Merge pull request #149 from cita-777/fix/codex-responses-runtime-parity

fix: align codex responses continuity handling

src/server/routes/proxy/responses.codex-oauth.test.ts

@@ -193,9 +193,7 @@ describe('responses proxy codex oauth refresh', () => {
     expect(secondOptions.headers['Chatgpt-Account-Id'] || secondOptions.headers['chatgpt-account-id']).toBe('chatgpt-account-123');
     expect(secondOptions.headers.Version || secondOptions.headers.version).toBe('0.101.0');
     expect(String(secondOptions.headers.Session_id || secondOptions.headers.session_id || '')).toMatch(/^[0-9a-f-]{36}$/i);
-    expect(secondOptions.headers.Conversation_id || secondOptions.headers.conversation_id).toBe(
+    expect(secondOptions.headers.Conversation_id || secondOptions.headers.conversation_id).toBeUndefined();
-      secondOptions.headers.Session_id || secondOptions.headers.session_id,
-    );
     expect(secondOptions.headers['User-Agent'] || secondOptions.headers['user-agent']).toBe('codex_cli_rs/0.101.0 (Mac OS 26.0.1; arm64) Apple_Terminal/464');
     expect(secondOptions.headers.Accept || secondOptions.headers.accept).toBe('text/event-stream');
     expect(secondOptions.headers.Connection || secondOptions.headers.connection).toBe('Keep-Alive');
@@ -230,9 +228,7 @@ describe('responses proxy codex oauth refresh', () => {
     const [, options] = fetchMock.mock.calls[0] as [string, any];
     const forwardedBody = JSON.parse(options.body);
     expect(forwardedBody.instructions).toBe('');
-    expect(forwardedBody.prompt_cache_key).toBe(
+    expect(forwardedBody.prompt_cache_key).toBeUndefined();
-      options.headers.Session_id || options.headers.session_id,
-    );
     expect(forwardedBody.stream).toBe(true);
     expect(forwardedBody.input).toEqual([
       {
@@ -243,6 +239,39 @@ describe('responses proxy codex oauth refresh', () => {
     ]);
   });
 
+  it('preserves explicit prompt_cache_key for codex responses requests', async () => {
+    fetchMock.mockResolvedValue(new Response(JSON.stringify({
+      id: 'resp_codex_with_cache_key',
+      object: 'response',
+      model: 'gpt-5.2-codex',
+      status: 'completed',
+      output_text: 'ok with cache key',
+      usage: { input_tokens: 4, output_tokens: 2, total_tokens: 6 },
+    }), {
+      status: 200,
+      headers: { 'content-type': 'application/json' },
+    }));
+
+    const response = await app.inject({
+      method: 'POST',
+      url: '/v1/responses',
+      payload: {
+        model: 'gpt-5.2-codex',
+        prompt_cache_key: 'codex-cache-123',
+        input: 'hello codex',
+      },
+    });
+
+    expect(response.statusCode).toBe(200);
+    expect(fetchMock).toHaveBeenCalledTimes(1);
+
+    const [, options] = fetchMock.mock.calls[0] as [string, any];
+    const forwardedBody = JSON.parse(options.body);
+    expect(options.headers.Session_id || options.headers.session_id).toBe('codex-cache-123');
+    expect(options.headers.Conversation_id || options.headers.conversation_id).toBe('codex-cache-123');
+    expect(forwardedBody.prompt_cache_key).toBe('codex-cache-123');
+  });
+
   it('records codex usage_limit_reached reset hints on upstream 429 failures', async () => {
     fetchMock.mockResolvedValue(new Response(JSON.stringify({
       error: {

src/server/routes/proxy/responses.ts

@@ -124,24 +124,11 @@ function carriesResponsesFileUrlInput(value: unknown): boolean {
 
 type UsageSummary = ReturnType<typeof parseProxyUsage>;
 
