@agent-native/core 0.84.0 → 0.84.2

package/corpus/README.md

@@ -28,4 +28,4 @@ rg -n "defineAction|useActionQuery" node_modules/@agent-native/core/corpus
 ## Generated Counts
 
 - core files: 2043
-- template files: 4877
+- template files: 4902

package/corpus/core/CHANGELOG.md

@@ -1,5 +1,18 @@
 # @agent-native/core
 
+## 0.84.2
+
+### Patch Changes
+
+- 3ff4f55: Extend Builder gateway timeouts for local and background agent-chat runs, and surface recoverable gateway timeout failures instead of silently retrying activity-only tool preparation loops.
+- 3ff4f55: Allow scoped chat surfaces to hide the composer scope badge while keeping thread context attached.
+
+## 0.84.1
+
+### Patch Changes
+
+- 87806ef: Report intentionally gated eval cases as skipped, keep skipped suites from setting up agent runners, and avoid action/engine setup for eval cases that fully short-circuit with custom run handlers.
+
 ## 0.84.0
 
 ### Minor Changes

package/corpus/core/docs/content/evals.mdx

@@ -90,13 +90,14 @@ export default defineEval({
 
 An eval passes only when **every** scorer meets the threshold. Key `defineEval` fields:
 
-| Field       | Type                  | Notes                                                         |
-| ----------- | --------------------- | ------------------------------------------------------------- |
-| `name`      | string                | Required. Shown in the report.                                |
-| `input`     | `{ prompt, history }` | Required `prompt`; optional prior `{ role, text }` turns.     |
-| `scorers`   | `Scorer[]`            | Required, at least one.                                       |
-| `threshold` | number `0..1`         | Per-scorer pass bar. Default `0.5`; overridable from the CLI. |
-| `run`       | function              | Optional override for custom setup (seed data, multi-turn).   |
+| Field        | Type                  | Notes                                                                                                           |
+| ------------ | --------------------- | --------------------------------------------------------------------------------------------------------------- |
+| `name`       | string                | Required. Shown in the report.                                                                                  |
+| `input`      | `{ prompt, history }` | Required `prompt`; optional prior `{ role, text }` turns.                                                       |
+| `scorers`    | `Scorer[]`            | Required, at least one, unless `skipReason` is set.                                                             |
+| `threshold`  | number `0..1`         | Per-scorer pass bar. Default `0.5`; overridable from the CLI.                                                   |
+| `run`        | function              | Optional override for custom setup (seed data, multi-turn).                                                     |
+| `skipReason` | string                | Optional reason to skip this eval without running the agent; useful for manual/nightly gates that need secrets. |
 
 The agent run handed to scorers is small and transport-agnostic:
 
@@ -225,11 +226,32 @@ The command discovers `**/*.eval.ts` and `evals/*.ts` under the current app, run
 
 Exit codes:
 
-| Code | Meaning                                                         |
-| ---- | --------------------------------------------------------------- |
-| `0`  | All evals passed — _or_ no eval files were found (CI-friendly). |
-| `1`  | At least one eval scored below threshold, or the suite errored. |
-| `2`  | Bad arguments (e.g. `--threshold` outside `[0, 1]`).            |
+| Code | Meaning                                                                          |
+| ---- | -------------------------------------------------------------------------------- |
+| `0`  | All evals passed, evals were skipped, or no eval files were found (CI-friendly). |
+| `1`  | At least one eval scored below threshold, or the suite errored.                  |
+| `2`  | Bad arguments (e.g. `--threshold` outside `[0, 1]`).                             |
+
+### Skipping gated evals {#skipping}
+
+Use `skipReason` when an eval suite is intentionally gated behind manual,
+nightly, or secret-backed execution:
+
+```ts
+export default defineEval({
+  name: "provider-backed source sync",
+  input: { prompt: "Refresh the source and summarize the changes." },
+  skipReason: process.env.SOURCE_SYNC_EVALS
+    ? undefined
+    : "Skipped because SOURCE_SYNC_EVALS is unset",
+  scorers: process.env.SOURCE_SYNC_EVALS ? [contains("updated")] : [],
+});
+```
+
+Skipped evals do not run the agent or scorers, do not fail the suite, and are
+reported distinctly with `status: "skipped"` and `skipReason` in JSON output.
+Use this for expensive or credential-dependent evals; keep cheap deterministic
+coverage in normal PR CI.
 
