glaip-sdk 0.0.20__py3-none-any.whl → 0.1.3__py3-none-any.whl

glaip_sdk/utils/rendering/renderer/base.py

@@ -8,25 +8,31 @@ from __future__ import annotations
 
 import json
 import logging
+from collections.abc import Iterable
 from dataclasses import dataclass, field
 from datetime import datetime, timezone
 from time import monotonic
 from typing import Any
 
-from rich.align import Align
 from rich.console import Console as RichConsole
 from rich.console import Group
 from rich.live import Live
 from rich.markdown import Markdown
+from rich.measure import Measurement
 from rich.spinner import Spinner
 from rich.text import Text
 
 from glaip_sdk.icons import ICON_AGENT, ICON_AGENT_STEP, ICON_DELEGATE, ICON_TOOL_STEP
 from glaip_sdk.rich_components import AIPPanel
 from glaip_sdk.utils.rendering.formatting import (
+    build_connector_prefix,
     format_main_title,
     get_spinner_char,
+    glyph_for_status,
     is_step_finished,
+    normalise_display_label,
+    pretty_args,
+    redact_sensitive,
 )
 from glaip_sdk.utils.rendering.models import RunStats, Step
 from glaip_sdk.utils.rendering.renderer.config import RendererConfig
@@ -44,13 +50,33 @@ from glaip_sdk.utils.rendering.renderer.progress import (
     is_delegation_tool,
 )
 from glaip_sdk.utils.rendering.renderer.stream import StreamProcessor
-from glaip_sdk.utils.rendering.steps import StepManager
+from glaip_sdk.utils.rendering.renderer.summary_window import clamp_step_nodes
+from glaip_sdk.utils.rendering.steps import UNKNOWN_STEP_DETAIL, StepManager
+
+DEFAULT_RENDERER_THEME = "dark"
+_NO_STEPS_TEXT = Text("No steps yet", style="dim")
 
 # Configure logger
 logger = logging.getLogger("glaip_sdk.run_renderer")
 
 # Constants
 LESS_THAN_1MS = "[<1ms]"
+FINISHED_STATUS_HINTS = {
+    "finished",
+    "success",
+    "succeeded",
+    "completed",
+    "failed",
+    "stopped",
+    "error",
+}
+RUNNING_STATUS_HINTS = {"running", "started", "pending", "working"}
+ARGS_VALUE_MAX_LEN = 160
+STATUS_ICON_STYLES = {
+    "success": "green",
+    "failed": "red",
+    "warning": "yellow",
+}
 
 
 def _coerce_received_at(value: Any) -> datetime | None:
@@ -72,6 +98,16 @@ def _coerce_received_at(value: Any) -> datetime | None:
     return None
 
 
+def _truncate_display(text: str | None, limit: int = 160) -> str:
+    """Return text capped at the given character limit with ellipsis."""
+    if not text:
+        return ""
+    stripped = str(text).strip()
+    if len(stripped) <= limit:
+        return stripped
+    return stripped[: limit - 1] + "…"
+
+
 @dataclass
 class RendererState:
     """Internal state for the renderer."""
@@ -96,6 +132,43 @@ class RendererState:
             self.buffer = []
 
 
+@dataclass
+class ThinkingScopeState:
+    """Runtime bookkeeping for deterministic thinking spans."""
+
+    anchor_id: str
+    task_id: str | None
+    context_id: str | None
+    anchor_started_at: float | None = None
+    anchor_finished_at: float | None = None
+    idle_started_at: float | None = None
+    idle_started_monotonic: float | None = None
+    active_thinking_id: str | None = None
+    running_children: set[str] = field(default_factory=set)
+    closed: bool = False
+
+
+class TrailingSpinnerLine:
+    """Render a text line with a trailing animated Rich spinner."""
+
+    def __init__(self, base_text: Text, spinner: Spinner) -> None:
+        """Initialize spinner line with base text and spinner component."""
+        self._base_text = base_text
+        self._spinner = spinner
+
+    def __rich_console__(self, console: RichConsole, options: Any) -> Any:
+        """Render the text with trailing animated spinner."""
+        spinner_render = self._spinner.render(console.get_time())
+        combined = Text.assemble(self._base_text.copy(), " ", spinner_render)
+        yield combined
+
+    def __rich_measure__(self, console: RichConsole, options: Any) -> Measurement:
+        """Measure the combined text and spinner dimensions."""
+        snapshot = self._spinner.render(0)
+        combined = Text.assemble(self._base_text.copy(), " ", snapshot)
+        return Measurement.get(console, options, combined)
+
+
 class RichStreamRenderer:
     """Live, modern terminal renderer for agent execution with rich visual output."""
 
