fix(runner): keep stopped provider checkpoint closed

packages/paperclip-runner/runner/crates/runner-core/src/acpx_provider_backend.rs

@@ -971,7 +971,11 @@ impl AcpxCommandExecutor {
             .as_mut()
             .expect("ACPX state remains available after provider termination");
         state.active_turn_id = None;
-        state.lifecycle = "suspended".to_owned();
+        // Keep the provider checkpoint closed while runnerd drains the
+        // already-persisted event suffix. `suspended` is a recoverable ACPX
+        // lifecycle and would make the following runner.drain command restart
+        // the provider generation that turn.stop just proved terminated.
+        state.lifecycle = "prepared".to_owned();
         self.save_state()?;
         Ok(CommandExecution::result(json!({
             "status": "stopped",

packages/paperclip-runner/runner/crates/runner-core/src/provider_backend.rs

@@ -2183,7 +2183,10 @@ impl CodexCommandExecutor {
         state.active_provider_result_fingerprint = None;
         state.active_provider_result_disposition = None;
         state.last_agent_message = None;
-        state.lifecycle = "provider_exited".to_owned();
+        // Do not let the runner.drain command that follows turn.stop restore a
+        // fresh provider process. `prepared` retains the durable thread while
+        // deferring the only authorized restart to the successor run.attach.
+        state.lifecycle = "prepared".to_owned();
         self.save_state()?;
         Ok(CommandExecution::result(json!({
             "status": "stopped",

packages/paperclip-runner/src/live/runnerd-codex-transport.test.ts

@@ -2129,7 +2129,7 @@ it("persists an active provider as settled before bounded suspension", async ()
       ),
     ) as Record<string, unknown>;
     expect(providerState).toMatchObject({
-      lifecycle: "provider_exited",
+      lifecycle: "prepared",
       activeProviderTurnId: null,
     });
   } finally {