fix(maestro): preserve dismissed recovery cleanup

crates/maestro-acp/src/lib.rs

@@ -158,6 +158,7 @@ enum Command {
 
 enum ChildCommand {
     Signal(libc::c_int),
+    Reap(oneshot::Sender<bool>),
 }
 
 /// Why the handshake did not produce a session id.
@@ -206,6 +207,18 @@ impl AcpSession {
         self.exit.borrow().is_some()
     }
 
+    /// Reap the child if it has already exited.
+    pub async fn reap_if_exited(&self) -> bool {
+        if self.has_exited() {
+            return true;
+        }
+        let (sent, received) = oneshot::channel();
+        if self.child_commands.send(ChildCommand::Reap(sent)).is_err() {
+            return self.has_exited();
+        }
+        received.await.unwrap_or_else(|_| self.has_exited())
+    }
+
     /// Wait for the child to exit.
     pub async fn wait(&self) -> Option<ExitStatus> {
         wait_exit(&self.exit).await
@@ -392,6 +405,21 @@ async fn supervise_child(
                     // so its PID cannot be reused in that window.
                     unsafe { libc::kill(pid as libc::pid_t, sig) };
                 }
+                Some(ChildCommand::Reap(done)) => match child.try_wait() {
+                    Ok(Some(status)) => {
+                        let _ = done.send(true);
+                        tracing::debug!(pid, ?status, "{ADAPTER_BIN} exited");
+                        let _ = exit_tx.send(Some(status));
+                        return;
+                    }
+                    Ok(None) => {
+                        let _ = done.send(false);
+                    }
+                    Err(error) => {
+                        tracing::warn!(pid, %error, "could not reap {ADAPTER_BIN}");
+                        let _ = done.send(false);
+                    }
+                },
                 None => {
                     // The session vanished without a command receiver. This
                     // task still owns the unreaped child, so this cannot hit

crates/maestro-daemon/src/session.rs

@@ -379,8 +379,26 @@ impl SessionManager {
         if children.is_empty() {
             return;
         }
-        tracing::info!(count = children.len(), "shutdown: ending adapter children");
-        let terminations = children
+        let reaped =
+            futures_util::future::join_all(children.iter().map(|(_, acp, _)| acp.reap_if_exited()))
+                .await;
+        let mut natural_exits = Vec::new();
+        let mut live_children = Vec::new();
+        for (child, exited) in children.iter().cloned().zip(reaped) {
+            if exited {
+                natural_exits.push(child.0);
+            } else {
+                live_children.push(child);
+            }
+        }
+        for id in natural_exits {
+            mark_session_dead_with_worktree_settlement(&self.inner, id, true).await;
+        }
+        tracing::info!(
+            count = live_children.len(),
+            "shutdown: ending adapter children"
+        );
+        let terminations = live_children
             .iter()
             .map(|(_, acp, _)| acp.terminate(maestro_acp::CANCEL_GRACE, maestro_acp::TERM_GRACE));
         futures_util::future::join_all(terminations).await;
@@ -389,7 +407,7 @@ impl SessionManager {
             .filter_map(|(_, _, done)| done.clone())
             .map(wait_for_event_pump);
         futures_util::future::join_all(drains).await;
-        for (id, _, _) in children {
+        for (id, _, _) in live_children {
             mark_session_dead_after_shutdown(&self.inner, id).await;
         }
     }
@@ -930,9 +948,9 @@ impl SessionManager {
             adopted += 1;
         }
 
-        // Structured sessions have no tmux process to discover. Their
-        // sidecars are the recovery inventory: when an ACP id exists, start a
-        // fresh adapter process and load that conversation into it.
+        // Structured sessions have no tmux process to discover. A durable
+        // marker written during graceful shutdown is the only evidence that a
+        // sidecar may be resumed; older and naturally-dead sidecars stay dead.
         for mut session in self
             .inner
             .sidecar
@@ -945,6 +963,7 @@ impl SessionManager {
             }
 
             let id = session.id;
+            let recoverable = self.inner.sidecar.take_recovery_marker(id)?;
             let mut session_env = sanitize_env(&session.env)?;
             let fifo_path = self.inner.fifo_dir.join(format!("{id}.fifo"));
             let bridge =
@@ -960,40 +979,50 @@ impl SessionManager {
                 session_env.insert("MAESTRO_URL".into(), url);
             }
 
