fix: preserve Claude Go affinity across combo selection and failover

docs-site/src/content/docs/guides/providers.md

@@ -393,6 +393,11 @@ Chat, bridged Chat, and Responses derive the same result. Explicit provider-conf
 session headers are operator overrides and are sent unchanged. Clients must keep the
 identifier stable within a conversation and distinct across conversations; requests
 without a session identifier cannot receive automatic session affinity.
+For Claude Messages, valid conversation identity in `metadata.user_id` supplies
+the fallback when no usable explicit session identifier exists. This fallback is
+applied to the final Go destination, including random combo selections and fallback
+attempts, rather than the preliminary route. Shared system-prompt cache keys do
+not identify conversations, and Go-specific identity is not sent to non-Go targets.
 Generated Pi provider configurations enable `compat.sendSessionAffinityHeaders`
 so Pi sends its per-session identity to the proxy. Existing manually managed Pi
 configurations can set this option on their `opencodex` provider as well.

src/server/claude-messages.ts

@@ -856,27 +856,26 @@ async function handleClaudeMessagesWithBudget(
       headers.set("chatgpt-account-id", token.chatgptAccountId);
     }
   }
-  if (opencodeGoRoute) {
-    const session = req.headers.get("x-opencode-session");
-    if (session) headers.set("x-opencode-session", session);
-  }
-  const hasExplicitGoSession = opencodeGoRoute
-    && (sessionLaneIdFromRequest(headers) !== undefined
-      || normalizeLogConversationId(headers.get("x-opencode-session")) !== undefined);
-  const synthesizeGoSession = opencodeGoRoute && !hasExplicitGoSession
+  // Carry Go identity out of band: a combo's preflight target may differ from its
+  // actual dispatch/fallback target. Never add Go-only identity to replay headers.
+  const metadataGoLane = cacheKeySource === "metadata"
+    && typeof internalBody.prompt_cache_key === "string"
     && isRec(anthropicBody)
-    && conversationIdFromClaudeMetadata(isRec(anthropicBody.metadata) ? anthropicBody.metadata : undefined) !== undefined;
-  // Go can also use the Responses adapter; its eligibility gate must win on both wires.
-  if (opencodeGoRoute ? synthesizeGoSession : nativeRoute) {
+    && conversationIdFromClaudeMetadata(isRec(anthropicBody.metadata) ? anthropicBody.metadata : undefined) !== undefined
+    ? normalizeLogConversationId(uuidFromHex(internalBody.prompt_cache_key))
+    : undefined;
+  const claudeGoSessionLane = sessionLaneIdFromRequest(headers)
+    ?? normalizeLogConversationId(req.headers.get("x-opencode-session"))
+    ?? metadataGoLane;
+  if (nativeRoute && !opencodeGoRoute) {
     // ChatGPT-backend prompt-cache affinity rides the session_id HEADER (codex
     // clients always send their session uuid; devlog 090 follow-up: body-level
     // prompt_cache_key alone still yielded cached_tokens:0). Claude Code never sends
     // the header, so synthesize a stable per-session uuid from the same cache key.
-    // Routed Go requests need this lane too for their x-opencode-session affinity —
-    // but ONLY for a real per-session key (metadata.user_id). The system-hash fallback
+    // Use ONLY a real per-session key (metadata.user_id). The system-hash fallback
     // key is shared across Desktop conversations, and a shared session_id's backend
     // semantics are unproven (audit 133 R2#3): body prompt_cache_key only there.
-    if (cacheKeySource === "metadata" && (synthesizeGoSession || !headers.has("session_id")) && typeof internalBody.prompt_cache_key === "string") {
+    if (cacheKeySource === "metadata" && !headers.has("session_id") && typeof internalBody.prompt_cache_key === "string") {
       headers.set("session_id", uuidFromHex(internalBody.prompt_cache_key));
     }
   }
@@ -922,6 +921,7 @@ async function handleClaudeMessagesWithBudget(
     // Without this the replay would look native and a Responses-scoped wire default
     // would fire, disagreeing with the pre-flight decision above.
     inboundWire: "anthropic",
+    claudeGoAffinity: { sessionLane: claudeGoSessionLane },
     stripClaudeMainAuthForNoncanonicalForward: true,
     translatorBudget,
     ...(logIds ? { onFirstOutput: () => recordFirstOutput(logCtx, logIds.start) } : {}),