 ### As a CI deploy gate {#ci}
 

package/corpus/core/docs/content/locales/ar-SA/evals.mdx

@@ -231,6 +231,13 @@ agent-native eval --threshold 0.8    # override every eval's pass threshold (0..
 | `1`   | تم تسجيل تقييم واحد على الأقل أقل من الحد الأدنى، أو حدث خطأ في المجموعة.           |
 | `2`   | وسيطات غير صالحة (على سبيل المثال، `--threshold` خارج `[0, 1]`).                    |
 
+### تخطي التقييمات المحكومة {#skipping}
+
+استخدم `skipReason` عندما تكون مجموعة التقييمات مخصصة عمدا للتشغيل اليدوي أو
+الليلي أو المعتمد على الأسرار. لا تشغل التقييمات المتخطاة الوكيل أو أدوات
+التسجيل، ولا تفشل المجموعة، وتظهر في JSON مع `status: "skipped"` و
+`skipReason`. أبق التغطية الحتمية الرخيصة في CI العادي لطلبات السحب.
+
 ### كبوابة نشر CI {#ci}
 
 قم بإضافته إلى المسار الذي يتم تشغيله قبل النشر:

package/corpus/core/docs/content/locales/de-DE/evals.mdx

@@ -231,6 +231,14 @@ Exit-Codes:
 | `1`  | Mindestens eine Bewertung lag unter dem Schwellenwert oder die Suite ist fehlerhaft.                |
 | `2`  | Ungültige Argumente (z. B. `--threshold` außerhalb von `[0, 1]`).                                   |
 
+### Gated Evals überspringen {#skipping}
+
+Verwenden Sie `skipReason`, wenn eine Eval-Suite absichtlich nur manuell,
+nachts oder mit Secrets ausgeführt werden soll. Übersprungene Evals führen
+weder Agent noch Scorer aus, lassen die Suite nicht fehlschlagen und erscheinen
+in der JSON-Ausgabe mit `status: "skipped"` und `skipReason`. Günstige
+deterministische Abdeckung bleibt im normalen PR-CI.
+
 ### Als CI-Deploy-Gate {#ci}
 
 Fügen Sie es der Pipeline hinzu, die vor einer Bereitstellung ausgeführt wird:

package/corpus/core/docs/content/locales/es-ES/evals.mdx

@@ -231,6 +231,14 @@ Códigos de salida:
 | `1`    | Al menos una evaluación obtuvo una puntuación inferior al umbral o la suite tuvo un error.                |
 | `2`    | Malos argumentos (por ejemplo, `--threshold` fuera de `[0, 1]`).                                          |
 
+### Omitir evaluaciones gated {#skipping}
+
+Use `skipReason` cuando una suite de evaluaciones esté reservada
+intencionalmente para ejecución manual, nocturna o con secretos. Las
+evaluaciones omitidas no ejecutan el agente ni los scorers, no hacen fallar la
+suite y aparecen en JSON con `status: "skipped"` y `skipReason`. Mantenga la
+cobertura determinista barata en el CI normal de PR.
+
 ### Como puerta de implementación de CI {#ci}
 
 Agréguelo al proceso que se ejecuta antes de una implementación:

package/corpus/core/docs/content/locales/fr-FR/evals.mdx

@@ -231,6 +231,14 @@ Codes de sortie :
 | `1`  | Au moins une évaluation a obtenu un score inférieur au seuil, ou la suite a commis une erreur.      |
 | `2`  | Mauvais arguments (par exemple `--threshold` en dehors de `[0, 1]`).                                |
 
+### Ignorer les évaluations gated {#skipping}
+
+Utilisez `skipReason` lorsqu'une suite d'évaluations est volontairement réservée
+à une exécution manuelle, nocturne ou dépendante de secrets. Les évaluations
+ignorées n'exécutent ni l'agent ni les scorers, ne font pas échouer la suite, et
+apparaissent dans la sortie JSON avec `status: "skipped"` et `skipReason`.
+Gardez la couverture déterministe peu coûteuse dans le CI normal des PR.
+
 ### En tant que porte de déploiement CI {#ci}
 
 Ajoutez-le au pipeline qui s'exécute avant un déploiement :

package/corpus/core/docs/content/locales/hi-IN/evals.mdx

@@ -229,6 +229,14 @@ agent-native eval --threshold 0.8    # override every eval's pass threshold (0..
 | `1` | कम से कम एक ईवल ने सीमा से नीचे स्कोर किया, या सुइट में त्रुटि हुई। |
 | `2` | खराब तर्क (उदाहरण के लिए `[0, 1]` के बाहर `--threshold`)।           |
 
