fix: fall back from native compact 404 to Responses compaction

docs-site/src/content/docs/fr/guides/pi.md

@@ -27,6 +27,9 @@ d’exportation de la variable d’environnement et le nombre de modèles dotés
       "baseUrl": "http://127.0.0.1:10100/v1",
       "api": "openai-completions",
       "apiKey": "$OPENCODEX_API_KEY",
+      "compat": {
+        "sendSessionAffinityHeaders": true
+      },
       "models": [
         {
           "id": "anthropic/claude-opus-5",
@@ -41,6 +44,8 @@ d’exportation de la variable d’environnement et le nombre de modèles dotés
 }
 ```
 
+Les fournisseurs Pi générés activent `compat.sendSessionAffinityHeaders`. Conservez ce réglage lors de la fusion ou de la modification manuelle du fournisseur : Pi transmet un identifiant de session stable, dont OpenCodex dérive l’affinité pour la destination canonique OpenCode Go. Pi peut omettre cet identifiant lorsque `cacheRetention` vaut `none`.
+
 Les identifiants de modèle sont les sélecteurs canoniques du proxy : les modèles routés apparaissent donc sous la forme `provider/model`
 (`anthropic/claude-opus-5`) et les slugs natifs OpenAI restent sans préfixe (`gpt-5.6-sol`). Le `name`
 suffixe — `(anthropic)`, `(native)`, `(routed)` — permet de distinguer, dans le sélecteur de Pi, deux modèles de même nom

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

@@ -27,6 +27,9 @@ export line, and how many models carry authoritative context limits.
       "baseUrl": "http://127.0.0.1:10100/v1",
       "api": "openai-completions",
       "apiKey": "$OPENCODEX_API_KEY",
+      "compat": {
+        "sendSessionAffinityHeaders": true
+      },
       "models": [
         {
           "id": "anthropic/claude-opus-5",
@@ -41,6 +44,8 @@ export line, and how many models carry authoritative context limits.
 }
 ```
 
+Generated Pi providers enable `compat.sendSessionAffinityHeaders`. Keep this flag when merging or manually editing the provider: Pi supplies a stable session identity and OpenCodex derives canonical OpenCode Go affinity from it. Pi may omit the identity when `cacheRetention` is `none`.
+
 Model ids are the proxy's canonical selectors, so routed models appear as `provider/model`
 (`anthropic/claude-opus-5`) and native OpenAI slugs stay unprefixed (`gpt-5.6-sol`). The `name`
 suffix — `(anthropic)`, `(native)`, `(routed)` — is what makes two same-named models from

docs-site/src/content/docs/ja/guides/pi.md

@@ -23,6 +23,9 @@ ocx export --client pi
       "baseUrl": "http://127.0.0.1:10100/v1",
       "api": "openai-completions",
       "apiKey": "$OPENCODEX_API_KEY",
+      "compat": {
+        "sendSessionAffinityHeaders": true
+      },
       "models": [
         {
           "id": "anthropic/claude-opus-5",
@@ -37,6 +40,8 @@ ocx export --client pi
 }
 ```
 
+生成される Pi プロバイダーでは `compat.sendSessionAffinityHeaders` が有効です。設定をマージしたり手動で編集したりする際も、このフラグを保持してください。Pi が送る安定したセッション識別子から、OpenCodex が正規の OpenCode Go 接続先用の affinity を生成します。`cacheRetention` が `none` の場合、Pi は識別子を送信しないことがあります。
+
 モデル ID はプロキシの正規セレクターであるため、ルーティングされたモデルは `provider/model` (`anthropic/claude-opus-5`) として表示され、ネイティブ OpenAI スラグはプレフィックスなし (`gpt-5.6-sol`) のままになります。 `name` サフィックス (`(anthropic)`、`(native)`、`(routed)`) により、異なるアップストリームの 2 つの同じ名前のモデルが Pi のピッカーで区別できるようになります。
 
 ## どこへ行くのか

docs-site/src/content/docs/ko/guides/pi.md

@@ -27,6 +27,9 @@ ocx export --client pi
       "baseUrl": "http://127.0.0.1:10100/v1",
       "api": "openai-completions",
       "apiKey": "$OPENCODEX_API_KEY",
+      "compat": {
+        "sendSessionAffinityHeaders": true
+      },
       "models": [
         {
           "id": "anthropic/claude-opus-5",
@@ -41,6 +44,8 @@ ocx export --client pi
 }
 ```
 