src/server/responses/core.ts

@@ -1652,6 +1652,8 @@ export interface ConsumedComboFailure {
 
 
 export interface HandleResponsesOptions {
+  /** Internal Claude replay identity; consumed only by the final canonical Go transport. */
+  claudeGoAffinity?: { sessionLane?: string };
   /** Original live policy owner; separate from caller-specific routing/sidecar snapshots. */
   codexAuthPolicy?: CodexAuthPolicyConfig;
   turnAdmissionLease?: AdmissionLease;
@@ -2312,6 +2314,7 @@ async function applyFinalRouteRequestNormalization(args: {
   logCtx: RequestLogContext;
   inboundWire: InboundWire;
   inboundTransport?: "websocket";
+  claudeGoAffinity?: HandleResponsesOptions["claudeGoAffinity"];
 }): Promise<void> {
   const { parsed, route, config, req, logCtx, inboundWire, inboundTransport } = args;
   const effortSelector = prepareEffortNormalization(parsed, route);
@@ -2340,7 +2343,8 @@ async function applyFinalRouteRequestNormalization(args: {
   // Settle the wire once so logging, fast-mode, auth, and sidecars read the adapter
   // this request will actually use (#404).
   route.provider = resolveOpenCodeGoTransport(route.provider,
-    sessionLaneIdFromRequest(req.headers) ?? normalizeLogConversationId(req.headers.get("x-opencode-session")));
+    args.claudeGoAffinity ? args.claudeGoAffinity.sessionLane
+      : sessionLaneIdFromRequest(req.headers) ?? normalizeLogConversationId(req.headers.get("x-opencode-session")));
   route.provider = resolveWireProtocolOverride(route.providerName, route.modelId, route.provider, inboundWire);
   if (preserveAnthropicResponseModel) parsed._responseModelId = responseModelId;
   logCtx.model = route.modelId;
@@ -3686,6 +3690,7 @@ async function handleResponsesInner(
     logCtx,
     inboundWire,
     inboundTransport: options.inboundTransport,
+    claudeGoAffinity: options.claudeGoAffinity,
   });
   // Attribute local auth/cooldown failures to the public selector too; exact auth may fail before
   // the normal post-resolution provider label is assigned.

tests/providers/opencode-go-session-header.test.ts

@@ -1,4 +1,5 @@
-import { afterEach, describe, expect, test } from "bun:test";
+import { afterEach, describe, expect, spyOn, test } from "bun:test";
+import { clearComboSelectionState, clearComboTargetCooldowns } from "../../src/combos";
 import { providerConfigSeed } from "../../src/providers/derive";
 import { resolveOpenCodeGoTransport } from "../../src/providers/opencode-go-transport";
 import { getProviderRegistryEntry } from "../../src/providers/registry";
@@ -120,6 +121,109 @@ describe("OpenCode Go session affinity (#3344)", () => {
   const originalFetch = globalThis.fetch;
   afterEach(() => { globalThis.fetch = originalFetch; });
 