-function deriveCodexSessionCacheKey(input: {
+function deriveCodexExplicitSessionId(body: Record<string, unknown>): string | null {
-  body: Record<string, unknown>;
+  const promptCacheKey = typeof body.prompt_cache_key === 'string'
-  requestedModel: string;
+    ? body.prompt_cache_key.trim()
-  proxyToken: string | null;
-}): string | null {
-  const promptCacheKey = typeof input.body.prompt_cache_key === 'string'
-    ? input.body.prompt_cache_key.trim()
     : '';
-  if (promptCacheKey) {
+  return promptCacheKey || null;
-    return `${input.requestedModel}:responses:${promptCacheKey}`;
-  }
-
-  const proxyToken = typeof input.proxyToken === 'string' ? input.proxyToken.trim() : '';
-  if (proxyToken) {
-    return `${input.requestedModel}:proxy:${proxyToken}`;
-  }
-
-  return null;
 }
 
 export async function responsesProxyRoute(app: FastifyInstance) {
@@ -271,11 +258,7 @@ export async function responsesProxyRoute(app: FastifyInstance) {
       );
       const buildEndpointRequest = (endpoint: 'chat' | 'messages' | 'responses') => {
         const upstreamStream = isStream || (isCodexSite && endpoint === 'responses');
-        const codexSessionCacheKey = deriveCodexSessionCacheKey({
+        const codexExplicitSessionId = deriveCodexExplicitSessionId(normalizedResponsesBody);
-          body: normalizedResponsesBody,
-          requestedModel,
-          proxyToken: getProxyAuthContext(request)?.token || null,
-        });
         const endpointRequest = buildUpstreamEndpointRequest({
           endpoint,
           modelName,
@@ -290,7 +273,7 @@ export async function responsesProxyRoute(app: FastifyInstance) {
           responsesOriginalBody: normalizedResponsesBody,
           downstreamHeaders: request.headers as Record<string, unknown>,
           providerHeaders: buildProviderHeaders(),
-          codexSessionCacheKey,
+          codexExplicitSessionId,
         });
         const upstreamPath = (
           isCompactRequest && endpoint === 'responses'

src/server/routes/proxy/upstreamEndpoint.test.ts

@@ -350,6 +350,7 @@ describe('buildUpstreamEndpointRequest', () => {
         Originator: 'codex_cli_rs',
         'Chatgpt-Account-Id': 'chatgpt-account-123',
       },
+      codexSessionCacheKey: 'gpt-5.2-codex:proxy:test-key',
     } as any);
 
     expect(request.path).toBe('/responses');
@@ -417,6 +418,56 @@ describe('buildUpstreamEndpointRequest', () => {
     expect(firstRequest.body.prompt_cache_key).toBe(secondRequest.body.prompt_cache_key);
   });
 
+  it('does not synthesize prompt_cache_key or conversation_id for native codex responses requests without one', () => {
+    const request = buildUpstreamEndpointRequest({
+      endpoint: 'responses',
+      modelName: 'gpt-5.4',
+      stream: false,
+      tokenValue: 'oauth-access-token',
+      sitePlatform: 'codex',
+      siteUrl: 'https://chatgpt.com/backend-api/codex',
+      openaiBody: {},
+      downstreamFormat: 'responses',
+      responsesOriginalBody: {
+        model: 'gpt-5.4',
+        input: 'hello codex',
+      },
+      providerHeaders: {
+        Originator: 'codex_cli_rs',
+      },
+    } as any);
+
+    expect(request.headers.Session_id).toMatch(/^[0-9a-f-]{36}$/i);
+    expect(request.headers.Conversation_id).toBeUndefined();
+    expect(request.body.prompt_cache_key).toBeUndefined();
+  });
+
+  it('preserves explicit prompt_cache_key for native codex responses requests', () => {
+    const request = buildUpstreamEndpointRequest({
+      endpoint: 'responses',
+      modelName: 'gpt-5.4',
+      stream: false,
+      tokenValue: 'oauth-access-token',
+      sitePlatform: 'codex',
+      siteUrl: 'https://chatgpt.com/backend-api/codex',
+      openaiBody: {},
+      downstreamFormat: 'responses',
+      responsesOriginalBody: {
+        model: 'gpt-5.4',
+        prompt_cache_key: 'codex-cache-123',
+        input: 'hello codex',
+      },
+      providerHeaders: {
+        Originator: 'codex_cli_rs',
+      },
+      codexExplicitSessionId: 'codex-cache-123',
+    } as any);
+
+    expect(request.headers.Session_id).toBe('codex-cache-123');
+    expect(request.headers.Conversation_id).toBe('codex-cache-123');
+    expect(request.body.prompt_cache_key).toBe('codex-cache-123');
+  });
+
   it('builds gemini-cli native requests with project envelope and bearer headers', () => {
     const request = buildUpstreamEndpointRequest({
       endpoint: 'responses',