@@ -122,17 +195,18 @@ class RichStreamRenderer:
         self.state = RendererState()
 
         # Initialize step manager and other state
-        self.steps = StepManager()
+        self.steps = StepManager(max_steps=self.cfg.summary_max_steps)
         # Live display instance (single source of truth)
         self.live: Live | None = None
+        self._step_spinners: dict[str, Spinner] = {}
 
-        # Context and tool tracking
-        self.context_order: list[str] = []
-        self.context_parent: dict[str, str] = {}
-        self.tool_order: list[str] = []
-        self.context_panels: dict[str, list[str]] = {}
-        self.context_meta: dict[str, dict[str, Any]] = {}
+        # Tool tracking and thinking scopes
         self.tool_panels: dict[str, dict[str, Any]] = {}
+        self._thinking_scopes: dict[str, ThinkingScopeState] = {}
+        self._root_agent_friendly: str | None = None
+        self._root_agent_step_id: str | None = None
+        self._root_query: str | None = None
+        self._root_query_attached: bool = False
 
         # Timing
         self._started_at: float | None = None
@@ -145,6 +219,17 @@ class RichStreamRenderer:
         # Output formatting constants
         self.OUTPUT_PREFIX: str = "**Output:**\n"
 
+        # Transcript toggling
+        self._transcript_mode_enabled: bool = False
+        self._transcript_render_cursor: int = 0
+        self.transcript_controller: Any | None = None
+        self._transcript_hint_message = "[dim]Transcript view · Press Ctrl+T to return to the summary.[/dim]"
+        self._summary_hint_message = "[dim]Press Ctrl+T to inspect raw transcript events.[/dim]"
+        self._summary_hint_printed_once: bool = False
+        self._transcript_hint_printed_once: bool = False
+        self._transcript_header_printed: bool = False
+        self._transcript_enabled_message_printed: bool = False
+
     def on_start(self, meta: dict[str, Any]) -> None:
         """Handle renderer start event."""
         if self.cfg.live:
@@ -158,6 +243,20 @@ class RichStreamRenderer:
         except Exception:
             self.state.meta = dict(meta)
 
+        meta_payload = meta or {}
+        self.steps.set_root_agent(meta_payload.get("agent_id"))
+        self._root_agent_friendly = self._humanize_agent_slug(meta_payload.get("agent_name"))
+        self._root_query = _truncate_display(
+            meta_payload.get("input_message")
+            or meta_payload.get("query")
+            or meta_payload.get("message")
+            or (meta_payload.get("meta") or {}).get("input_message")
+            or ""
+        )
+        if not self._root_query:
+            self._root_query = None
+        self._root_query_attached = False
+
         # Print compact header and user request (parity with old renderer)
         self._render_header(meta)
         self._render_user_query(meta)
@@ -207,20 +306,73 @@ class RichStreamRenderer:
         except Exception:
             logger.exception("Failed to print header fallback")
 
