fix(pi): close history-only runtimes after replay

Archived timeline reads resume a Pi RPC process to reconstruct the saved
conversation. The history-only session remained open after replay, so
browsing or synchronizing archived agents left one idle Pi process per
agent.

Mark history-purpose resumes for closure and close the runtime in
streamHistory's finally path. Interactive sessions remain resident, and
idempotent close keeps later manager cleanup safe.

packages/server/src/server/agent/providers/pi/agent.test.ts

@@ -1600,6 +1600,46 @@ describe("PiRpcAgentSession", () => {
     ]);
   });
 
+  test("closes a history-only Pi runtime after replaying its timeline", async () => {
+    const pi = new FakePi();
+    pi.queueSessionSetup((runtimeSession) => {
+      runtimeSession.messages = [
+        { role: "user", content: "Old question" },
+        { role: "assistant", content: [{ type: "text", text: "Old answer" }] },
+      ];
+    });
+
+    const client: import("../../agent-sdk-types.js").AgentClient = createClient(pi);
+    const session = await client.resumeSession(
+      {
+        provider: "pi",
+        sessionId: "pi-session-1",
+        nativeHandle: "/tmp/native-pi-session",
+        metadata: { cwd: "/workspace/project" },
+      },
+      {},
+      undefined,
+      { purpose: "history" },
+    );
+
+    const history: AgentStreamEvent[] = [];
+    for await (const event of session.streamHistory()) history.push(event);
+
+    expect(history).toEqual(
+      expect.arrayContaining([
+        expect.objectContaining({
+          type: "timeline",
+          provider: "pi",
+          item: expect.objectContaining({ type: "assistant_message", text: "Old answer" }),
+        }),
+      ]),
+    );
+    expect(pi.latestSession().closeCalls).toBe(1);
+
+    await session.close();
+    expect(pi.latestSession().closeCalls).toBe(1);
+  });
+
   test("resumes a session whose model was removed on the model Pi falls back to", async () => {
     const pi = new FakePi();
     pi.removeModel("9router/deepseek-v4-flash");

packages/server/src/server/agent/providers/pi/agent.ts

@@ -26,6 +26,7 @@ import {
   type AgentRunOptions,
   type AgentRunResult,
   type AgentRuntimeInfo,
+  type AgentResumeSessionOptions,
   type AgentSelectOption,
   type AgentSession,
   type AgentSessionConfig,
@@ -253,6 +254,7 @@ interface PiRpcAgentSessionOptions {
   currentModeId?: string | null;
   persistedSessionFile?: string;
   cleanup?: () => void;
+  closeAfterHistory?: boolean;
   extensionTimeoutMs?: number;
   logger: Logger;
   usagePollScheduler?: PiUsagePollScheduler;
@@ -1255,6 +1257,7 @@ export class PiRpcAgentSession implements AgentSession {
     this.currentModeId = options.currentModeId ?? null;
     this.persistedSessionFile = options.persistedSessionFile;
     this.cleanup = options.cleanup;
+    this.closeAfterHistory = options.closeAfterHistory ?? false;
     this.extensionTimeoutMs = options.extensionTimeoutMs ?? DEFAULT_PI_EXTENSION_RESULT_TIMEOUT_MS;
     this.logger = options.logger;
     this.extensionHost = createPiExtensionHost(this.logger);
@@ -1286,6 +1289,7 @@ export class PiRpcAgentSession implements AgentSession {
   private readonly config: AgentSessionConfig;
   private readonly persistedSessionFile?: string;
   private readonly cleanup?: () => void;
+  private readonly closeAfterHistory: boolean;
   private readonly extensionTimeoutMs: number;
 
   get id(): string | null {
@@ -1454,16 +1458,20 @@ export class PiRpcAgentSession implements AgentSession {
   }
 
   async *streamHistory(): AsyncGenerator<AgentStreamEvent> {
-    await this.requestEntryCapture("history");
-    yield* streamPiHistory(
-      this.provider,
-      await this.runtimeSession.getMessages(),
-      this.contextUserEntries,
-      {},
-      // At most eight 2 MiB child files per replay; later cards retain their summaries.
-      createPiExtensionHost(this.logger, undefined, 16 * 1024 * 1024),
-      this.closeController.signal,
-    );
+    try {
+      await this.requestEntryCapture("history");
+      yield* streamPiHistory(
+        this.provider,
+        await this.runtimeSession.getMessages(),
+        this.contextUserEntries,
+        {},
+        // At most eight 2 MiB child files per replay; later cards retain their summaries.
+        createPiExtensionHost(this.logger, undefined, 16 * 1024 * 1024),
+        this.closeController.signal,
+      );
+    } finally {
+      if (this.closeAfterHistory) await this.close();
+    }
   }
 
   async getRuntimeInfo(): Promise<AgentRuntimeInfo> {
@@ -2738,6 +2746,7 @@ export class PiRpcAgentClient implements AgentClient {
         capabilities: capabilitiesForSession(mcp !== null),
         persistedSessionFile: sessionCopy ? sessionFile : undefined,
         cleanup: cleanupTempFiles,
+        closeAfterHistory: options?.purpose === "history",
         extensionTimeoutMs: providerOptions.extensionTimeoutMs,
         logger: this.logger,
         usagePollScheduler: this.usagePollScheduler,

packages/server/src/server/agent/providers/pi/test-utils/fake-pi.ts

@@ -153,6 +153,7 @@ export class FakePiSession implements PiRuntimeSession {
   // The user entries on the current branch that getMessages() replays.
   contextUserEntries: FakePiUserEntry[] = [];
   abortRequested = false;
+  closeCalls = 0;
   readonly canceledExtensionUiRequests: string[] = [];
   readonly extensionUiResponses: Array<{
     id: string;
@@ -422,7 +423,9 @@ export class FakePiSession implements PiRuntimeSession {
     this.respondToExtensionUiRequest(id, { cancelled: true });
   }
 
-  async close(): Promise<void> {}
+  async close(): Promise<void> {
+    this.closeCalls += 1;
+  }
 
   emit(event: PiRuntimeEvent): void {
     for (const subscriber of this.subscribers) {