huskies: merge 1196 bug Inactivity watchdog kills agents mid tool-call: awaited MCP calls produce no PTY output

server/src/agents/pty/mod.rs

@@ -247,6 +247,104 @@ mod tests {
         );
     }
 
+    // ── story 1196: in-flight MCP tool calls suspend the inactivity deadline ──
+
+    /// A `tool_use` block in an `assistant` message marks a call as in
+    /// flight; the inactivity deadline must be suspended while it is
+    /// outstanding (so a slow MCP tool like `run_tests` doesn't get the
+    /// agent killed), and must resume as soon as the matching `tool_result`
+    /// arrives in a `user` message.
+    ///
+    /// Script: emits a tool_use, sleeps 2s (would fail a 1s timeout without
+    /// suspension), emits the matching tool_result, then sleeps 2s again
+    /// (now with no call in flight, the 1s timeout must fire).
+    #[tokio::test]
+    async fn tool_call_in_flight_suspends_then_resumes_inactivity_deadline() {
+        use std::os::unix::fs::PermissionsExt;
+
+        let tmp = tempfile::tempdir().unwrap();
+        let script = tmp.path().join("tool_call_then_silence.sh");
+        let body = "#!/bin/sh\n\
+             printf '%s\\n' '{\"type\":\"assistant\",\"message\":{\"content\":[{\"type\":\"tool_use\",\"id\":\"tool1\",\"name\":\"run_tests\",\"input\":{}}]}}'\n\
+             sleep 2\n\
+             printf '%s\\n' '{\"type\":\"user\",\"message\":{\"content\":[{\"type\":\"tool_result\",\"tool_use_id\":\"tool1\",\"content\":\"ok\"}]}}'\n\
+             sleep 2\n";
+        std::fs::write(&script, body).unwrap();
+        std::fs::set_permissions(&script, std::fs::Permissions::from_mode(0o755)).unwrap();
+
+        let (tx, _rx) = broadcast::channel::<AgentEvent>(64);
+        let (watcher_tx, _watcher_rx) = broadcast::channel::<WatcherEvent>(16);
+        let event_log = Arc::new(Mutex::new(Vec::new()));
+
+        let result = run_agent_pty_streaming(
+            "1196_story_tool_call_in_flight",
+            "coder-1",
+            "sh",
+            &[script.to_string_lossy().to_string()],
+            "--",
+            "/tmp",
+            &tx,
+            &event_log,
+            None,
+            1, // inactivity_timeout_secs = 1s
+            watcher_tx,
+            None,
+            None,
+        )
+        .await;
+
+        match result {
+            Err(err) => assert!(
+                err.contains("inactivity timeout"),
+                "expected an inactivity timeout error after the tool call resolved, got: {err}"
+            ),
+            Ok(_) => panic!(
+                "agent must still be killed once the tool call resolves and \
+                 the process falls genuinely silent again"
+            ),
+        }
+    }
+
+    /// A genuinely hung agent (no output at all, no tool call in flight)
+    /// must still be killed by the inactivity watchdog after the configured
+    /// timeout — the suspension added for in-flight tool calls must not
+    /// mask a real hang.
+    #[tokio::test]
+    async fn genuinely_silent_agent_with_no_in_flight_call_is_killed() {
+        use std::os::unix::fs::PermissionsExt;
+
+        let tmp = tempfile::tempdir().unwrap();
+        let script = tmp.path().join("silent.sh");
+        std::fs::write(&script, "#!/bin/sh\nsleep 2\n").unwrap();
+        std::fs::set_permissions(&script, std::fs::Permissions::from_mode(0o755)).unwrap();
+
+        let (tx, _rx) = broadcast::channel::<AgentEvent>(64);
+        let (watcher_tx, _watcher_rx) = broadcast::channel::<WatcherEvent>(16);
+        let event_log = Arc::new(Mutex::new(Vec::new()));
+
+        let result = run_agent_pty_streaming(
+            "1196_story_genuine_hang",
+            "coder-1",
+            "sh",
+            &[script.to_string_lossy().to_string()],
+            "--",
+            "/tmp",
+            &tx,
+            &event_log,
+            None,
+            1, // inactivity_timeout_secs = 1s
+            watcher_tx,
+            None,
+            None,
+        )
+        .await;
+
+        match result {
+            Err(err) => assert!(err.contains("inactivity timeout")),
+            Ok(_) => panic!("a genuinely silent agent with no in-flight tool call must be killed"),
+        }
+    }
+
     #[test]
     fn test_emit_event_writes_to_log_writer() {
         let tmp = tempfile::tempdir().unwrap();