+    def _extract_query_from_meta(self, meta: dict[str, Any] | None) -> str | None:
+        """Extract the primary query string from a metadata payload."""
+        if not meta:
+            return None
+        query = (
+            meta.get("input_message")
+            or meta.get("query")
+            or meta.get("message")
+            or (meta.get("meta") or {}).get("input_message")
+        )
+        if isinstance(query, str) and query.strip():
+            return query
+        return None
+
+    def _build_user_query_panel(self, query: str) -> AIPPanel:
+        """Create the panel used to display the user request."""
+        return AIPPanel(
+            Markdown(f"**Query:** {query}"),
+            title="User Request",
+            border_style="#d97706",
+            padding=(0, 1),
+        )
+
     def _render_user_query(self, meta: dict[str, Any]) -> None:
         """Render the user query panel."""
-        query = meta.get("input_message") or meta.get("query") or meta.get("message")
+        query = self._extract_query_from_meta(meta)
         if not query:
             return
+        self.console.print(self._build_user_query_panel(query))
+
+    def _render_summary_static_sections(self) -> None:
+        """Re-render header and user query when returning to summary mode."""
+        meta = getattr(self.state, "meta", None)
+        if meta:
+            self._render_header(meta)
+        elif self.header_text and not self._render_header_rule():
+            self._render_header_fallback()
 
-        self.console.print(
-            AIPPanel(
-                Markdown(f"**Query:** {query}"),
-                title="User Request",
-                border_style="#d97706",
-                padding=(0, 1),
-            )
+        query = self._extract_query_from_meta(meta) or self._root_query
+        if query:
+            self.console.print(self._build_user_query_panel(query))
+
+    def _render_summary_after_transcript_toggle(self) -> None:
+        """Render the summary panel after leaving transcript mode."""
+        if self.state.finalizing_ui:
+            self._render_final_summary_panels()
+        elif self.live:
+            self._refresh_live_panels()
+        else:
+            self._render_static_summary_panels()
+
+    def _render_final_summary_panels(self) -> None:
+        """Render a static summary and disable live mode for final output."""
+        self.cfg.live = False
+        self.live = None
+        self._render_static_summary_panels()
+
+    def _render_static_summary_panels(self) -> None:
+        """Render the steps and main panels in a static (non-live) layout."""
+        steps_renderable = self._render_steps_text()
+        steps_panel = AIPPanel(
+            steps_renderable,
+            title="Steps",
+            border_style="blue",
         )
+        self.console.print(steps_panel)
+        self.console.print(self._render_main_panel())
 
     def _ensure_streaming_started_baseline(self, timestamp: float) -> None:
         """Synchronize streaming start state across renderer components."""
@@ -237,10 +389,12 @@ class RichStreamRenderer:
         self._sync_stream_start(ev, received_at)
 
         metadata = self.stream_processor.extract_event_metadata(ev)
-        self.stream_processor.update_timing(metadata["context_id"])
 
         self._maybe_render_debug(ev, received_at)
-        self._dispatch_event(ev, metadata)
+        try:
+            self._dispatch_event(ev, metadata)
+        finally:
+            self.stream_processor.update_timing(metadata.get("context_id"))
 
     def _resolve_received_timestamp(self, ev: dict[str, Any]) -> datetime:
         """Return the timestamp an event was received, normalising inputs."""
@@ -253,9 +407,7 @@ class RichStreamRenderer:
 
         return received_at
 
-    def _sync_stream_start(
-        self, ev: dict[str, Any], received_at: datetime | None
-    ) -> None:
+    def _sync_stream_start(self, ev: dict[str, Any], received_at: datetime | None) -> None:
         """Ensure renderer and stream processor share a streaming baseline."""
         baseline = self.state.streaming_started_at
         if baseline is None:
@@ -275,12 +427,14 @@ class RichStreamRenderer:
         if not self.verbose:
             return
 
+        self._ensure_transcript_header()
         render_debug_event(
             ev,
             self.console,
             received_ts=received_at,
             baseline_ts=self.state.streaming_started_event_ts,
         )
+        self._print_transcript_hint()
 
     def _dispatch_event(self, ev: dict[str, Any], metadata: dict[str, Any]) -> None:
         """Route events to the appropriate renderer handlers."""
@@ -294,7 +448,7 @@ class RichStreamRenderer:
         elif kind == "final_response":
             self._handle_final_response_event(content, metadata)
         elif kind in {"agent_step", "agent_thinking_step"}:
-            self._handle_agent_step_event(ev)
+            self._handle_agent_step_event(ev, metadata)
         else:
             self._ensure_live()
 
@@ -310,21 +464,32 @@ class RichStreamRenderer:
             self.state.buffer.append(content)
             self._ensure_live()
 
