refactor(agent): build the InfiniteSession once in the serve loop, thread it

hive-ag3nt/src/bin/hive.rs

@@ -503,6 +503,9 @@ async fn serve_loop<S: Surface>(
 ) -> Result<()> {
     tracing::info!(socket = %socket.display(), "harness serve");
     S::requeue_inflight(socket).await;
+    // The durable claude session, built once and reused for every turn +
+    // idle compaction below (it's effectively stateless).
+    let session = turn::make_session(&bus);
     // Set when a turn calls `request_next_turn` and no real work is
     // pending — the next iteration drives this synthetic message
     // in-process instead of long-polling the broker. Never
@@ -519,7 +522,7 @@ async fn serve_loop<S: Surface>(
                     // (the in-flight case is handled at the end of drive_turn).
                     let compacted = {
                         let _guard = turn_lock.lock().await;
-                        turn::run_pending_compact(files, &bus).await
+                        turn::run_pending_compact(files, &bus, &session).await
                     };
                     if !compacted {
                         tokio::time::sleep(interval).await;
@@ -545,6 +548,7 @@ async fn serve_loop<S: Surface>(
                         stats.as_ref(),
                         files,
                         &turn_lock,
+                        &session,
                         graceful_stop_message(),
                     )
                     .await;
@@ -553,7 +557,8 @@ async fn serve_loop<S: Surface>(
                 }
             },
         };
-        let ctrl = handle_turn::<S>(socket, &bus, stats.as_ref(), files, &turn_lock, next).await;
+        let ctrl =
+            handle_turn::<S>(socket, &bus, stats.as_ref(), files, &turn_lock, &session, next).await;
         if ctrl.auth_failed {
             *login_state.lock().unwrap() = LoginState::NeedsLogin;
             login::wait_for_login(
@@ -582,6 +587,7 @@ async fn handle_turn<S: Surface>(
     stats: Option<&TurnStats>,
     files: &turn::TurnFiles,
     turn_lock: &TurnLock,
+    session: &turn::AgentSession,
     first: hive_sh4re::DeliveredMessage,
 ) -> TurnControl {
     let from = first.from;
@@ -603,7 +609,7 @@ async fn handle_turn<S: Surface>(
     let prompt = serve_common::format_wake_prompt(msg_id, &from, &body, unread, redelivered);
     let outcome = {
         let _guard = turn_lock.lock().await;
-        turn::drive_turn(&prompt, files, bus).await
+        turn::drive_turn(&prompt, files, bus, session).await
     };
     turn::emit_turn_end(bus, &outcome);
     bus.set_state(TurnState::Idle);

hive-ag3nt/src/turn.rs

@@ -230,11 +230,20 @@ fn compact_percent() -> u8 {
     u8::try_from(pct.min(100)).unwrap_or(DEFAULT_COMPACT_PERCENT)
 }
 
-/// Build the agent's durable session: constant title + on-disk store + a
+/// The agent's durable session type: the constant-title [`InfiniteSession`]
+/// with hyperhive's percent-of-window compaction policy. Built once by the
+/// serve loop (see [`make_session`]) and threaded through the turns, rather
+/// than rebuilt each time — it's effectively stateless, so one instance serves
+/// the whole run.
+pub type AgentSession = InfiniteSession<PercentPolicy>;
+
+/// Construct the agent's durable session: constant title + on-disk store + a
 /// percent-of-window compaction policy that checkpoints (`CHECKPOINT_PROMPT`)
-/// before compacting. `default_window` feeds the policy the effective window
+/// before compacting. Called once at serve-loop start. `percent` comes from a
-/// for turns where the model didn't itself report one.
+/// boot-time env var and `default_window` is only a fallback for turns where
-fn infinite_session(bus: &Bus) -> InfiniteSession<PercentPolicy> {
+/// the model didn't report a window, so a single build at startup is fine.
+#[must_use]
+pub fn make_session(bus: &Bus) -> AgentSession {
     InfiniteSession::new(
         session_title(),
         session_store(),
@@ -260,8 +269,14 @@ fn infinite_session(bus: &Bus) -> InfiniteSession<PercentPolicy> {
 ///   the whole turn is retried a single time before bubbling `AuthFailed` to
 ///   the serve loop (which parks for re-login).
 ///
-/// Called once per turn by the `hive` serve loop.
+/// Called once per turn by the `hive` serve loop, which owns the shared
-pub async fn drive_turn(prompt: &str, files: &TurnFiles, bus: &Bus) -> TurnOutcome {
+/// `session` ([`make_session`]) and threads it in.
+pub async fn drive_turn(
+    prompt: &str,
+    files: &TurnFiles,
+    bus: &Bus,
+    session: &AgentSession,
+) -> TurnOutcome {
     if bus.take_session_reset() {
         // Operator-requested (deferred from `POST /api/new-session`).
         bus.emit(LiveEvent::Note {
@@ -274,7 +289,6 @@ pub async fn drive_turn(prompt: &str, files: &TurnFiles, bus: &Bus) -> TurnOutco
     }
     let config = claude_config(bus, files);
     let sink = BusSink::new(bus);
-    let session = infinite_session(bus);
     let mut result = session.run(&config, prompt, &sink).await;
     if matches!(result, Err(hive_claude::Error::AuthFailed)) {
         bus.emit(LiveEvent::Note {
@@ -398,7 +412,7 @@ pub fn emit_turn_end(bus: &Bus, outcome: &TurnOutcome) {
 /// case is handled at the end of [`drive_turn`].) Resume-only via
 /// [`InfiniteSession::compact`]: a missing session is a harmless no-op. Returns
 /// `true` if a compaction ran.
-pub async fn run_pending_compact(files: &TurnFiles, bus: &Bus) -> bool {
+pub async fn run_pending_compact(files: &TurnFiles, bus: &Bus, session: &AgentSession) -> bool {
     if !bus.take_compact() {
         return false;
     }
@@ -408,7 +422,7 @@ pub async fn run_pending_compact(files: &TurnFiles, bus: &Bus) -> bool {
     bus.set_state(crate::events::TurnState::Compacting);
     let config = claude_config(bus, files);
     let sink = BusSink::new(bus);
-    match infinite_session(bus).compact(&config, &sink).await {
+    match session.compact(&config, &sink).await {
         Ok(()) => bus.emit(LiveEvent::Note {
             text: "/compact done".into(),
         }),