fix(agent): cancel sandbox work on host termination

docs/reference/commands.mdx

@@ -1277,6 +1277,8 @@ Otherwise, it writes the captured output to the corresponding host streams and r
 The in-sandbox NemoClaw plugin writes its registration banner to `stderr`, so the banner does not prefix the agent reply on `stdout` in non-JSON mode.
 Because a delivered turn always writes to one of the two streams, the wrapper reports a dispatch with status `0` and no output as a failure.
 The wrapper prints recovery guidance to `stderr` and exits with status `1`.
+Pressing `Ctrl+C` interrupts the OpenShell child, and sending `SIGTERM` to the host wrapper forwards `SIGTERM` to that child.
+NemoClaw waits for OpenShell to stop the in-sandbox turn, replays captured output, and returns status `130` for `SIGINT` or `143` for `SIGTERM`.
 The diagnostic shell-quotes the sandbox name and forwarded arguments, then redacts detected credential values before writing the recovery command to `stderr`.
 If redaction changes the recovery command, the diagnostic tells you not to replay it; otherwise, it labels the command as runnable inside the sandbox.
 For a registered sandbox, both captured paths pin the sandbox's recorded gateway with an explicit `-g`.
@@ -4516,7 +4518,7 @@ OpenClaw-specific onboarding configuration:
 | `NEMOCLAW_WEB_SEARCH_PROVIDER` | `brave`, `tavily`, or `none` | Selects Brave Search or Tavily Search in non-interactive onboarding, or disables web search explicitly. When unset, `BRAVE_API_KEY` implicitly selects Brave before `TAVILY_API_KEY` can implicitly select Tavily. |
 | `BRAVE_API_KEY` | Brave Search API key | Supplies and implicitly selects Brave Search when no web search provider is set. NemoClaw validates the key and stores it in OpenShell rather than the sandbox. |
 | `TAVILY_API_KEY` | Tavily Search API key | Supplies and implicitly selects Tavily Search when no provider is set and no Brave key is available. NemoClaw validates the key and stores it in OpenShell rather than the sandbox. |
-| `NEMOCLAW_AGENT_TIMEOUT` | positive integer (seconds) | Overrides `agents.defaults.timeoutSeconds` and `models.providers.<provider-id>.timeoutSeconds` in the built OpenClaw config. Raise for slow inference. |
+| `NEMOCLAW_AGENT_TIMEOUT` | positive integer (seconds) | Build-time setting that overrides `agents.defaults.timeoutSeconds` and `models.providers.<provider-id>.timeoutSeconds` in the built OpenClaw config. Set it before onboarding builds the sandbox image. Setting it only for a later `$$nemoclaw <name> agent` invocation does not change the existing image. Raise for slow inference. |
 | `NEMOCLAW_MCP_SHADOW_DIAGNOSTICS` | literal `1` to enable | Forwards opt-in successful Streamable HTTP MCP timing diagnostics to a newly created or rebuilt OpenClaw sandbox. It does not change timeouts, retries, requests, or responses. Unset it and rebuild after evidence collection to restore failure-only logging. Other values are ignored. |
 | `NEMOCLAW_AUTO_PAIR_SLOW_INTERVAL_SECS` | positive number of seconds | Sets the post-pairing poll cadence for the in-sandbox OpenClaw auto-pair watcher. Defaults to `5` so late allowlisted CLI and browser scope upgrades are approved before clients time out. Raise only on load-sensitive gateways. |
 | `NEMOCLAW_AUTO_PAIR_FAST_REENTRY_POLLS` | positive integer | Sets how many fast polls run after the watcher observes a fresh allowlisted scope-upgrade request. Defaults to `5`; set lower only when you need to reduce gateway polling. |

src/lib/actions/sandbox/agent/passthrough-dispatch.test.ts

@@ -1,13 +1,115 @@
 // SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 // SPDX-License-Identifier: Apache-2.0
 