src/server/routes/proxy/upstreamEndpoint.ts

@@ -283,16 +283,26 @@ function extractClaudeBetasFromBody(body: Record<string, unknown>): {
 function buildCodexRuntimeHeaders(input: {
   baseHeaders: Record<string, string>;
   providerHeaders?: Record<string, string>;
+  explicitSessionId?: string | null;
   continuityKey?: string | null;
 }): Record<string, string> {
   const originator = getInputHeader(input.providerHeaders, 'originator') || 'codex_cli_rs';
   const accountId = getInputHeader(input.providerHeaders, 'chatgpt-account-id');
+  const explicitSessionId = asTrimmedString(input.explicitSessionId);
+  const continuityKey = asTrimmedString(input.continuityKey);
   const sessionId = (
     getInputHeader(input.baseHeaders, 'session_id')
     || getInputHeader(input.baseHeaders, 'session-id')
-    || (input.continuityKey ? uuidFromSeed(`metapi:codex:${input.continuityKey}`) : null)
+    || explicitSessionId
+    || (continuityKey ? uuidFromSeed(`metapi:codex:${continuityKey}`) : null)
     || randomUUID()
   );
+  const conversationId = (
+    getInputHeader(input.baseHeaders, 'conversation_id')
+    || getInputHeader(input.baseHeaders, 'conversation-id')
+    || explicitSessionId
+    || (continuityKey ? sessionId : null)
+  );
 
   return {
     ...input.baseHeaders,
@@ -300,7 +310,7 @@ function buildCodexRuntimeHeaders(input: {
     Originator: originator,
     Version: CODEX_CLIENT_VERSION,
     Session_id: sessionId,
-    Conversation_id: sessionId,
+    ...(conversationId ? { Conversation_id: conversationId } : {}),
     'User-Agent': CODEX_DEFAULT_USER_AGENT,
     Accept: 'text/event-stream',
     Connection: 'Keep-Alive',
@@ -753,6 +763,7 @@ export function buildUpstreamEndpointRequest(input: {
   downstreamHeaders?: Record<string, unknown>;
   providerHeaders?: Record<string, string>;
   codexSessionCacheKey?: string | null;
+  codexExplicitSessionId?: string | null;
 }): {
   path: string;
   headers: Record<string, string>;
@@ -985,6 +996,7 @@ export function buildUpstreamEndpointRequest(input: {
           ...responsesHeaders,
         },
         providerHeaders: input.providerHeaders,
+        explicitSessionId: asTrimmedString(input.codexExplicitSessionId) || null,
         continuityKey: asTrimmedString(input.codexSessionCacheKey) || null,
       })
       : ensureStreamAcceptHeader({
@@ -994,7 +1006,12 @@ export function buildUpstreamEndpointRequest(input: {
     const codexSessionId = sitePlatform === 'codex'
       ? (getInputHeader(headers, 'session_id') || getInputHeader(headers, 'session-id'))
       : null;
-    const runtimeBody = sitePlatform === 'codex' && codexSessionId
+    const shouldInjectDerivedPromptCacheKey = sitePlatform === 'codex'
+      && !!codexSessionId
+      && !asTrimmedString((body as Record<string, unknown>).prompt_cache_key)
+      && !asTrimmedString(input.codexExplicitSessionId)
+      && !!asTrimmedString(input.codexSessionCacheKey);
+    const runtimeBody = shouldInjectDerivedPromptCacheKey
       ? {
         ...body,
         prompt_cache_key: codexSessionId,