+### gated evals छोड़ना {#skipping}
+
+जब कोई eval suite केवल manual, nightly, या secret-backed execution के लिए हो,
+तो `skipReason` का उपयोग करें। छोड़े गए evals agent या scorers नहीं चलाते,
+suite को fail नहीं करते, और JSON output में `status: "skipped"` और
+`skipReason` के साथ दिखते हैं। सस्ती deterministic coverage को सामान्य PR CI
+में रखें।
+
 ### सीआई परिनियोजन गेट के रूप में {#ci}
 
 इसे उस पाइपलाइन में जोड़ें जो परिनियोजन से पहले चलती है:

package/corpus/core/docs/content/locales/ja-JP/evals.mdx

@@ -231,6 +231,13 @@ agent-native eval --threshold 0.8    # override every eval's pass threshold (0..
 | `1`    | 少なくとも 1 つの評価スコアがしきい値を下回ったか、スイートでエラーが発生しました。             |
 | `2`    | 不正な引数 (例: `[0, 1]` の外の `--threshold`)。                                                |
 
+### gated eval をスキップする {#skipping}
+
+手動、夜間、または secret が必要な eval スイートには `skipReason` を使い
+ます。スキップされた eval はエージェントや scorer を実行せず、スイート
+を失敗させず、JSON 出力では `status: "skipped"` と `skipReason` とし
+て表示されます。安価で決定的なカバレッジは通常の PR CI に残します。
+
 ### CI デプロイゲートとして {#ci}
 
 デプロイ前に実行されるパイプラインに追加します。

package/corpus/core/docs/content/locales/ko-KR/evals.mdx

@@ -231,6 +231,13 @@ agent-native eval --threshold 0.8    # override every eval's pass threshold (0..
 | `1`  | 최소 하나의 평가가 임계값 미만의 점수를 얻었거나 제품군에 오류가 발생했습니다.   |
 | `2`  | 잘못된 인수(예: `[0, 1]` 외부의 `--threshold`).                                  |
 
+### 게이트된 평가 건너뛰기 {#skipping}
+
+수동, 야간 또는 시크릿이 필요한 평가 스위트에는 `skipReason`를 사용하세요.
+건너뛴 평가는 에이전트나 scorer를 실행하지 않고, 스위트를 실패시키지
+않으며, JSON 출력에 `status: "skipped"` 및 `skipReason`로 표시됩니다.
+저렴한 결정적 커버리지는 일반 PR CI에 유지하세요.
+
 ### CI 배포 게이트로 {#ci}
 
 배포 전에 실행되는 파이프라인에 추가하세요:

package/corpus/core/docs/content/locales/pt-BR/evals.mdx

@@ -231,6 +231,14 @@ Códigos de saída:
 | `1`    | Pelo menos uma avaliação com pontuação abaixo do limite ou o conjunto apresentou erro.                     |
 | `2`    | Argumentos incorretos (por exemplo, `--threshold` fora de `[0, 1]`).                                       |
 
+### Ignorar evals gated {#skipping}
+
+Use `skipReason` quando uma suíte de evals for intencionalmente reservada para
+execução manual, noturna ou dependente de secrets. Evals ignorados não executam
+o agente nem os scorers, não fazem a suíte falhar e aparecem no JSON com
+`status: "skipped"` e `skipReason`. Mantenha cobertura determinística barata no
+CI normal de PR.
+
 ### Como porta de implantação de CI {#ci}
 
 Adicione-o ao pipeline executado antes da implantação:

package/corpus/core/docs/content/locales/zh-CN/evals.mdx

@@ -229,6 +229,13 @@ agent-native eval --threshold 0.8    # override every eval's pass threshold (0..
 | `1`  | 至少有一项评估得分低于阈值，或者套件出错。            |
 | `2`  | 错误的参数（例如 `--threshold` 位于 `[0, 1]` 之外）。 |
 
+### 跳过受控评估 {#skipping}
+
+当某个评估套件只应在手动、夜间或具备密钥的环境中运行时，请使用
+`skipReason`。被跳过的评估不会运行代理或 scorer，不会让套件失败，并会
+在 JSON 输出中显示 `status: "skipped"` 和 `skipReason`。廉价的确定性覆
+盖应保留在普通 PR CI 中。
+
 ### 作为 CI 部署门 {#ci}
 