-    def _handle_final_response_event(
-        self, content: str, metadata: dict[str, Any]
-    ) -> None:
+    def _handle_final_response_event(self, content: str, metadata: dict[str, Any]) -> None:
         """Handle final response events."""
         if content:
             self.state.buffer.append(content)
             self.state.final_text = content
 
             meta_payload = metadata.get("metadata") or {}
-            self._update_final_duration(meta_payload.get("time"))
+            final_time = self._coerce_server_time(meta_payload.get("time"))
+            self._update_final_duration(final_time)
+            self._close_active_thinking_scopes(final_time)
+            self._finish_running_steps()
+            self._finish_tool_panels()
+            self._normalise_finished_icons()
 
-            self._ensure_live()
-            self._print_final_panel_if_needed()
+        self._ensure_live()
+        self._print_final_panel_if_needed()
+
+    def _normalise_finished_icons(self) -> None:
+        """Ensure finished steps do not keep spinner icons."""
+        for step in self.steps.by_id.values():
+            if getattr(step, "status", None) == "finished" and getattr(step, "status_icon", None) == "spinner":
+                step.status_icon = "success"
+            if getattr(step, "status", None) != "running":
+                self._step_spinners.pop(step.step_id, None)
 
-    def _handle_agent_step_event(self, ev: dict[str, Any]) -> None:
+    def _handle_agent_step_event(self, ev: dict[str, Any], metadata: dict[str, Any]) -> None:
         """Handle agent step events."""
         # Extract tool information
         (
@@ -334,22 +499,376 @@ class RichStreamRenderer:
             tool_calls_info,
         ) = self.stream_processor.parse_tool_calls(ev)
 
-        # Track tools and sub-agents
-        self.stream_processor.track_tools_and_agents(
-            tool_name, tool_calls_info, is_delegation_tool
-        )
+        payload = metadata.get("metadata") or {}
+
+        tracked_step: Step | None = None
+        try:
+            tracked_step = self.steps.apply_event(ev)
+        except ValueError:
+            logger.debug("Malformed step event skipped", exc_info=True)
+        else:
+            self._record_step_server_start(tracked_step, payload)
+            self._update_thinking_timeline(tracked_step, payload)
+            self._maybe_override_root_agent_label(tracked_step, payload)
+            self._maybe_attach_root_query(tracked_step)
+
+        # Track tools and sub-agents for transcript/debug context
+        self.stream_processor.track_tools_and_agents(tool_name, tool_calls_info, is_delegation_tool)
 
         # Handle tool execution
-        self._handle_agent_step(ev, tool_name, tool_args, tool_out, tool_calls_info)
+        self._handle_agent_step(
+            ev,
+            tool_name,
+            tool_args,
+            tool_out,
+            tool_calls_info,
+            tracked_step=tracked_step,
+        )
 