+생성된 Pi provider에는 `compat.sendSessionAffinityHeaders`가 활성화됩니다. provider를 병합하거나 직접 수정할 때 이 설정을 유지하세요. Pi가 안정적인 세션 식별자를 보내면 OpenCodex가 이를 바탕으로 정규 OpenCode Go 대상의 affinity를 계산합니다. `cacheRetention`이 `none`이면 Pi가 식별자를 보내지 않을 수 있습니다.
+
 모델 id는 프록시의 정규 선택자이므로, 라우팅된 모델은 `provider/model`
 (`anthropic/claude-opus-5`) 형태로 나타나고, 네이티브 OpenAI slug는 접두사 없이
 (`gpt-5.6-sol`) 유지됩니다. `name` 접미사인 `(anthropic)`, `(native)`, `(routed)`는

docs-site/src/content/docs/reference/proxy-formats.md

@@ -402,9 +402,16 @@ conversation.
 
 | Route type | Behavior |
 | --- | --- |
-| Canonical ChatGPT or official OpenAI route | Forwards the request to the native `/responses/compact` endpoint with the resolved account and model authentication |
+| Canonical ChatGPT or official OpenAI route | Tries the native `/responses/compact` endpoint with the resolved account and model authentication; HTTP 404 falls back to a regular Responses compaction turn |
 | Other routed model | Runs an internal, non-streaming, no-tools compaction turn with a `compaction_trigger`; requires exactly one synthetic `compaction` item whose `encrypted_content` is an `ocx1:` envelope; decodes that summary into v1 replacement history |
 
+If the native compact endpoint returns HTTP 404, OpenCodex retries compaction through a regular
+Responses turn with the same model selector and session headers. Canonical ChatGPT fallback
+turns use upstream SSE; the compact caller still receives JSON. A completed native opaque
+compaction item is preserved, while an `ocx1:` summary is decoded into replacement user history.
+Failed or incomplete fallback turns return an error instead of replacement history. Other
+native compact statuses retain their existing handling.
+
 Codex names a bare OpenAI-family model (for example `gpt-5.6-sol`) for its compaction turns
 regardless of which provider the operator routes ordinary turns to. Ordinary requests reserve
 such ids for the canonical `openai` provider. On the compaction surface only — `POST

docs-site/src/content/docs/ru/guides/pi.md

@@ -27,6 +27,9 @@ ocx export --client pi
       "baseUrl": "http://127.0.0.1:10100/v1",
       "api": "openai-completions",
       "apiKey": "$OPENCODEX_API_KEY",
+      "compat": {
+        "sendSessionAffinityHeaders": true
+      },
       "models": [
         {
           "id": "anthropic/claude-opus-5",
@@ -41,6 +44,8 @@ ocx export --client pi
 }
 ```
 
+В создаваемой конфигурации Pi включён `compat.sendSessionAffinityHeaders`. Сохраняйте этот флаг при объединении или ручном редактировании провайдера: Pi передаёт стабильный идентификатор сессии, из которого OpenCodex формирует affinity для канонического OpenCode Go. При `cacheRetention: none` Pi может не передавать идентификатор.
+
 Id моделей — это канонические селекторы прокси, поэтому маршрутизируемые модели появляются как
 `provider/model` (`anthropic/claude-opus-5`), а нативные slug OpenAI остаются без префикса
 (`gpt-5.6-sol`). Суффикс в `name` — `(anthropic)`, `(native)`, `(routed)` — как раз и позволяет

docs-site/src/content/docs/tr/guides/pi.md

@@ -31,6 +31,9 @@ export line, and how many models carry authoritative context limits.
       "baseUrl": "http://127.0.0.1:10100/v1",
       "api": "openai-completions",
       "apiKey": "$OPENCODEX_API_KEY",
+      "compat": {
+        "sendSessionAffinityHeaders": true
+      },
       "models": [
         {
           "id": "anthropic/claude-opus-5",
@@ -45,6 +48,8 @@ export line, and how many models carry authoritative context limits.
 }
 ```
 