 将其添加到部署之前运行的管道：

package/corpus/core/docs/content/locales/zh-TW/evals.mdx

@@ -229,6 +229,13 @@ agent-native eval --threshold 0.8    # override every eval's pass threshold (0..
 | `1`    | 至少有一項評估得分低於閾值，或者套件出錯。            |
 | `2`    | 錯誤的參數（例如 `--threshold` 位於 `[0, 1]` 之外）。 |
 
+### 跳過受控評估 {#skipping}
+
+當某個評估套件只應在手動、夜間或具備密鑰的環境中執行時，請使用
+`skipReason`。被跳過的評估不會執行代理或 scorer，不會讓套件失敗，並會
+在 JSON 輸出中顯示 `status: "skipped"` 和 `skipReason`。廉價的確定性覆
+蓋應保留在普通 PR CI 中。
+
 ### 作為 CI 部署門 {#ci}
 
 將其新增到部署之前執行的管道：

package/corpus/core/docs/design/durable-agent-runs.md

@@ -83,10 +83,11 @@ terminate with an untruthful "done" state.
 not buy more time — it just converts a graceful hand-off into a hard kill. The
 walls, in order:
 
-1. **Builder model gateway hard cap — ~45s.**
-   `MAX_BUILDER_GATEWAY_TIMEOUT_MS = 45_000` in
-   `packages/core/src/agent/engine/builder-engine.ts`. A single model call is
-   killed at the gateway after 45s. **Not raisable** by the framework.
+1. **Builder model gateway foreground cap — ~45s.**
+   Hosted foreground calls keep a 45s cap in
+   `packages/core/src/agent/engine/builder-engine.ts`. Local/non-hosted and
+   proven background-function calls use the longer background-style cap because
+   they are not constrained by the synchronous function wall.
 2. **Serverless function kill — ~60–65s.** The hosting function is terminated
    shortly after; the heartbeat then reaps the run row as `stale_run`.
 
@@ -366,8 +367,8 @@ The whole `auto_continue` / 40s soft-timeout dance exists to stay under the
 **serverless function wall** (~60–65s synchronous on Netlify), not under any
 model limit. Evidence:
 
-- `builder-engine.ts:62` `MAX_LOCAL_BUILDER_GATEWAY_TIMEOUT_MS = 180_000`
-  (3 min) is allowed locally; the run loop has no inherent reason to stop at 40s.
+- `builder-engine.ts` allows the long background gateway cap in local/non-hosted
+  runs; the run loop has no inherent reason to stop at 40s.
 - `run-manager.ts:58,68` `DEFAULT_HOSTED_RUN_SOFT_TIMEOUT_MS` /
   `HOSTED_SOFT_TIMEOUT_CEILING_MS = 40_000` are pinned just under the function
   wall, and `templates/brain/netlify.toml` sets `[functions."*"] timeout = 75`.
@@ -379,10 +380,9 @@ function there is no ~60s wall, so:
 
 - The agent loop can run for minutes in a single invocation with **few or no
   `auto_continue` continuations**.
-- The per-model-call gateway cap (`MAX_BUILDER_GATEWAY_TIMEOUT_MS = 45_000`,
-  `builder-engine.ts:60`) still applies per call — see
-  [Per-model-call gateway cap](#per-model-call-gateway-cap) — but the _run_ is no
-  longer chopped into 40s chunks.
+- The foreground hosted per-model-call gateway cap still applies per call — see
+  [Per-model-call gateway cap](#per-model-call-gateway-cap) — but the _run_ is
+  no longer chopped into 40s chunks.
 
 This is exactly the Layer 1 worker with a concrete long-lived host: Netlify is
 the durable worker, reached through the existing self-dispatch primitive — the
@@ -527,29 +527,27 @@ Minimal client change, because the reconnect machinery already exists:
 
 ## Per-model-call gateway cap
 
-Even in a 15-min function, a _single_ model call is still capped by the Builder
-gateway: `MAX_BUILDER_GATEWAY_TIMEOUT_MS = 45_000` (`builder-engine.ts:60`),
-resolved by `getBuilderGatewayTimeoutMs()` (`:770`) /
-`resolveMaxBuilderGatewayTimeoutMs()` (`:758`). Locally the cap is already
-`180_000` (`:62`) when the gateway base URL is localhost.
+Builder gateway calls now use a runtime-aware cap. Hosted foreground calls keep
+the 45s cap so the synchronous function can still checkpoint before its hard
+wall. Local/non-hosted runs use a longer local cap, and proven Netlify
+background-function runs may use a background cap below the 15-minute function
+wall. The run-manager's 13-minute background soft-timeout should fire before
+the gateway cap in normal durable background operation.
 
