty string = use the 8192 default).
+  // Kept as a string so the numeric input can be cleared; parsed on save.
+  const [localCodexMaxOutputTokens, setLocalCodexMaxOutputTokens] =
+    useState<string>(
+      typeof initialData?.meta?.maxOutputTokens === "number" &&
+        initialData.meta.maxOutputTokens > 0
+        ? String(initialData.meta.maxOutputTokens)
+        : "",
+    );
+
  const { configError: codexConfigError, debouncedValidate } =
    useCodexTomlValidation();

@@ -1502,6 +1508,28 @@ function ProviderFormFull({
        category !== "official" &&
        localApiKeyField !== "ANTHROPIC_AUTH_TOKEN"
          ? localApiKeyField
+          : appId === "codex" &&
+              category !== "official" &&
+              localCodexApiFormat === "anthropic" &&
+              localCodexAnthropicAuthField !== "ANTHROPIC_AUTH_TOKEN"
+            ? localCodexAnthropicAuthField
+            : undefined,
+      // Off by default; persist true only for codex+anthropic when the user explicitly enables it
+      impersonateClaudeCode:
+        appId === "codex" &&
+        category !== "official" &&
+        localCodexApiFormat === "anthropic" &&
+        localCodexImpersonateClaudeCode
+          ? true
+          : undefined,
+      // Persist only for codex+anthropic when a positive value was entered
+      maxOutputTokens:
+        appId === "codex" &&
+        category !== "official" &&
+        localCodexApiFormat === "anthropic" &&
+        localCodexMaxOutputTokens.trim() !== "" &&
+        Number(localCodexMaxOutputTokens) > 0
+          ? Number(localCodexMaxOutputTokens)
          : undefined,
      isFullUrl:
        supportsFullUrl && category !== "official" && localIsFullUrl
@@ -2142,6 +2170,12 @@ function ProviderFormFull({
              onAutoSelectChange={setEndpointAutoSelect}
              apiFormat={localCodexApiFormat}
              onApiFormatChange={handleCodexApiFormatChange}
+              anthropicAuthField={localCodexAnthropicAuthField}
+              onAnthropicAuthFieldChange={setLocalCodexAnthropicAuthField}
+              impersonateClaudeCode={localCodexImpersonateClaudeCode}
+              onImpersonateClaudeCodeChange={setLocalCodexImpersonateClaudeCode}
+              maxOutputTokens={localCodexMaxOutputTokens}
+              onMaxOutputTokensChange={setLocalCodexMaxOutputTokens}
              codexChatReasoning={codexChatReasoning}
              onCodexChatReasoningChange={setCodexChatReasoning}
              catalogModels={codexCatalogModels}