-            let (acp, failure) = match session.acp_session_id.clone() {
-                Some(acp_session_id) => {
-                    let config = SpawnConfig::new(
-                        &session.cwd,
-                        session_env
-                            .iter()
-                            .map(|(key, value)| (key.clone(), value.clone()))
-                            .collect(),
-                    )
-                    .load_session(acp_session_id);
-                    match maestro_acp::spawn(config).await {
-                        Ok(acp) => (Some(Arc::new(acp)), None),
-                        Err(error) => {
-                            tracing::warn!(%id, %error, "structured session: adapter failed to resume");
-                            (None, Some(error.to_string()))
+            let (acp, failure) = if recoverable {
+                match session.acp_session_id.clone() {
+                    Some(acp_session_id) => {
+                        let config = SpawnConfig::new(
+                            &session.cwd,
+                            session_env
+                                .iter()
+                                .map(|(key, value)| (key.clone(), value.clone()))
+                                .collect(),
+                        )
+                        .load_session(acp_session_id);
+                        match maestro_acp::spawn(config).await {
+                            Ok(acp) => (Some(Arc::new(acp)), None),
+                            Err(error) => {
+                                tracing::warn!(%id, %error, "structured session: adapter failed to resume");
+                                (None, Some(error.to_string()))
+                            }
                         }
                     }
+                    None => (
+                        None,
+                        Some("structured session has no stored ACP session id".into()),
+                    ),
                 }
-                None => (
-                    None,
-                    Some("structured session has no stored ACP session id".into()),
-                ),
+            } else {
+                (None, None)
             };
-            session.status = if failure.is_some() {
+            session.status = if recoverable {
+                if failure.is_some() {
+                    SessionStatus::Dead
+                } else {
+                    SessionStatus::Running
+                }
+            } else {
                 SessionStatus::Dead
-            } else {
-                SessionStatus::Running
             };
-            session.activity = if failure.is_some() {
-                ActivityState::Error
-            } else {
-                ActivityState::Active
-            };
-            session.activity_detail = failure;
+            if recoverable {
+                session.activity = if failure.is_some() {
+                    ActivityState::Error
+                } else {
+                    ActivityState::Active
+                };
+                session.activity_detail = failure;
+            }
 
             let acp_events = if let Some(acp) = &acp {
                 match self.inner.sidecar.open_events(id) {
@@ -1723,14 +1752,16 @@ async fn mark_session_dead(inner: &Arc<Inner>, id: Uuid) {
 /// A daemon restart terminates ACP children, but it is not a natural adapter
 /// exit: the sidecar is the recovery inventory, including its worktree.
 async fn mark_session_dead_after_shutdown(inner: &Arc<Inner>, id: Uuid) {
-    mark_session_dead_with_worktree_settlement(inner, id, false).await;
+    if mark_session_dead_with_worktree_settlement(inner, id, false).await {
+        let _ = inner.sidecar.mark_for_recovery(id);
+    }
 }
 
 async fn mark_session_dead_with_worktree_settlement(
     inner: &Arc<Inner>,
     id: Uuid,
     settle_worktree: bool,
-) {
+) -> bool {
     let flipped = {
         let mut sessions = inner.sessions.lock().await;
         match sessions.get_mut(&id) {
@@ -1750,7 +1781,7 @@ async fn mark_session_dead_with_worktree_settlement(
         }
     };
     let Some((mut session, worktree)) = flipped else {
-        return;
+        return false;
     };
     tracing::info!(%id, "session process gone — marking dead");
 
@@ -1781,6 +1812,7 @@ async fn mark_session_dead_with_worktree_settlement(
 
     let _ = inner.sidecar.write(&session);
     let _ = inner.events.send(CatalogEvent::SessionUpdated { session });
+    true
 }
 
 /// Apply the exit rule (#82) to a vanished session's worktree: probe what

crates/maestro-daemon/src/sidecar.rs

@@ -56,6 +56,28 @@ impl SidecarStore {
         self.dir.join(format!("{id}.events.jsonl"))
     }
 