 Options, in scope-order:
 
-1. **In scope / safe now:** keep the 45s per-call cap. A 15-min function makes
-   _many_ 45s calls; the cap bounds one call, not the run. This alone delivers
-   the goal (long multi-step runs finish) with **zero gateway changes**.
-2. **Flag, out of scope until confirmed:** raise the per-call cap toward 180s in
-   the background context. `getBuilderGatewayTimeoutMs()` already honors
-   `AGENT_NATIVE_BUILDER_GATEWAY_TIMEOUT_MS`, but it is hard-clamped to
-   `resolveMaxBuilderGatewayTimeoutMs()` = 45s for non-localhost gateways. Raising
-   it requires the **hosted Builder gateway to actually allow >45s upstream** —
-   unverified, and not the framework's call. Until confirmed, do not raise it.
-3. **Escape hatch:** a direct Anthropic engine (no Builder gateway) has no 45s
-   cap, so a single very long call could use it. Out of scope for this change;
-   note it as the lever if a single model call truly needs >45s.
-
-**Recommendation:** ship with option 1. The win is removing the _run-level_
-ceiling; the per-call cap is orthogonal and rarely the binding constraint.
+1. **Foreground hosted:** keep the 45s cap. The synchronous function wall is the
+   constraint, and a larger cap would turn graceful checkpointing into a hard
+   platform kill.
+2. **Local/non-hosted:** allow longer calls by default. Local development should
+   not inherit a serverless wall it does not have.
+3. **Durable background:** allow longer calls only when the runtime proves it is
+   inside the emitted background function. Keep the cap below 15 minutes and
+   above the 13-minute run soft-timeout so background checkpointing still owns
+   logical-turn continuation.
+
+**Recommendation:** preserve this split. Do not raise the foreground cap; use
+durable background for long-running tool-input generation.
 
 ## Idempotency / dedup
 
@@ -626,7 +624,8 @@ only): foreground handler inserts the run row, dispatches, returns the SSE
 stream from `subscribeToRun`. The background `_process-run` claims the run and
 calls the same `startRun` + `runAgentLoop` the inline path uses, with
 `softTimeoutMs ≈ 13min` (new `backgroundFunction` option in
-`resolveRunSoftTimeoutMs`). Per-call gateway cap stays at 45s (option 1).
+`resolveRunSoftTimeoutMs`). Builder gateway calls use the background cap only
+when the worker proves it is inside the emitted background function.
 **Done when:** a long multi-step turn that thrashes today completes in one
 background invocation with zero `auto_continue`, events streaming to the client
 via the SQL-poll path, terminal `done` persisted.
@@ -645,10 +644,10 @@ dispatches, and background→background `auto_continue` chaining for the rare
 > 13-min turn (mirror `agent-teams.ts:1886`). Internal checkpointing (Option A)
 > for monotonic progress across any continuation.
 
-**Slice 4 (optional, separate decision) — raise the per-call gateway cap.**
-Only after confirming the hosted Builder gateway accepts >45s upstream; wire
-`AGENT_NATIVE_BUILDER_GATEWAY_TIMEOUT_MS` through with a background-context max.
-Out of scope until that confirmation exists.
+**Slice 4 — foreground remains capped.**
+Do not raise the hosted foreground per-call gateway cap. Any future tuning
+should stay in the local/background timeout regime unless the synchronous
+platform wall changes.
 