+Oluşturulan Pi sağlayıcılarında `compat.sendSessionAffinityHeaders` etkinleştirilir. Sağlayıcıyı birleştirirken veya elle düzenlerken bu ayarı koruyun: Pi sabit bir oturum kimliği gönderir ve OpenCodex bu kimlikten kanonik OpenCode Go hedefi için oturum yakınlığı üretir. `cacheRetention` değeri `none` olduğunda Pi kimliği göndermeyebilir.
+
 Model ids are the proxy's canonical selectors, so routed models appear as
 `provider/model`
 (`anthropic/claude-opus-5`) and native OpenAI slugs stay unprefixed

docs-site/src/content/docs/zh-cn/guides/pi.md

@@ -23,6 +23,9 @@ ocx export --client pi
       "baseUrl": "http://127.0.0.1:10100/v1",
       "api": "openai-completions",
       "apiKey": "$OPENCODEX_API_KEY",
+      "compat": {
+        "sendSessionAffinityHeaders": true
+      },
       "models": [
         {
           "id": "anthropic/claude-opus-5",
@@ -37,6 +40,8 @@ ocx export --client pi
 }
 ```
 
+生成的 Pi 提供方配置启用了 `compat.sendSessionAffinityHeaders`。合并或手动编辑提供方时请保留该设置：Pi 提供稳定的会话标识，OpenCodex 据此为规范的 OpenCode Go 目标生成会话亲和标识。`cacheRetention` 为 `none` 时，Pi 可能不发送会话标识。
+
 模型 id 是代理的规范选择器，因此已路由模型会显示为 `provider/model`（`anthropic/claude-opus-5`），而原生 OpenAI slug 会保持不带前缀（`gpt-5.6-sol`）。`name` 后缀 - `(anthropic)`、`(native)`、`(routed)` - 负责让两个同名但来自不同上游的模型在 Pi 的选择器中可区分。
 
 ## 放置位置

docs-site/src/content/docs/zh-tw/guides/pi.md

@@ -23,6 +23,9 @@ ocx export --client pi
       "baseUrl": "http://127.0.0.1:10100/v1",
       "api": "openai-completions",
       "apiKey": "$OPENCODEX_API_KEY",
+      "compat": {
+        "sendSessionAffinityHeaders": true
+      },
       "models": [
         {
           "id": "anthropic/claude-opus-5",
@@ -37,6 +40,8 @@ ocx export --client pi
 }
 ```
 
+產生的 Pi 供應商設定會啟用 `compat.sendSessionAffinityHeaders`。合併或手動編輯供應商時請保留此設定：Pi 提供穩定的工作階段識別碼，OpenCodex 據此為標準 OpenCode Go 目標產生工作階段親和識別碼。當 `cacheRetention` 為 `none` 時，Pi 可能不傳送識別碼。
+
 模型 id 是代理的規範選擇器，因此路由模型顯示為 `provider/model`（`anthropic/claude-opus-5`），而原生 OpenAI slug 保持無前綴（`gpt-5.6-sol`）。`name` 後綴 — `(anthropic)`、`(native)`、`(routed)` — 正是讓來自不同上游的兩個同名模型在 Pi 的 picker 中可區分的關鍵。
 
 ## 放置位置

src/clients/config-export.ts

@@ -695,6 +695,7 @@ export interface PiProviderBlock {
   baseUrl: string;
   api: string;
   apiKey: string;
+  compat?: { sendSessionAffinityHeaders: boolean };
   models: PiModelEntry[];
 }
 