-import { describe, expect, it } from "vitest";
+import { EventEmitter } from "node:events";
+
+import { describe, expect, it, vi } from "vitest";
 
 import {
+  type AgentDispatchChild,
   agentDispatchStdio,
   isSilentAgentDispatch,
+  runAgentDispatch,
   SILENT_AGENT_DISPATCH_EXIT_CODE,
 } from "./passthrough-dispatch";
+import { computeExitCode, type SandboxExecSignalSource } from "../exec";
+
+function dispatchHarness() {
+  const childEvents = new EventEmitter();
+  const signalEvents = new EventEmitter();
+  const stderr = new EventEmitter();
+  const stdout = new EventEmitter();
+  const child: AgentDispatchChild = {
+    exitCode: null,
+    signalCode: null,
+    kill: vi.fn((signal) => {
+      child.signalCode = signal;
+      queueMicrotask(() => childEvents.emit("close", null, signal));
+      return true;
+    }),
+    once: ((event: string, listener: (...args: unknown[]) => void) =>
+      childEvents.once(event, listener)) as AgentDispatchChild["once"],
+    stderr,
+    stdout,
+  };
+  const signalSource: SandboxExecSignalSource = {
+    add: (signal, listener) => signalEvents.on(signal, listener),
+    remove: (signal, listener) => signalEvents.off(signal, listener),
+  };
+  return { child, signalEvents, signalSource, stderr, stdout };
+}
+
+describe("runAgentDispatch", () => {
+  it("forwards host SIGTERM to OpenShell and captures output before signal exit (#8723)", async () => {
+    const harness = dispatchHarness();
+    const pending = runAgentDispatch(
+      "openshell",
+      ["sandbox", "exec", "--name", "alpha", "--", "openclaw", "agent"],
+      { stdinIsTty: true },
+      { signalSource: harness.signalSource, spawnChild: () => harness.child },
+    );
+
+    harness.stdout.emit("data", "partial response\n");
+    harness.stderr.emit("data", Buffer.from("gateway timeout pending\n"));
+    harness.signalEvents.emit("SIGTERM");
+
+    const result = await pending;
+    expect(harness.child.kill).toHaveBeenCalledOnce();
+    expect(harness.child.kill).toHaveBeenCalledWith("SIGTERM");
+    expect(result).toMatchObject({
+      status: null,
+      signal: "SIGTERM",
+      stdout: "partial response\n",
+      stderr: "gateway timeout pending\n",
+    });
+    expect(harness.signalEvents.listenerCount("SIGTERM")).toBe(0);
+    expect(harness.signalEvents.listenerCount("SIGINT")).toBe(0);
+  });
+
+  it("terminates the OpenShell child when captured output exceeds its bound", async () => {
+    const harness = dispatchHarness();
+    const pending = runAgentDispatch(
+      "openshell",
+      ["sandbox", "exec", "--name", "alpha", "--", "openclaw", "agent"],
+      { maxBufferBytes: 4, stdinIsTty: false },
+      { signalSource: harness.signalSource, spawnChild: () => harness.child },
+    );
+
+    harness.stdout.emit("data", "12345");
+
+    const result = await pending;
+    expect(harness.child.kill).toHaveBeenCalledWith("SIGTERM");
+    expect(result.error).toEqual(
+      new Error("agent output exceeded the 4-byte combined capture limit"),
+    );
+    expect(computeExitCode(result)).toEqual({
+      code: 1,
+      errorMessage: "agent output exceeded the 4-byte combined capture limit",
+    });
+    expect(result.stdout).toBe("");
+  });
+
+  it("enforces one capture bound across stdout and stderr", async () => {
+    const harness = dispatchHarness();
+    const pending = runAgentDispatch(
+      "openshell",
+      ["sandbox", "exec", "--name", "alpha", "--", "openclaw", "agent"],
+      { maxBufferBytes: 6, stdinIsTty: false },
+      { signalSource: harness.signalSource, spawnChild: () => harness.child },
+    );
+
+    harness.stdout.emit("data", "1234");
+    harness.stderr.emit("data", "567");
+
+    const result = await pending;
+    expect(harness.child.kill).toHaveBeenCalledWith("SIGTERM");
+    expect(result.error).toEqual(
+      new Error("agent output exceeded the 6-byte combined capture limit"),
+    );
+    expect(result.stdout).toBe("1234");
+    expect(result.stderr).toBe("");
+  });
+});
 