 ## Open risks / unknowns
 

package/corpus/core/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@agent-native/core",
-  "version": "0.84.0",
+  "version": "0.84.2",
   "description": "Framework for agent-native application development — where AI agents and UI share SQL state, actions, and context",
   "homepage": "https://github.com/BuilderIO/agent-native#readme",
   "bugs": {

package/corpus/core/src/agent/engine/builder-engine.ts

@@ -24,6 +24,7 @@ import {
   normalizeReasoningEffortForModel,
   type ReasoningEffort,
 } from "../../shared/reasoning-effort.js";
+import { isInBackgroundFunctionRuntime } from "../durable-background.js";
 import { BUILDER_MODEL_CONFIG } from "../model-config.js";
 import { getBuilderGatewayRequestHeaders } from "./builder-gateway-headers.js";
 import {
@@ -53,13 +54,20 @@ export const BUILDER_CAPABILITIES: EngineCapabilities = {
 
 export const BUILDER_SUPPORTED_MODELS = BUILDER_MODEL_CONFIG.supportedModels;
 
-// Keep the gateway timeout below the hosted run soft timeout so the agent loop
-// can append a continuation and persist terminal state before serverless hosts
-// (Netlify synchronous Functions are 60s) hard-kill the invocation.
+// Keep the foreground hosted gateway timeout below the synchronous serverless
+// wall so the agent loop can append a continuation and persist terminal state
+// before the host hard-kills the invocation.
 const DEFAULT_BUILDER_GATEWAY_TIMEOUT_MS = 45_000;
-const MAX_BUILDER_GATEWAY_TIMEOUT_MS = 45_000;
-/** Local ai-services has no serverless wall; allow longer streams in dev. */
-const MAX_LOCAL_BUILDER_GATEWAY_TIMEOUT_MS = 180_000;
+const MAX_HOSTED_FOREGROUND_BUILDER_GATEWAY_TIMEOUT_MS = 45_000;
+/**
+ * Netlify background functions have a 15-minute wall. Keep the Builder gateway
+ * ceiling below that, but above the run-manager's 13-minute soft-timeout so
+ * durable background runs checkpoint before this timer wins.
+ */
+const MAX_BACKGROUND_BUILDER_GATEWAY_TIMEOUT_MS = 14 * 60_000;
+/** Local and non-hosted runtimes have no synchronous serverless wall. */
+const MAX_LOCAL_BUILDER_GATEWAY_TIMEOUT_MS =
+  MAX_BACKGROUND_BUILDER_GATEWAY_TIMEOUT_MS;
 const BUILDER_GATEWAY_NETWORK_ERROR_CODE = "builder_gateway_network_error";
 
 export const BUILDER_DEFAULT_MODEL = BUILDER_MODEL_CONFIG.defaultModel;
@@ -315,6 +323,7 @@ class BuilderEngine implements AgentEngine {
       }
 
       yield* parseJsonlStream(reader, opts.model, {
+        abortSignal: gatewayAbort.signal,
         didGatewayTimeout: gatewayAbort.didTimeout,
         gatewayTimeoutMs,
         gatewayUrl,
@@ -430,11 +439,12 @@ async function* emitHttpError(response: Response): AsyncIterable<EngineEvent> {
 // a complete line and the client must still process it.
 async function* readJsonlLines(
   reader: ReadableStreamDefaultReader<Uint8Array>,
+  abortSignal?: AbortSignal,
 ): AsyncIterable<string> {
   const decoder = new TextDecoder();
   let buffer = "";
   while (true) {
-    const { done, value } = await reader.read();
+    const { done, value } = await readStreamChunk(reader, abortSignal);
     if (done) break;
     buffer += decoder.decode(value, { stream: true });
     let newlineIdx = buffer.indexOf("\n");
@@ -457,6 +467,7 @@ async function* parseJsonlStream(
   reader: ReadableStreamDefaultReader<Uint8Array>,
   model: string,
   captureContext: {
+    abortSignal?: AbortSignal;
     didGatewayTimeout?: () => boolean;
     gatewayTimeoutMs?: number;
     gatewayUrl?: URL;
@@ -487,7 +498,10 @@ async function* parseJsonlStream(
   };
 
   try {
-    for await (const line of readJsonlLines(reader)) {
+    for await (const line of readJsonlLines(
+      reader,
+      captureContext.abortSignal,
+    )) {
       let event: any;
       try {
         event = JSON.parse(line);
@@ -722,6 +736,32 @@ async function* parseJsonlStream(
   }
 }
 