+    /// Durable evidence that graceful daemon shutdown ended this structured
+    /// session and the next daemon may resume it.
+    pub fn recovery_path_for(&self, id: Uuid) -> PathBuf {
+        self.dir.join(format!("{id}.recover"))
+    }
+
+    /// Mark `id` as eligible for one structured-session recovery attempt.
+    pub fn mark_for_recovery(&self, id: Uuid) -> Result<(), SidecarError> {
+        std::fs::create_dir_all(&self.dir)?;
+        std::fs::write(self.recovery_path_for(id), [])?;
+        Ok(())
+    }
+
+    /// Consume the graceful-shutdown recovery marker for `id`.
+    pub fn take_recovery_marker(&self, id: Uuid) -> Result<bool, SidecarError> {
+        match std::fs::remove_file(self.recovery_path_for(id)) {
+            Ok(()) => Ok(true),
+            Err(error) if error.kind() == std::io::ErrorKind::NotFound => Ok(false),
+            Err(error) => Err(error.into()),
+        }
+    }
+
     /// Open the event log for `id` for appending, creating it (and the
     /// directory) when missing. One handle per live session: the caller
     /// keeps it for the session's lifetime rather than reopening per
@@ -164,10 +186,11 @@ impl SidecarStore {
         }
     }
 
-    /// Delete the sidecar and the event log for `id`. No-op for whichever
-    /// does not exist.
+    /// Delete the sidecar, event log, and recovery marker for `id`. No-op for
+    /// whichever does not exist.
     pub fn delete(&self, id: Uuid) -> Result<(), SidecarError> {
         remove_if_present(&self.events_path_for(id))?;
+        remove_if_present(&self.recovery_path_for(id))?;
         remove_if_present(&self.path_for(id))
     }
 

crates/maestro-daemon/tests/acp.rs

@@ -232,6 +232,12 @@ async fn recovery_loads_the_acp_session_without_duplicating_history() {
     let old_pid = fake.pid();
     fx.manager.shutdown().await;
     wait_for_exit(old_pid).await;
+    assert!(
+        fx.sidecar_dir()
+            .join(format!("{}.recover", session.id))
+            .exists(),
+        "graceful shutdown must record recovery provenance"
+    );
 
     let revived = SessionManager::new(
         Arc::new(fx.tmux.tmux.clone()),
@@ -498,6 +504,100 @@ async fn shutdown_reaps_every_adapter_child() {
     assert_eq!(sidecar.status, SessionStatus::Dead);
 }
 