         # Update live display
         self._ensure_live()
 
+    def _maybe_attach_root_query(self, step: Step | None) -> None:
+        """Attach the user query to the root agent step for display."""
+        if not step or self._root_query_attached or not self._root_query or step.kind != "agent" or step.parent_id:
+            return
+
+        args = dict(getattr(step, "args", {}) or {})
+        args.setdefault("query", self._root_query)
+        step.args = args
+        self._root_query_attached = True
+
+    def _record_step_server_start(self, step: Step | None, payload: dict[str, Any]) -> None:
+        """Store server-provided start times for elapsed calculations."""
+        if not step:
+            return
+        server_time = payload.get("time")
+        if not isinstance(server_time, (int, float)):
+            return
+        self._step_server_start_times.setdefault(step.step_id, float(server_time))
+
+    def _maybe_override_root_agent_label(self, step: Step | None, payload: dict[str, Any]) -> None:
+        """Ensure the root agent row uses the human-friendly name and shows the ID."""
+        if not step or step.kind != "agent" or step.parent_id:
+            return
+        friendly = self._root_agent_friendly or self._humanize_agent_slug((payload or {}).get("agent_name"))
+        if not friendly:
+            return
+        agent_identifier = step.name or step.step_id
+        if not agent_identifier:
+            return
+        step.display_label = normalise_display_label(f"{ICON_AGENT} {friendly} ({agent_identifier})")
+        if not self._root_agent_step_id:
+            self._root_agent_step_id = step.step_id
+
+    def _update_thinking_timeline(self, step: Step | None, payload: dict[str, Any]) -> None:
+        """Maintain deterministic thinking spans for each agent/delegate scope."""
+        if not self.cfg.render_thinking or not step:
+            return
+
+        now_monotonic = monotonic()
+        server_time = self._coerce_server_time(payload.get("time"))
+        status_hint = (payload.get("status") or "").lower()
+
+        if self._is_scope_anchor(step):
+            self._update_anchor_thinking(
+                step=step,
+                server_time=server_time,
+                status_hint=status_hint,
+                now_monotonic=now_monotonic,
+            )
+            return
+
+        self._update_child_thinking(
+            step=step,
+            server_time=server_time,
+            status_hint=status_hint,
+            now_monotonic=now_monotonic,
+        )
+
+    def _update_anchor_thinking(
+        self,
+        *,
+        step: Step,
+        server_time: float | None,
+        status_hint: str,
+        now_monotonic: float,
+    ) -> None:
+        """Handle deterministic thinking bookkeeping for agent/delegate anchors."""
+        scope = self._get_or_create_scope(step)
+        if scope.anchor_started_at is None and server_time is not None:
+            scope.anchor_started_at = server_time
+
+        if not scope.closed and scope.active_thinking_id is None:
+            self._start_scope_thinking(
+                scope,
+                start_server_time=scope.anchor_started_at or server_time,
+                start_monotonic=now_monotonic,
+            )
+
+        is_anchor_finished = status_hint in FINISHED_STATUS_HINTS or (not status_hint and is_step_finished(step))
+        if is_anchor_finished:
+            scope.anchor_finished_at = server_time or scope.anchor_finished_at
+            self._finish_scope_thinking(scope, server_time, now_monotonic)
+            scope.closed = True
+
+        parent_anchor_id = self._resolve_anchor_id(step)
+        if parent_anchor_id:
+            self._cascade_anchor_update(
+                parent_anchor_id=parent_anchor_id,
+                child_step=step,
+                server_time=server_time,
+                now_monotonic=now_monotonic,
+                is_finished=is_anchor_finished,
+            )
+
+    def _cascade_anchor_update(
+        self,
+        *,
+        parent_anchor_id: str,
+        child_step: Step,
+        server_time: float | None,
+        now_monotonic: float,
+        is_finished: bool,
+    ) -> None:
+        """Propagate anchor state changes to the parent scope."""
+        parent_scope = self._thinking_scopes.get(parent_anchor_id)
+        if not parent_scope or parent_scope.closed:
+            return
+        if is_finished:
+            self._mark_child_finished(parent_scope, child_step.step_id, server_time, now_monotonic)
+        else:
+            self._mark_child_running(parent_scope, child_step, server_time, now_monotonic)
+
+    def _update_child_thinking(
+        self,
+        *,
+        step: Step,
+        server_time: float | None,
+        status_hint: str,
+        now_monotonic: float,
+    ) -> None:
+        """Update deterministic thinking state for non-anchor steps."""
+        anchor_id = self._resolve_anchor_id(step)
+        if not anchor_id:
+            return
+
+        scope = self._thinking_scopes.get(anchor_id)
+        if not scope or scope.closed or step.kind == "thinking":
+            return
+
+        is_finish_event = status_hint in FINISHED_STATUS_HINTS or (not status_hint and is_step_finished(step))
+        if is_finish_event:
+            self._mark_child_finished(scope, step.step_id, server_time, now_monotonic)
+        else:
+            self._mark_child_running(scope, step, server_time, now_monotonic)
+
+    def _resolve_anchor_id(self, step: Step) -> str | None:
+        """Return the nearest agent/delegate ancestor for a step."""
+        parent_id = step.parent_id
+        while parent_id:
+            parent = self.steps.by_id.get(parent_id)
+            if not parent:
+                return None
+            if self._is_scope_anchor(parent):
+                return parent.step_id
+            parent_id = parent.parent_id
+        return None
+
+    def _get_or_create_scope(self, step: Step) -> ThinkingScopeState:
+        """Fetch (or create) thinking state for the given anchor step."""
+        scope = self._thinking_scopes.get(step.step_id)
+        if scope:
+            if scope.task_id is None:
+                scope.task_id = step.task_id
+            if scope.context_id is None:
+                scope.context_id = step.context_id
+            return scope
+        scope = ThinkingScopeState(
+            anchor_id=step.step_id,
+            task_id=step.task_id,
+            context_id=step.context_id,
+        )
+        self._thinking_scopes[step.step_id] = scope
+        return scope
+
+    def _is_scope_anchor(self, step: Step) -> bool:
+        """Return True when a step should host its own thinking timeline."""
+        if step.kind in {"agent", "delegate"}:
+            return True
+        name = (step.name or "").lower()
+        return name.startswith(("delegate_to_", "delegate_", "delegate "))
+
+    def _start_scope_thinking(
+        self,
+        scope: ThinkingScopeState,
+        *,
+        start_server_time: float | None,
+        start_monotonic: float,
+    ) -> None:
+        """Open a deterministic thinking node beneath the scope anchor."""
+        if scope.closed or scope.active_thinking_id or not scope.anchor_id:
+            return
+        step = self.steps.start_or_get(
+            task_id=scope.task_id,
+            context_id=scope.context_id,
+            kind="thinking",
+            name=f"agent_thinking_step::{scope.anchor_id}",
+            parent_id=scope.anchor_id,
+            args={"reason": "deterministic_timeline"},
+        )
+        step.display_label = "💭 Thinking…"
+        step.status_icon = "spinner"
+        scope.active_thinking_id = step.step_id
+        scope.idle_started_at = start_server_time
+        scope.idle_started_monotonic = start_monotonic
+
+    def _finish_scope_thinking(
+        self,
+        scope: ThinkingScopeState,
+        end_server_time: float | None,
+        end_monotonic: float,
+    ) -> None:
+        """Close the currently running thinking node if one exists."""
+        if not scope.active_thinking_id:
+            return
+        thinking_step = self.steps.by_id.get(scope.active_thinking_id)
+        if not thinking_step:
+            scope.active_thinking_id = None
+            scope.idle_started_at = None
+            scope.idle_started_monotonic = None
+            return
+
+        duration = self._calculate_timeline_duration(
+            scope.idle_started_at,
+            end_server_time,
+            scope.idle_started_monotonic,
+            end_monotonic,
+        )
+        thinking_step.display_label = thinking_step.display_label or "💭 Thinking…"
+        if duration is not None:
+            thinking_step.finish(duration, source="timeline")
+        else:
+            thinking_step.finish(None, source="timeline")
+        thinking_step.status_icon = "success"
+        scope.active_thinking_id = None
+        scope.idle_started_at = None
+        scope.idle_started_monotonic = None
+
+    def _mark_child_running(
+        self,
+        scope: ThinkingScopeState,
+        step: Step,
+        server_time: float | None,
+        now_monotonic: float,
+    ) -> None:
+        """Mark a direct child as running and close any open thinking node."""
+        if step.step_id in scope.running_children:
+            return
+        scope.running_children.add(step.step_id)
+        if not scope.active_thinking_id:
+            return
+
+        start_server = self._step_server_start_times.get(step.step_id)
+        if start_server is None:
+            start_server = server_time
+        self._finish_scope_thinking(scope, start_server, now_monotonic)
+
+    def _mark_child_finished(
+        self,
+        scope: ThinkingScopeState,
+        step_id: str,
+        server_time: float | None,
+        now_monotonic: float,
+    ) -> None:
+        """Handle completion for a scope child and resume thinking if idle."""
+        if step_id in scope.running_children:
+            scope.running_children.discard(step_id)
+        if scope.running_children or scope.closed:
+            return
+        self._start_scope_thinking(
+            scope,
+            start_server_time=server_time,
+            start_monotonic=now_monotonic,
+        )
+
+    def _close_active_thinking_scopes(self, server_time: float | None) -> None:
+        """Finish any in-flight thinking nodes during finalization."""
+        now = monotonic()
+        for scope in self._thinking_scopes.values():
+            if not scope.active_thinking_id:
+                continue
+            self._finish_scope_thinking(scope, server_time, now)
+            scope.closed = True
+            # Parent scopes resume thinking via _cascade_anchor_update
+
+    def _apply_root_duration(self, duration_seconds: float | None) -> None:
+        """Propagate the final run duration to the root agent step."""
+        if duration_seconds is None or not self._root_agent_step_id:
+            return
+        root_step = self.steps.by_id.get(self._root_agent_step_id)
+        if not root_step:
+            return
+        try:
+            duration_ms = max(0, int(round(float(duration_seconds) * 1000)))
+        except Exception:
+            return
+        root_step.duration_ms = duration_ms
+        root_step.duration_source = root_step.duration_source or "run"
+        root_step.status = "finished"
+
+    @staticmethod
+    def _coerce_server_time(value: Any) -> float | None:
+        """Convert a raw SSE time payload into a float if possible."""
+        if isinstance(value, (int, float)):
+            return float(value)
+        try:
+            return float(value)
+        except (TypeError, ValueError):
+            return None
+
+    @staticmethod
+    def _calculate_timeline_duration(
+        start_server: float | None,
+        end_server: float | None,
+        start_monotonic: float | None,
+        end_monotonic: float,
+    ) -> float | None:
+        """Pick the most reliable pair of timestamps to derive duration seconds."""
+        if start_server is not None and end_server is not None:
+            return max(0.0, float(end_server) - float(start_server))
+        if start_monotonic is not None:
+            try:
+                return max(0.0, float(end_monotonic) - float(start_monotonic))
+            except Exception:
+                return None
+        return None
+
+    @staticmethod
+    def _humanize_agent_slug(value: Any) -> str | None:
+        """Convert a slugified agent name into Title Case."""
+        if not isinstance(value, str):
+            return None
+        cleaned = value.replace("_", " ").replace("-", " ").strip()
+        if not cleaned:
+            return None
+        parts = [part for part in cleaned.split() if part]
+        return " ".join(part[:1].upper() + part[1:] for part in parts)
+
     def _finish_running_steps(self) -> None:
         """Mark any running steps as finished to avoid lingering spinners."""
         for st in self.steps.by_id.values():
             if not is_step_finished(st):
-                st.finish(None)
+                self._mark_incomplete_step(st)
+
+    def _mark_incomplete_step(self, step: Step) -> None:
+        """Mark a lingering step as incomplete/warning with unknown duration."""
+        step.status = "finished"
+        step.duration_unknown = True
+        if step.duration_ms is None:
+            step.duration_ms = 0
+        step.duration_source = step.duration_source or "unknown"
+        step.status_icon = "warning"
 