+function readStreamChunk(
+  reader: ReadableStreamDefaultReader<Uint8Array>,
+  abortSignal?: AbortSignal,
+): Promise<ReadableStreamReadResult<Uint8Array>> {
+  if (!abortSignal) return reader.read();
+  if (abortSignal.aborted) {
+    return Promise.reject(abortSignal.reason ?? new Error("Stream aborted"));
+  }
+  return new Promise((resolve, reject) => {
+    const onAbort = () => {
+      reject(abortSignal.reason ?? new Error("Stream aborted"));
+    };
+    abortSignal.addEventListener("abort", onAbort, { once: true });
+    reader.read().then(
+      (value) => {
+        abortSignal.removeEventListener("abort", onAbort);
+        resolve(value);
+      },
+      (err) => {
+        abortSignal.removeEventListener("abort", onAbort);
+        reject(err);
+      },
+    );
+  });
+}
+
 function normalizeGatewayErrorText(raw: string, status: number): string {
   const text = raw.trim();
   const looksHtml = /<html[\s>]|<body[\s>]|<head[\s>]/i.test(text);
@@ -761,6 +801,14 @@ export function createBuilderEngine(
 }
 
 function resolveMaxBuilderGatewayTimeoutMs(): number {
+  if (isInBackgroundFunctionRuntime()) {
+    return MAX_BACKGROUND_BUILDER_GATEWAY_TIMEOUT_MS;
+  }
+
+  if (!isHostedBuilderRuntime()) {
+    return MAX_LOCAL_BUILDER_GATEWAY_TIMEOUT_MS;
+  }
+
   try {
     const base = getBuilderGatewayBaseUrl();
     if (/^https?:\/\/(localhost|127\.0\.0\.1)([:/]|$)/i.test(base)) {
@@ -769,7 +817,31 @@ function resolveMaxBuilderGatewayTimeoutMs(): number {
   } catch {
     // ignore malformed override
   }
-  return MAX_BUILDER_GATEWAY_TIMEOUT_MS;
+  return MAX_HOSTED_FOREGROUND_BUILDER_GATEWAY_TIMEOUT_MS;
+}
+
+function isHostedBuilderRuntime(): boolean {
+  if (
+    process.env.NETLIFY &&
+    process.env.NETLIFY !== "false" &&
+    process.env.NETLIFY_LOCAL !== "true"
+  ) {
+    return true;
+  }
+  if (
+    process.env.AWS_LAMBDA_FUNCTION_NAME &&
+    process.env.NETLIFY_LOCAL !== "true"
+  ) {
+    return true;
+  }
+  return Boolean(
+    process.env.CF_PAGES ||
+    process.env.VERCEL ||
+    process.env.VERCEL_ENV ||
+    process.env.RENDER ||
+    process.env.FLY_APP_NAME ||
+    process.env.K_SERVICE,
+  );
 }
 
 function getBuilderGatewayTimeoutMs(): number {

package/corpus/core/src/agent/run-manager.ts

@@ -46,8 +46,8 @@ const CLEANUP_DELAY_MS = 5 * 60 * 1000;
  *
  * This MUST fire before the two upstream hard walls that otherwise kill a run
  * mid-turn with no chance to hand off:
- *   1. The Builder model gateway hard-caps a single model call at 45s
- *      (builder-engine.ts MAX_BUILDER_GATEWAY_TIMEOUT_MS) — not raisable.
+ *   1. The Builder model gateway keeps a 45s cap only for hosted foreground
+ *      runs; local and proven background-function runs use longer caps.
  *   2. Serverless functions are hard-killed around 60-65s (the heartbeat then
  *      reaps the row as a stale_run).
  * Production data showed every cutoff landing in the 44-70s window with ZERO
@@ -61,11 +61,12 @@ export const DEFAULT_HOSTED_RUN_SOFT_TIMEOUT_MS = 40_000;
 
 /**
  * Hard ceiling for the hosted soft timeout. On a hosted runtime the
- * auto_continue soft timeout can never usefully exceed this — the gateway
- * (45s) / function (~60s) walls kill the run first, so a larger configured or
- * env value just guarantees the cutoff is a hard error instead of a graceful
- * hand-off. Any resolved value above this is clamped down when hosted. Local
- * dev (non-hosted) is left alone so long-running local turns aren't chunked.
+ * foreground auto_continue soft timeout can never usefully exceed this — the
+ * synchronous function (~60s) wall kills the run first, so a larger configured
+ * or env value just guarantees the cutoff is a hard error instead of a graceful
+ * hand-off. Any resolved value above this is clamped down for hosted foreground
+ * runs. Local dev (non-hosted) is left alone so long-running local turns aren't
+ * chunked.
  *
  * IMPORTANT: this clamp is for the INTERACTIVE / foreground path and must NOT
  * be raised. The foreground POST still rides a synchronous serverless function