refactor: Refactor pause/resume functionality and improve state handling in CLI (#8152)

openhands/core/cli.py

@@ -101,7 +101,7 @@ async def run_session(
 
     sid = str(uuid4())
     is_loaded = asyncio.Event()
-    is_paused = asyncio.Event()
+    is_paused = asyncio.Event()  # Event to track agent pause requests
 
     # Show runtime initialization message
     display_runtime_initialization_message(config.runtime)
@@ -157,20 +157,15 @@ async def run_session(
         display_event(event, config)
         update_usage_metrics(event, usage_metrics)
 
-        # Pause the agent if the pause event is set (if Ctrl-P is pressed)
-        if is_paused.is_set():
-            event_stream.add_event(
-                ChangeAgentStateAction(AgentState.PAUSED),
-                EventSource.USER,
-            )
-            is_paused.clear()
-
         if isinstance(event, AgentStateChangedObservation):
             if event.agent_state in [
                 AgentState.AWAITING_USER_INPUT,
                 AgentState.FINISHED,
-                AgentState.PAUSED,
             ]:
+                # If the agent is paused, do not prompt for input as it's already handled by PAUSED state change
+                if is_paused.is_set():
+                    return
+
                 # Reload microagents after initialization of repo.md
                 if reload_microagents:
                     microagents: list[BaseMicroagent] = (
@@ -181,25 +176,32 @@ async def run_session(
                 await prompt_for_next_task(event.agent_state)
 
             if event.agent_state == AgentState.AWAITING_USER_CONFIRMATION:
-                # Only display the confirmation prompt if the agent is not paused
-                if not is_paused.is_set():
-                    user_confirmed = await read_confirmation_input()
-                    if user_confirmed:
-                        event_stream.add_event(
-                            ChangeAgentStateAction(AgentState.USER_CONFIRMED),
-                            EventSource.USER,
-                        )
-                    else:
-                        event_stream.add_event(
-                            ChangeAgentStateAction(AgentState.USER_REJECTED),
-                            EventSource.USER,
-                        )
+                # If the agent is paused, do not prompt for confirmation
+                # The confirmation step will re-run after the agent has been resumed
+                if is_paused.is_set():
+                    return
+
+                user_confirmed = await read_confirmation_input()
+                if user_confirmed:
+                    event_stream.add_event(
+                        ChangeAgentStateAction(AgentState.USER_CONFIRMED),
+                        EventSource.USER,
+                    )
+                else:
+                    event_stream.add_event(
+                        ChangeAgentStateAction(AgentState.USER_REJECTED),
+                        EventSource.USER,
+                    )
+
+            if event.agent_state == AgentState.PAUSED:
+                is_paused.clear()  # Revert the event state before prompting for user input
+                await prompt_for_next_task(event.agent_state)
 
             if event.agent_state == AgentState.RUNNING:
-                # Enable pause/resume functionality only if the confirmation mode is disabled
-                if not config.security.confirmation_mode:
-                    display_agent_running_message()
-                    loop.create_task(process_agent_pause(is_paused))
+                display_agent_running_message()
+                loop.create_task(
+                    process_agent_pause(is_paused, event_stream)
+                )  # Create a task to track agent pause requests from the user
 
     def on_event(event: Event) -> None:
         loop.create_task(on_event_async(event))

openhands/core/cli_tui.py

@@ -25,10 +25,11 @@ from prompt_toolkit.widgets import Frame, TextArea
 from openhands import __version__
 from openhands.core.config import AppConfig
 from openhands.core.schema import AgentState
-from openhands.events import EventSource
+from openhands.events import EventSource, EventStream
 from openhands.events.action import (
     Action,
     ActionConfirmationStatus,
+    ChangeAgentStateAction,
     CmdRunAction,
     FileEditAction,
     MessageAction,
@@ -60,7 +61,7 @@ COMMANDS = {
     '/status': 'Display session details and usage metrics',
     '/new': 'Create a new session',
     '/settings': 'Display and modify current settings',
-    '/resume': 'Resume the agent',
+    '/resume': 'Resume the agent when paused',
 }
 
 
@@ -396,7 +397,7 @@ def display_status(usage_metrics: UsageMetrics, session_id: str):
 def display_agent_running_message():
     print_formatted_text('')
     print_formatted_text(
-        HTML('<gold>Agent running...</gold> <grey>(Ctrl-P to pause)</grey>')
+        HTML('<gold>Agent running...</gold> <grey>(Press Ctrl-P to pause)</grey>')
     )
 
 
@@ -405,7 +406,7 @@ def display_agent_paused_message(agent_state: str):
         return
     print_formatted_text('')
     print_formatted_text(
-        HTML('<gold>Agent paused</gold> <grey>(type /resume to resume)</grey>')
+        HTML('<gold>Agent paused...</gold> <grey>(Enter /resume to continue)</grey>')
     )
 
 
@@ -430,7 +431,7 @@ class CommandCompleter(Completer):
                         command,
                         start_position=-len(text),
                         display_meta=description,
-                        style='bg:ansidarkgray fg:ansiwhite',
+                        style='bg:ansidarkgray fg:gold',
                     )
 
 
@@ -488,7 +489,7 @@ async def read_confirmation_input() -> bool:
         return False
 
 
-async def process_agent_pause(done: asyncio.Event) -> None:
+async def process_agent_pause(done: asyncio.Event, event_stream: EventStream) -> None:
     input = create_input()
 
     def keys_ready():
@@ -496,6 +497,10 @@ async def process_agent_pause(done: asyncio.Event) -> None:
             if key_press.key == Keys.ControlP:
                 print_formatted_text('')
                 print_formatted_text(HTML('<gold>Pausing the agent...</gold>'))
+                event_stream.add_event(
+                    ChangeAgentStateAction(AgentState.PAUSED),
+                    EventSource.USER,
+                )
                 done.set()
 
     with input.raw_mode():