fix: keep codex stdio server alive across turns

src/selftest/cases/50-hwlab-manual-dispatch.ts

@@ -89,14 +89,19 @@ console.log(JSON.stringify({ apiVersion: manifest.apiVersion, kind: manifest.kin
     assert.equal(((resumedRun.sessionRef as JsonRecord).threadId), "thread_selftest_1");
 
     const multiTurn = await createHwlabRun(client, context, bundle, "hwlab-session-multiturn", "hello first turn", "hwlab-command-multiturn-1");
-    const multiturnRunner = runOnce({ managerUrl: server.baseUrl, runId: multiTurn.runId, commandId: multiTurn.commandId, codexCommand: context.fakeCodexCommand, codexArgs: context.fakeCodexArgs, codexHome: context.codexHome, env: { CODEX_HOME: context.codexHome, AGENTRUN_WORKSPACE_ROOT: path.join(context.tmp, "workspaces-multiturn") }, idleTimeoutMs: 500, pollIntervalMs: 50 });
+    const fakeCodexStartFile = path.join(context.tmp, "fake-codex-starts-multiturn.txt");
+    const multiturnRunner = runOnce({ managerUrl: server.baseUrl, runId: multiTurn.runId, commandId: multiTurn.commandId, codexCommand: context.fakeCodexCommand, codexArgs: context.fakeCodexArgs, codexHome: context.codexHome, env: { CODEX_HOME: context.codexHome, AGENTRUN_WORKSPACE_ROOT: path.join(context.tmp, "workspaces-multiturn"), AGENTRUN_FAKE_CODEX_START_FILE: fakeCodexStartFile }, idleTimeoutMs: 500, pollIntervalMs: 50 });
     await waitForCommandState(client, multiTurn.runId, multiTurn.commandId, "completed");
     const secondCommand = await client.post(`/api/v1/runs/${multiTurn.runId}/commands`, { type: "turn", payload: { prompt: "hello second turn", traceId: "hwlab-command-multiturn-2" }, idempotencyKey: "hwlab-command-multiturn-2" }) as { id: string };
     await waitForCommandState(client, multiTurn.runId, secondCommand.id, "completed");
     const multiturnResult = await multiturnRunner as JsonRecord;
     assert.equal(multiturnResult.commandsProcessed, 2);
+    const starts = (await readTextIfExists(fakeCodexStartFile)).trim().split("\n").filter(Boolean);
+    assert.equal(starts.length, 1, "same-run multiturn runner must keep one codex app-server process alive instead of restarting per command");
     const multiEventsResponse = await client.get(`/api/v1/runs/${multiTurn.runId}/events?afterSeq=0&limit=200`) as { items?: Array<{ type?: string; payload?: JsonRecord }> };
     const multiEvents = multiEventsResponse.items ?? [];
+    assert.equal(multiEvents.filter((event) => event.type === "backend_status" && event.payload?.phase === "codex-app-server-starting").length, 1);
+    assert.equal(multiEvents.filter((event) => event.type === "backend_status" && event.payload?.phase === "codex-app-server:reused").length, 1);
     assert.equal(multiEvents.filter((event) => event.type === "backend_status" && event.payload?.phase === "resource-bundle-materialized").length, 1);
     assert.equal(multiEvents.filter((event) => event.type === "backend_status" && event.payload?.phase === "command-terminal").length, 2);
     const secondEnvelope = await client.get(`/api/v1/runs/${multiTurn.runId}/commands/${secondCommand.id}/result`) as JsonRecord;
@@ -159,4 +164,12 @@ async function waitForCommandState(client: ManagerClient, runId: string, command
   throw new Error(`command ${commandId} did not reach ${state}`);
 }
 
+async function readTextIfExists(filePath: string): Promise<string> {
+  try {
+    return await readFile(filePath, "utf8");
+  } catch {
+    return "";
+  }
+}
+
 export default selfTest;

src/selftest/fake-codex-app-server.ts

@@ -1,7 +1,9 @@
 import * as readline from "node:readline";
+import { appendFileSync } from "node:fs";
 
 const rl = readline.createInterface({ input: process.stdin, crlfDelay: Infinity });
 const mode = process.env.AGENTRUN_FAKE_CODEX_MODE ?? "success";
+if (process.env.AGENTRUN_FAKE_CODEX_START_FILE) appendFileSync(process.env.AGENTRUN_FAKE_CODEX_START_FILE, `${process.pid}\n`);
 let threadCounter = 0;
 let turnCounter = 0;
 let observedThreadModel = false;