@@ -816,7 +817,7 @@ export interface GajaeGeneratedConfig {
  * model. The rest of this contract (omitting `cost`) is still ours rather than
  * a claim about Pi's acceptance.
  */
-function buildPiClientConfig(ctx: ExportContext): PiGeneratedConfig {
+function buildPiClientConfig(ctx: ExportContext, sendSessionAffinityHeaders = false): PiGeneratedConfig {
   const models: PiModelEntry[] = [];
   for (const model of normalizeExportModels(ctx.models)) {
     // Text is the one modality every routed model supports; anything richer must come
@@ -859,6 +860,7 @@ function buildPiClientConfig(ctx: ExportContext): PiGeneratedConfig {
         baseUrl: ctx.baseUrl,
         api: PI_API_DIALECT,
         apiKey: LOOPBACK_API_KEY_PLACEHOLDER,
+        ...(sendSessionAffinityHeaders ? { compat: { sendSessionAffinityHeaders: true } } : {}),
         models,
       },
     },
@@ -1031,7 +1033,7 @@ function buildOpencodeContribution(ctx: ExportContext): ManagedContribution {
 }
 
 function buildPiContribution(ctx: ExportContext): ManagedContribution {
-  const doc = buildPiClientConfig(ctx);
+  const doc = buildPiClientConfig(ctx, true);
   return singleFragment("pi", ["providers", OPENCODE_PROVIDER_ID], doc.providers[OPENCODE_PROVIDER_ID]);
 }
 
@@ -1127,7 +1129,7 @@ export const EXPORT_CLIENTS: Record<ExportClientId, ExportClientSpec> = {
     destination: env => piConfigPath(env),
     apiKeyEnv: "",
     exportHint: "Pi reads a non-secret placeholder from models.json; loopback needs no key.",
-    build: buildPiClientConfig,
+    build: ctx => buildPiClientConfig(ctx, true),
     format: "json",
     summarize: summarizePi,
     buildContribution: buildPiContribution,

src/server/chat-completions.ts

@@ -25,6 +25,8 @@ import { estimateTokens } from "../lib/token-estimate";
 import { NoEligiblePolicyCandidateError, UnknownRoutingPolicyError, routeModel } from "../router";
 import { evidenceFromBody } from "../routing/request-evidence";
 import { resolveWireProtocolOverride } from "./adapter-resolve";
+import { resolveOpenCodeGoTransport } from "../providers/opencode-go-transport";
+import { normalizeLogConversationId, sessionLaneIdFromRequest } from "./request-log-conversation";
 import type { OcxConfig } from "../types";
 import { readJsonRequestBody } from "./request-decompress";
 import {
@@ -136,6 +138,8 @@ async function handleChatCompletionsWithBudget(
   let chatNativeRoute: ReturnType<typeof routeModel> | null = null;
   try {
     const route = routeModel(config, chatBody.model as string, evidenceFromBody(chatBody));
+    route.provider = resolveOpenCodeGoTransport(route.provider,
+      sessionLaneIdFromRequest(req.headers) ?? normalizeLogConversationId(req.headers.get("x-opencode-session")));
     // Settle the wire once so every branch below reads the adapter this model will
     // actually use, not the provider-wide default (#404).
     route.provider = resolveWireProtocolOverride(route.providerName, route.modelId, route.provider, "chat");
@@ -237,6 +241,9 @@ async function handleChatCompletionsWithBudget(
     return chatCompletionsErrorResponse(400, CODEX_RESERVE_HELPER_UNSUPPORTED_MESSAGE, "invalid_request_error");
   }
   const headers = new Headers({ "content-type": "application/json" });
+  // Internal bridge metadata; the Go resolver scopes and hashes it before upstream use.
+  const openCodeSession = req.headers.get("x-opencode-session");
+  if (openCodeSession) headers.set("x-opencode-session", openCodeSession);
   for (const name of FORWARD_HEADERS) {
     if (name === "authorization" && !directRoute) continue;
     const value = req.headers.get(name);

