fix(interactive-subagents): harden parallel lifecycle delivery

extensions/interactive-subagents/pi-extension/subagents/surface.ts

@@ -22,6 +22,7 @@ type SurfaceBackend = {
   sendCommand(surface: string, command: string): void;
   readScreen(surface: string, lines?: number): string;
   readScreenAsync(surface: string, lines?: number): Promise<string>;
+  isSurfaceAlive(surface: string, signal?: AbortSignal): Promise<boolean>;
   closeSurface(surface: string): void;
 };
 
@@ -87,6 +88,10 @@ export function readScreenAsync(surface: string, lines = 50): Promise<string> {
   return getBackend().readScreenAsync(surface, lines);
 }
 
+export function isSurfaceAlive(surface: string, signal?: AbortSignal): Promise<boolean> {
+  return getBackend().isSurfaceAlive(surface, signal);
+}
+
 export function closeSurface(surface: string): void {
   getBackend().closeSurface(surface);
 }
@@ -106,7 +111,7 @@ export function wrapWithCompletionFile(command: string, completionFile: string):
 export function sendLongCommand(
   surface: string,
   command: string,
-  options?: { scriptPath?: string; scriptPreamble?: string },
+  options?: { scriptPath?: string; scriptPreamble?: string; replaceShell?: boolean },
 ): string {
   const scriptPath =
     options?.scriptPath ??
@@ -122,7 +127,8 @@ export function sendLongCommand(
   scriptParts.push(command);
 
   writeFileSync(scriptPath, scriptParts.join("\n") + "\n", { mode: 0o755 });
-  sendCommand(surface, `bash ${shellEscape(scriptPath)}`);
+  const launch = `${options?.replaceShell ? "exec " : ""}bash ${shellEscape(scriptPath)}`;
+  sendCommand(surface, launch);
   return scriptPath;
 }
 
@@ -159,6 +165,10 @@ type PollOptions = {
   sentinelFile?: string;
   completionFile?: string;
   onTick?: (elapsed: number) => void;
+  onExitDetected?: () => void;
+  livenessCheck?: (surface: string, signal?: AbortSignal) => Promise<boolean>;
+  livenessIntervalMs?: number;
+  livenessFailureThreshold?: number;
 };
 
 /** Avoid launching a terminal CLI reader when a local completion signal is authoritative. */
@@ -177,6 +187,11 @@ export async function pollForExit(
   readSurface: (surface: string, lines?: number) => Promise<string> = readScreenAsync,
 ): Promise<PollResult> {
   const start = Date.now();
+  const livenessCheck = options.livenessCheck ?? isSurfaceAlive;
+  const livenessIntervalMs = options.livenessIntervalMs ?? 5_000;
+  const livenessFailureThreshold = options.livenessFailureThreshold ?? 3;
+  let nextLivenessCheck = start + livenessIntervalMs;
+  let livenessFailures = 0;
 
   for (;;) {
     if (signal.aborted) throw new Error("Aborted while waiting for subagent to finish");
@@ -186,6 +201,7 @@ export async function pollForExit(
         const exitFile = `${options.sessionFile}.exit`;
         if (existsSync(exitFile)) {
           const result = interpretExitSidecar(JSON.parse(readFileSync(exitFile, "utf8")));
+          options.onExitDetected?.();
           rmSync(exitFile, { force: true });
           return result;
         }
@@ -195,6 +211,7 @@ export async function pollForExit(
     if (options.sentinelFile) {
       try {
         if (existsSync(options.sentinelFile)) {
+          options.onExitDetected?.();
           return { reason: "sentinel", exitCode: 0 };
         }
       } catch {}
@@ -203,7 +220,10 @@ export async function pollForExit(
     if (options.completionFile) {
       try {
         const result = readCompletionFile(options.completionFile);
-        if (result) return result;
+        if (result) {
+          options.onExitDetected?.();
+          return result;
+        }
       } catch {}
     }
 
@@ -211,10 +231,24 @@ export async function pollForExit(
       try {
         const screen = await readSurface(surface, 5);
         const match = screen.match(/__SUBAGENT_DONE_(\d+)__/);
-        if (match) return { reason: "sentinel", exitCode: Number.parseInt(match[1], 10) };
+        if (match) {
+          options.onExitDetected?.();
+          return { reason: "sentinel", exitCode: Number.parseInt(match[1], 10) };
+        }
       } catch {
         // Surface reads are a best-effort fallback when no completion file is available.
       }
+    } else if (Date.now() >= nextLivenessCheck) {
+      nextLivenessCheck = Date.now() + livenessIntervalMs;
+      try {
+        livenessFailures = await livenessCheck(surface, signal) ? 0 : livenessFailures + 1;
+      } catch {
+        livenessFailures += 1;
+      }
+      if (signal.aborted) throw new Error("Aborted while checking subagent surface liveness");
+      if (livenessFailures >= livenessFailureThreshold) {
+        throw new Error("Subagent terminal surface closed before completion.");
+      }
     }
 
     const elapsed = Math.floor((Date.now() - start) / 1000);