     def _finish_tool_panels(self) -> None:
         """Mark unfinished tool panels as finished."""
@@ -376,11 +895,14 @@ class RichStreamRenderer:
         if not body:
             return
 
+        if getattr(self, "_transcript_mode_enabled", False):
+            return
+
         if self.verbose:
             final_panel = create_final_panel(
                 body,
                 title=self._final_panel_title(),
-                theme=self.cfg.theme,
+                theme=DEFAULT_RENDERER_THEME,
             )
             self.console.print(final_panel)
             self.state.printed_final_output = True
@@ -389,25 +911,37 @@ class RichStreamRenderer:
         """Handle completion event."""
         self.state.finalizing_ui = True
 
-        if isinstance(stats, RunStats):
+        self._handle_stats_duration(stats)
+        self._close_active_thinking_scopes(self.state.final_duration_seconds)
+        self._cleanup_ui_elements()
+        self._finalize_display()
+        self._print_completion_message()
+
+    def _handle_stats_duration(self, stats: RunStats) -> None:
+        """Handle stats processing and duration calculation."""
+        if not isinstance(stats, RunStats):
+            return
+
+        duration = None
+        try:
+            if stats.finished_at is not None and stats.started_at is not None:
+                duration = max(0.0, float(stats.finished_at) - float(stats.started_at))
+        except Exception:
             duration = None
-            try:
-                if stats.finished_at is not None and stats.started_at is not None:
-                    duration = max(
-                        0.0, float(stats.finished_at) - float(stats.started_at)
-                    )
-            except Exception:
-                duration = None
 
