acpmux: wait for live daemon peers

cmux-tui/crates/acpmux/src/cli/command.rs

@@ -246,8 +246,8 @@ pub enum Command {
         /// What to do when the turn asks for a permission: wait (default), deny, or fail (exit 5).
         #[arg(long, default_value = "wait")]
         on_permission: String,
-        /// Report prompt_stalled (exit 1) when nothing happens for this many seconds after sending; 0 disables.
-        #[arg(long, default_value_t = 30)]
+        /// Report prompt_stalled (exit 1) when nothing happens for this many seconds after sending; 0 disables. The default waits for the live peer.
+        #[arg(long, default_value_t = 0)]
         stall: u64,
         /// Send this prompt id again (from a `daemon_closed` error): the daemon runs an id once and
         /// answers a repeat with the first run's outcome.
@@ -2003,3 +2003,27 @@ mod config_set_tests {
         }
     }
 }
+
+#[cfg(test)]
+mod deadline_defaults_tests {
+    use super::{Cli, Command};
+    use clap::Parser;
+
+    #[test]
+    fn send_waits_for_a_live_peer_by_default() {
+        let cli = Cli::try_parse_from(["acpmux", "send", "session", "slow reply"]).unwrap();
+        match cli.command {
+            Some(Command::Send { stall, .. }) => assert_eq!(stall, 0),
+            _ => panic!("expected send command"),
+        }
+    }
+
+    #[test]
+    fn explicit_stall_remains_a_caller_choice() {
+        let cli = Cli::try_parse_from(["acpmux", "send", "session", "--stall", "7"]).unwrap();
+        match cli.command {
+            Some(Command::Send { stall, .. }) => assert_eq!(stall, 7),
+            _ => panic!("expected send command"),
+        }
+    }
+}

cmux-tui/crates/acpmux/src/cli/command_new.rs

@@ -68,8 +68,8 @@ pub struct NewArgs {
     /// Retry the turn on an agent-internal error, only when it produced nothing yet.
     #[arg(long, default_value_t = 0)]
     pub retries: u32,
-    /// Report prompt_stalled (exit 1) when nothing happens for this many seconds; 0 disables.
-    #[arg(long, default_value_t = 30)]
+    /// Report prompt_stalled (exit 1) when nothing happens for this many seconds; 0 disables. The default waits for the live peer.
+    #[arg(long, default_value_t = 0)]
     pub stall: u64,
     /// Send the first prompt with this id again (from a `daemon_closed` error); it runs once.
     #[arg(long)]

cmux-tui/crates/acpmux/src/cli/run/standard.rs

@@ -173,17 +173,11 @@ pub(super) async fn run(cmd: Command, json_out: bool, suppress_reads: bool) -> R
         }
         Command::Last { session, count } => {
             let client = connect(true).await?;
-            // One budget includes resolution and every transcript page. Each forwarded read
-            // has its own bound too, but many slow pages must not multiply this CLI's wait.
-            let (id, replies) = tokio::time::timeout(std::time::Duration::from_secs(30), async {
-                let id = resolve_id(&client, &session).await?;
-                let replies = last_replies(&client, &id, count).await?;
-                Ok::<_, anyhow::Error>((id, replies))
-            })
-            .await
-            .map_err(|_| anyhow!(
-                "reading last replies for {session:?} did not finish within 30 seconds; check the peer connection and retry this read (the session may still be running)"
-            ))??;
+            // The daemon connection is the liveness boundary: a live peer may
+            // take as long as it needs to page a large transcript, while EOF
+            // still fails the request promptly.
+            let id = resolve_id(&client, &session).await?;
+            let replies = last_replies(&client, &id, count).await?;
             if json_out {
                 print_json(&json!({"sessionId": id, "replies": replies}));
             } else {