+  for (const model of [CHAT_MODEL, MUSE_MODEL]) {
+    for (const preliminaryAdapter of ["openai-chat", "openai-responses"] as const) {
+    for (const strategy of ["random", "failover"] as const) {
+      for (const identity of [
+        { name: "metadata", headers: {}, metadata: "user_test_account__session_conversation-a", expected: "ocx_a89540229ef781fd5f7adf92a711b436" },
+        { name: "explicit Go header", headers: { [SESSION_HEADER]: "client-session-a" }, metadata: "other-session", expected: "ocx_516d593899f34b7baca2db37c7b0c8c5" },
+        { name: "explicit lane", headers: { session_id: "native-client-session", [SESSION_HEADER]: "client-session-a" }, metadata: "other-session", expected: "ocx_a197dbb87311c29a5fbe51140e3845ce" },
+        { name: "operator override", headers: {}, metadata: "user_test_account__session_conversation-a", operator: true, expected: "operator-session" },
+        { name: "invalid explicit lane", headers: { session_id: "invalid\tidentity", [SESSION_HEADER]: "invalid\tidentity" }, metadata: "user_test_account__session_conversation-a", expected: "ocx_a89540229ef781fd5f7adf92a711b436" },
+        { name: "invalid metadata", headers: {}, metadata: "invalid\u0000identity", expected: null },
+        { name: "shared system only", headers: {}, metadata: undefined, expected: null },
+      ]) {
+      test(`Claude ${strategy} ${preliminaryAdapter} to Go uses ${identity.name} on ${model}`, async () => {
+        clearComboSelectionState();
+        clearComboTargetCooldowns();
+        const requests: Array<{ url: string; headers: Headers }> = [];
+        globalThis.fetch = (async (input: RequestInfo | URL, init?: RequestInit) => {
+          const url = String(input);
+          requests.push({ url, headers: new Headers(init?.headers) });
+          if (url.startsWith("https://other.example")) {
+            return Response.json({ error: { message: "model retired", code: "model_not_found" } }, { status: 404 });
+          }
+          return upstreamResponse(url, true);
+        }) as typeof fetch;
+        const config = {
+          providers: {
+            other: { adapter: preliminaryAdapter, authMode: "key", baseUrl: "https://other.example/v1", apiKey: "test-key", models: ["other"] },
+            "renamed-go": opencodeGo(identity.operator ? { headers: { "X-OpenCode-Session": "operator-session" } } : {}),
+          },
+          combos: { affinity: { strategy, targets: [
+            { provider: "other", model: "other" }, { provider: "renamed-go", model },
+          ] } },
+        } as unknown as OcxConfig;
+        const entropy = spyOn(Math, "random").mockReturnValue(0.9);
+        // Preliminary route checks the first target; dispatch independently picks Go.
+        entropy.mockReturnValueOnce(0);
+        try {
+          const response = await handleClaudeMessages(new Request("http://localhost/v1/messages", {
+            method: "POST", headers: { "content-type": "application/json", ...identity.headers } as Record<string, string>,
+            body: JSON.stringify({ model: "combo/affinity", max_tokens: 64, stream: false,
+              messages: [{ role: "user", content: "ping" }],
+              system: "Shared system prompt is not a session.",
+              metadata: { user_id: identity.metadata } }),
+          }), config, { model: "", provider: "" });
+          await response.text();
+          expect(response.status).toBe(200);
+          expect(requests.at(-1)?.url).toStartWith("https://opencode.ai/zen/go/v1/");
+          expect(requests.at(-1)?.headers.get(SESSION_HEADER)).toBe(identity.expected);
+          if (strategy === "failover") {
+            expect(requests).toHaveLength(2);
+            expect(requests[0]?.headers.has(SESSION_HEADER)).toBe(false);
+          } else {
+            expect(requests).toHaveLength(1);
+          }
+        } finally {
+          entropy.mockRestore();
+          clearComboSelectionState();
+          clearComboTargetCooldowns();
+        }
+      });
+      }
+    }
+    }
+
+    test(`Claude random Go preflight does not leak affinity to a final non-Go Responses target (${model})`, async () => {
+      clearComboSelectionState();
+      clearComboTargetCooldowns();
+      const requests: Headers[] = [];
+      globalThis.fetch = (async (input: RequestInfo | URL, init?: RequestInit) => {
+        expect(String(input)).toBe("https://other.example/v1/responses");
+        requests.push(new Headers(init?.headers));
+        return upstreamResponse(String(input));
+      }) as typeof fetch;
+      const config = {
+        providers: {
+          other: { adapter: "openai-responses", authMode: "key", baseUrl: "https://other.example/v1", apiKey: "test-key", models: ["other"] },
+          "renamed-go": opencodeGo(),
+        },
+        combos: { affinity: { strategy: "random", targets: [
+          { provider: "renamed-go", model }, { provider: "other", model: "other" },
+        ] } },
+      } as unknown as OcxConfig;
+      const entropy = spyOn(Math, "random").mockReturnValue(0.9).mockReturnValueOnce(0);
+      try {
+        const response = await handleClaudeMessages(new Request("http://localhost/v1/messages", {
+          method: "POST", headers: { "content-type": "application/json", [SESSION_HEADER]: "client-session-a" },
+          body: JSON.stringify({ model: "combo/affinity", max_tokens: 64, stream: false,
+            messages: [{ role: "user", content: "ping" }],
+            metadata: { user_id: "user_test_account__session_conversation-a" } }),
+        }), config, { model: "", provider: "" });
+        await response.text();
+        expect(response.status).toBe(200);
+        expect(requests).toHaveLength(1);
+        expect(requests[0]?.has(SESSION_HEADER)).toBe(false);
+        expect(requests[0]?.has("session_id")).toBe(false);
+      } finally {
+        entropy.mockRestore();
+        clearComboSelectionState();
+        clearComboTargetCooldowns();
+      }
+    });
+  }
+
   test("Claude metadata gives stable Go affinity across turns and distinct conversations", async () => {
     const input = { claude: true, model: CHAT_MODEL, metadataUserId: "user_test_account__session_conversation-a" };
     const first = await captureRequest(input);