src/server/responses/compact.ts

@@ -1094,7 +1094,8 @@ export async function handleResponsesCompact(
         }
       }
     }
-    return buffered;
+    // A native compact 404 falls back to a regular Responses compaction turn.
+    if (buffered.status !== 404) return buffered;
     } finally {
       releaseUpstreamHostAdmission(compactHostAdmissionLease);
       releaseCodexAuthContextProbeLease(authCtx);
@@ -1111,7 +1112,7 @@ export async function handleResponsesCompact(
     // the completed event back into the v1 compact JSON contract below. Combo-dispatched
     // turns also go out as SSE: failover can land on a canonical child that rejects a
     // non-streaming turn, and every combo-capable provider already serves streaming traffic.
-    stream: accountGatedCompactWireModel || route.combo ? true : false,
+    stream: isCanonicalOpenAiForwardProvider(route.provider) || accountGatedCompactWireModel || route.combo ? true : false,
     input: [...inputItems, { type: "compaction_trigger" }],
   };
   const internalHeaders = new Headers({ "content-type": "application/json" });

src/server/responses/core.ts

@@ -2304,7 +2304,8 @@ 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));
+  route.provider = resolveOpenCodeGoTransport(route.provider,
+    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;

tests/clients/prime-client.test.ts

@@ -37,18 +37,14 @@ function context(): ExportContext {
 }
 
 describe("Prime Agent client config", () => {
-  /**
-   * The load-bearing claim of this client: Prime Agent is the pi coding agent
-   * under a different brand, so it reads the SAME models.json contract rather
-   * than a lookalike. Locking the two documents together is what keeps that
-   * claim true — if a future Pi-only change diverges, this fails here instead
-   * of silently shipping Prime users a config their agent rejects.
-   */
-  test("generates byte-for-byte the document Pi generates", () => {
-    const prime = buildClientConfigText("prime", context());
-    const pi = buildClientConfigText("pi", context());
-    expect(prime.format).toBe("json");
-    expect(prime.text).toBe(pi.text);
+  test("shares Pi's model contract without opting Prime into session headers", () => {
+    const prime = buildClientConfig("prime", context()) as PiGeneratedConfig;
+    const pi = buildClientConfig("pi", context()) as PiGeneratedConfig;
+    expect(pi.providers[OPENCODE_PROVIDER_ID]!.compat).toEqual({ sendSessionAffinityHeaders: true });
+    delete pi.providers[OPENCODE_PROVIDER_ID]!.compat;
+    expect(prime).toEqual(pi);
+    expect(buildClientContribution("prime", context()).fragments[0]!.value)
+      .toEqual(prime.providers[OPENCODE_PROVIDER_ID]);
   });
 
   test("adds only providers.opencodex, wired to the loopback proxy", () => {

tests/config/client-config-export.test.ts

@@ -11,6 +11,7 @@ import {
   LOOPBACK_API_KEY_PLACEHOLDER,
   SCHEMA_REQUIRED_OUTPUT_BUDGET,
   buildClientConfig,
+  buildClientContribution,
   buildClientConfigText,
   isExportClientId,
   normalizeExportModels,
@@ -315,6 +316,8 @@ describe("Pi serializer (accept criterion 2)", () => {
     expect(provider.baseUrl).toBe(BASE_URL);
     expect(provider.api).toBe("openai-completions");
     expect(provider.apiKey).toBe(LOOPBACK_API_KEY_PLACEHOLDER);
+    expect(provider.compat?.sendSessionAffinityHeaders).toBe(true);
+    expect(buildClientContribution("pi", ctx()).fragments[0]!.value).toEqual(provider);
   });
 
   test("cost is omitted on every entry — zeros would assert routed models are free", () => {
@@ -899,7 +902,7 @@ describe("EXPORT_CLIENTS registry", () => {
 `);
   });
 
-  test("pi bytes are unchanged, to the last newline", () => {
+  test("pi bytes include session affinity, to the last newline", () => {
     const built = buildClientConfigText("pi", ctx({ config: cfg() }));
     expect(built.format).toBe("json");
     expect(built.text).toBe(`{
@@ -908,6 +911,9 @@ describe("EXPORT_CLIENTS registry", () => {
       "baseUrl": "http://127.0.0.1:10100/v1",
       "api": "openai-completions",
       "apiKey": "opencodex-loopback",
+      "compat": {
+        "sendSessionAffinityHeaders": true
+      },
       "models": [
         {
           "id": "anthropic/claude-opus-5",