 describe("isSilentAgentDispatch", () => {
   it("classifies a zero-exit dispatch with no bytes on either stream as silent", () => {

src/lib/actions/sandbox/agent/passthrough-dispatch.ts

@@ -3,10 +3,10 @@
 
 // Source-of-truth boundary for the agent dispatch contract (#8796).
 //
-// Both `nemoclaw <name> agent` transports capture the child's streams and
-// forward its exit code. That makes "the child exited 0" the only success
-// signal, so a dispatch that never ran the turn is indistinguishable from a
-// turn that ran and answered.
+// Both `nemoclaw <name> agent` transports capture the child's streams, forward
+// host termination signals, and return the child's exit status. OpenClaw can
+// still report status 0 when a dispatch produces no result, so the wrapper
+// must classify that ambiguous result before it reports success.
 //
 // 6. Empty-dispatch guard (delivery contract).
 //
@@ -41,25 +41,174 @@
 //    - Removal condition: drop the TTY carve-out if `openclaw agent` gains a
 //      documented interactive stdin mode reachable through this wrapper.
 //
+// 8. Host interruption propagation.
+//
+//    - Invalid state: the former synchronous transports blocked the Node.js
+//      event loop. A host SIGTERM stopped NemoClaw without notifying the
+//      OpenShell child, so the in-sandbox agent turn continued until its own
+//      deadline.
+//    - Source boundary: OpenShell owns remote command cancellation. NemoClaw
+//      owns its direct child and uses the shared sandbox exec supervisor to
+//      forward SIGTERM, wait for OpenShell to exit, and return exit 143.
+//    - Removal condition: none while NemoClaw owns the host-side OpenShell
+//      child lifecycle.
+//
 // Regression tests: `passthrough-dispatch.test.ts` owns the classifier and the
-// stdio shape; `passthrough-help.test.ts` owns the diagnostic text.
+// supervised process lifecycle; `passthrough-help.test.ts` owns the diagnostic
+// text.
 
-import type { StdioOptions } from "node:child_process";
+import { spawn, type StdioOptions } from "node:child_process";
 
 import { isStdinTty } from "../../../core/stdin";
+import { runSandboxExecChild, type SandboxExecChild, type SandboxExecSignalSource } from "../exec";
 
 /**
  * Exit code for a dispatch that reported success without delivering a turn.
  * Matches the wrapper's other non-recoverable dispatch failures.
  */
 export const SILENT_AGENT_DISPATCH_EXIT_CODE = 1;
 
-/** The subset of a `spawnSync` return the delivery classifier reads. */
+/** The subset of a child-process result the delivery classifier reads. */
 export type AgentDispatchOutcome = {
-  error?: unknown;
+  error?: Error;
   status: number | null;
+  signal?: NodeJS.Signals | null;
+};
+
+export type AgentDispatchResult = AgentDispatchOutcome & {
+  stderr: string;
+  stdout: string;
+};
+
+type AgentDispatchReadable = {
+  on(event: "data", listener: (chunk: Buffer | string) => void): unknown;
+};
+
+type AgentDispatchCaptureBudget = {
+  bytes: number;
+  overflowed: boolean;
+};
+
+export type AgentDispatchChild = SandboxExecChild & {
+  stderr: AgentDispatchReadable | null;
+  stdout: AgentDispatchReadable | null;
+};
+
+export type AgentDispatchSpawner = (
+  binary: string,
+  args: readonly string[],
+  stdio: StdioOptions,
+) => AgentDispatchChild;
+
+export type AgentDispatchRunner = (
+  binary: string,
+  args: readonly string[],
+  options?: {
+    maxBufferBytes?: number;
+    stdinIsTty?: boolean;
+  },
+) => Promise<AgentDispatchResult>;
+
+export type AgentDispatchRunDeps = {
+  signalSource?: SandboxExecSignalSource;
+  spawnChild?: AgentDispatchSpawner;
 };
 