server/src/agents/pty/runner.rs

@@ -246,14 +246,27 @@ fn run_agent_pty_blocking(
     // can distinguish a rate-limit exit from a genuine no-progress exit (bug 1053).
     let mut rate_limit_hard_block_seen = false;
     let mut rate_limit_reset_at_captured: Option<chrono::DateTime<chrono::Utc>> = None;
+    // Tool-use ids from `assistant` messages that haven't yet seen a matching
+    // `tool_result` in a `user` message.  While an MCP tool call (e.g.
+    // run_tests) is in flight, the server can legitimately run for many
+    // minutes with no PTY output at all — the CLI is blocked waiting on the
+    // MCP response, not hung.  The inactivity deadline is suspended entirely
+    // while this set is non-empty, and resumes as soon as the matching
+    // tool_result clears the last in-flight id (story 1196).
+    let mut tool_calls_in_flight: std::collections::HashSet<String> =
+        std::collections::HashSet::new();
 
     loop {
-        let effective_timeout = base_timeout.map(|base| {
-            let extra = block_until
-                .and_then(|t| (t - chrono::Utc::now()).to_std().ok())
-                .unwrap_or(std::time::Duration::ZERO);
-            base + extra
-        });
+        let effective_timeout = if !tool_calls_in_flight.is_empty() {
+            None
+        } else {
+            base_timeout.map(|base| {
+                let extra = block_until
+                    .and_then(|t| (t - chrono::Utc::now()).to_std().ok())
+                    .unwrap_or(std::time::Duration::ZERO);
+                base + extra
+            })
+        };
 
         let recv_result = match effective_timeout {
             Some(dur) => line_rx.recv_timeout(dur),
@@ -339,8 +352,31 @@ fn run_agent_pty_blocking(
             }
             // Complete assistant events are skipped for content extraction
             // because thinking and text already arrived via stream_event.
-            // The raw JSON is still forwarded as AgentJson below.
-            "assistant" | "user" => {}
+            // The raw JSON is still forwarded as AgentJson below.  A tool_use
+            // block marks an MCP call as in flight so the inactivity deadline
+            // is suspended until its tool_result arrives (story 1196).
+            "assistant" => {
+                if let Some(blocks) = json.pointer("/message/content").and_then(|c| c.as_array()) {
+                    for block in blocks {
+                        if block.get("type").and_then(|t| t.as_str()) == Some("tool_use")
+                            && let Some(id) = block.get("id").and_then(|i| i.as_str())
+                        {
+                            tool_calls_in_flight.insert(id.to_string());
+                        }
+                    }
+                }
+            }
+            "user" => {
+                if let Some(blocks) = json.pointer("/message/content").and_then(|c| c.as_array()) {
+                    for block in blocks {
+                        if block.get("type").and_then(|t| t.as_str()) == Some("tool_result")
+                            && let Some(id) = block.get("tool_use_id").and_then(|i| i.as_str())
+                        {
+                            tool_calls_in_flight.remove(id);
+                        }
+                    }
+                }
+            }
             "rate_limit_event" => {
                 let rate_limit_info = json.get("rate_limit_info");
                 let status = rate_limit_info