+#[tokio::test]
+async fn dismissal_during_shutdown_does_not_restore_recovery_artifacts() {
+    let fx = DaemonFixture::new();
+    let dismissed = Fake::new();
+    let blocker = Fake::new();
+    let cwd = TempDir::new().unwrap();
+    let mut env = dismissed.env();
+    env.insert("FAKE_ACP_EXIT_ON_CANCEL".into(), "1".into());
+
+    let session = fx
+        .manager
+        .create(structured(cwd.path(), env))
+        .await
+        .expect("create");
+    fx.manager
+        .create(structured(cwd.path(), blocker.env()))
+        .await
+        .expect("create blocker");
+
+    let manager = fx.manager.clone();
+    let shutdown = tokio::spawn(async move { manager.shutdown().await });
+    wait_for_async(Duration::from_secs(5), || async {
+        dismissed.log().contains("recv session/cancel")
+    })
+    .await;
+
+    fx.manager.kill(session.id, false).await.expect("dismiss");
+    assert!(
+        !fx.sidecar_dir()
+            .join(format!("{}.json", session.id))
+            .exists(),
+        "dismissal must finish before shutdown settles its stale snapshot"
+    );
+    shutdown.await.expect("shutdown task");
+
+    assert!(!fx.manager.has(session.id).await);
+    assert!(
+        !fx.sidecar_dir()
+            .join(format!("{}.recover", session.id))
+            .exists(),
+        "dismissal must leave no recovery marker"
+    );
+    assert!(
+        !fx.sidecar_dir()
+            .join(format!("{}.json", session.id))
+            .exists(),
+        "dismissal must leave no sidecar"
+    );
+    assert!(
+        !fx.sidecar_dir()
+            .join(format!("{}.events.jsonl", session.id))
+            .exists(),
+        "dismissal must leave no event log"
+    );
+}
+
+#[tokio::test]
+async fn shutdown_does_not_recover_an_adapter_that_already_exited() {
+    let fx = DaemonFixture::new();
+    let fake = Fake::new();
+    let cwd = TempDir::new().unwrap();
+    let mut env = fake.env();
+    env.insert("FAKE_ACP_EXIT_AFTER_NEW".into(), "1".into());
+
+    let session = fx
+        .manager
+        .create(structured(cwd.path(), env))
+        .await
+        .expect("create");
+    wait_for_exit(fake.pid()).await;
+
+    fx.manager.shutdown().await;
+    assert!(
+        !fx.sidecar_dir()
+            .join(format!("{}.recover", session.id))
+            .exists(),
+        "a naturally exited adapter is not recoverable"
+    );
+
+    let revived = SessionManager::new(
+        Arc::new(fx.tmux.tmux.clone()),
+        SidecarStore::new(fx.sidecar_dir()),
+        WorktreeManager::new(None),
+        fx.fifo_dir(),
+    );
+    assert_eq!(revived.recover().await.expect("recover"), 1);
+    assert_eq!(revived.status(session.id).await, Some(SessionStatus::Dead));
+    assert!(
+        !fake.log().contains("recv session/load"),
+        "a naturally dead adapter must not be resumed: {}",
+        fake.log()
+    );
+}
+
 #[tokio::test]
 async fn shutdown_waits_for_structured_creation_and_rejects_new_creates() {
     let fx = DaemonFixture::new();
@@ -600,7 +700,27 @@ async fn adapter_exit_marks_the_session_dead_and_keeps_it_listed() {
     })
     .await;
     assert!(fx.manager.has(session.id).await);
+    assert!(
+        !fx.sidecar_dir()
+            .join(format!("{}.recover", session.id))
+            .exists(),
+        "a naturally exited adapter is never marked for recovery"
+    );
 
-    fx.manager.kill(session.id, false).await.expect("dismiss");
-    assert!(!fx.manager.has(session.id).await);
+    let revived = SessionManager::new(
+        Arc::new(fx.tmux.tmux.clone()),
+        SidecarStore::new(fx.sidecar_dir()),
+        WorktreeManager::new(None),
+        fx.fifo_dir(),
+    );
+    assert_eq!(revived.recover().await.expect("recover"), 1);
+    assert_eq!(revived.status(session.id).await, Some(SessionStatus::Dead));
+    assert!(
+        !fake.log().contains("recv session/load"),
+        "a naturally dead adapter must not be resumed: {}",
+        fake.log()
+    );
+
+    revived.kill(session.id, false).await.expect("dismiss");
+    assert!(!revived.has(session.id).await);
 }

crates/maestro-daemon/tests/sidecar.rs

@@ -68,15 +68,30 @@ fn delete_removes_file_and_is_idempotent() {
     let s = sample(id);
 
     store.write(&s).unwrap();
+    store.mark_for_recovery(id).unwrap();
     assert!(store.path_for(id).exists());
+    assert!(store.recovery_path_for(id).exists());
 
     store.delete(id).unwrap();
     assert!(!store.path_for(id).exists());
+    assert!(!store.recovery_path_for(id).exists());
 
     // second delete is a no-op
     store.delete(id).unwrap();
 }
 
+#[test]
+fn recovery_marker_is_consumed_once() {
+    let dir = TempDir::new().unwrap();
+    let store = SidecarStore::new(dir.path());
+    let id = Uuid::new_v4();
+
+    assert!(!store.take_recovery_marker(id).unwrap());
+    store.mark_for_recovery(id).unwrap();
+    assert!(store.take_recovery_marker(id).unwrap());
+    assert!(!store.take_recovery_marker(id).unwrap());
+}
+
 #[test]
 fn load_all_recovers_every_sidecar_and_skips_garbage() {
     let dir = TempDir::new().unwrap();