+const DEFAULT_AGENT_DISPATCH_MAX_BUFFER_BYTES = 64 * 1024 * 1024;
+
+const defaultAgentDispatchSpawner: AgentDispatchSpawner = (binary, args, stdio) =>
+  spawn(binary, [...args], { stdio }) as unknown as AgentDispatchChild;
+
+function captureAgentDispatchStream(
+  stream: AgentDispatchReadable | null,
+  child: AgentDispatchChild,
+  chunks: Buffer[],
+  maxBufferBytes: number,
+  budget: AgentDispatchCaptureBudget,
+  setOverflowError: (error: Error) => void,
+): void {
+  stream?.on("data", (chunk) => {
+    if (budget.overflowed) return;
+    const data = Buffer.isBuffer(chunk) ? chunk : Buffer.from(chunk);
+    const nextSize = budget.bytes + data.byteLength;
+    if (nextSize > maxBufferBytes) {
+      budget.overflowed = true;
+      setOverflowError(
+        new Error(`agent output exceeded the ${maxBufferBytes}-byte combined capture limit`),
+      );
+      if (child.exitCode === null && child.signalCode === null) child.kill("SIGTERM");
+      return;
+    }
+    budget.bytes = nextSize;
+    chunks.push(data);
+  });
+}
+
+/**
+ * Capture one agent dispatch while the shared sandbox exec supervisor forwards
+ * host termination signals to OpenShell and waits for the child to exit.
+ */
+export async function runAgentDispatch(
+  binary: string,
+  args: readonly string[],
+  options: {
+    maxBufferBytes?: number;
+    stdinIsTty?: boolean;
+  } = {},
+  deps: AgentDispatchRunDeps = {},
+): Promise<AgentDispatchResult> {
+  const stderrChunks: Buffer[] = [];
+  const stdoutChunks: Buffer[] = [];
+  const captureBudget: AgentDispatchCaptureBudget = { bytes: 0, overflowed: false };
+  let overflowError: Error | undefined;
+  const maxBufferBytes = options.maxBufferBytes ?? DEFAULT_AGENT_DISPATCH_MAX_BUFFER_BYTES;
+  const spawnChild = deps.spawnChild ?? defaultAgentDispatchSpawner;
+  const result = await runSandboxExecChild(
+    binary,
+    args,
+    { tty: false },
+    (runBinary, runArgs) => {
+      const child = spawnChild(
+        runBinary,
+        runArgs,
+        agentDispatchStdio(options.stdinIsTty ?? isStdinTty()),
+      );
+      const setOverflowError = (error: Error) => {
+        overflowError ??= error;
+      };
+      captureAgentDispatchStream(
+        child.stdout,
+        child,
+        stdoutChunks,
+        maxBufferBytes,
+        captureBudget,
+        setOverflowError,
+      );
+      captureAgentDispatchStream(
+        child.stderr,
+        child,
+        stderrChunks,
+        maxBufferBytes,
+        captureBudget,
+        setOverflowError,
+      );
+      return child;
+    },
+    deps.signalSource,
+  );
+  try {
+    return {
+      status: result.status,
+      signal: result.signal,
+      ...(result.error || overflowError ? { error: result.error ?? overflowError } : {}),
+      stderr: Buffer.concat(stderrChunks).toString("utf-8"),
+      stdout: Buffer.concat(stdoutChunks).toString("utf-8"),
+    };
+  } finally {
+    result.releaseSignals?.();
+  }
+}
+
 /**
  * Stdio for a non-interactive agent dispatch. An interactive terminal is
  * withheld from fd 0; a genuine pipe or redirect is still forwarded so