-            if duration is not None:
-                self._update_final_duration(duration, overwrite=True)
+        if duration is not None:
+            self._update_final_duration(duration, overwrite=True)
 
+    def _cleanup_ui_elements(self) -> None:
+        """Clean up running UI elements."""
         # Mark any running steps as finished to avoid lingering spinners
         self._finish_running_steps()
 
         # Mark unfinished tool panels as finished
         self._finish_tool_panels()
 
+    def _finalize_display(self) -> None:
+        """Finalize live display and render final output."""
         # Final refresh
         self._ensure_live()
 
@@ -417,8 +951,24 @@ class RichStreamRenderer:
         # Render final output based on configuration
         self._print_final_panel_if_needed()
 
+    def _print_completion_message(self) -> None:
+        """Print completion message based on current mode."""
+        if self._transcript_mode_enabled:
+            try:
+                self.console.print(
+                    "[dim]Run finished. Press Ctrl+T to return to the summary view or stay here to inspect events. "
+                    "Use the post-run viewer for export.[/dim]"
+                )
+            except Exception:
+                pass
+        else:
+            # No transcript toggle in summary mode; nothing to print here.
+            return
+
     def _ensure_live(self) -> None:
         """Ensure live display is updated."""
+        if getattr(self, "_transcript_mode_enabled", False):
+            return
         if not self._ensure_live_stack():
             return
 
@@ -426,6 +976,12 @@ class RichStreamRenderer:
 
         if self.live:
             self._refresh_live_panels()
+            if (
+                not self._transcript_mode_enabled
+                and not self.state.finalizing_ui
+                and not self._summary_hint_printed_once
+            ):
+                self._print_summary_hint(force=True)
 
     def _ensure_live_stack(self) -> bool:
         """Guarantee the console exposes the internal live stack Rich expects."""
@@ -472,8 +1028,7 @@ class RichStreamRenderer:
             title="Steps",
             border_style="blue",
         )
-        tool_panels = self._render_tool_panels()
-        panels = self._build_live_panels(main_panel, steps_panel, tool_panels)
+        panels = self._build_live_panels(main_panel, steps_panel)
 
         self.live.update(Group(*panels))
 
@@ -481,17 +1036,12 @@ class RichStreamRenderer:
         self,
         main_panel: Any,
         steps_panel: Any,
-        tool_panels: list[Any],
     ) -> list[Any]:
         """Assemble the panel order for the live display."""
         if self.verbose:
-            return [main_panel, steps_panel, *tool_panels]
+            return [main_panel, steps_panel]
 
-        panels: list[Any] = [steps_panel]
-        if tool_panels:
-            panels.extend(tool_panels)
-        panels.append(main_panel)
-        return panels
+        return [steps_panel, main_panel]
 
     def _render_main_panel(self) -> Any:
         """Render the main content panel."""
@@ -503,11 +1053,11 @@ class RichStreamRenderer:
                 return create_final_panel(
                     final_content,
                     title=title,
-                    theme=self.cfg.theme,
+                    theme=DEFAULT_RENDERER_THEME,
                 )
         # Dynamic title with spinner + elapsed/hints
         title = self._format_enhanced_main_title()
-        return create_main_panel(body, title, self.cfg.theme)
+        return create_main_panel(body, title, DEFAULT_RENDERER_THEME)
 
     def _final_panel_title(self) -> str:
         """Compose title for the final result panel including duration."""
@@ -522,8 +1072,6 @@ class RichStreamRenderer:
             return
 
         self.verbose = verbose
-        self.cfg.style = "debug" if verbose else "pretty"
-
         desired_live = not verbose
         if desired_live != self.cfg.live:
             self.cfg.live = desired_live
@@ -538,9 +1086,123 @@ class RichStreamRenderer:
     # ------------------------------------------------------------------
     # Transcript helpers
     # ------------------------------------------------------------------
-    def _capture_event(
-        self, ev: dict[str, Any], received_at: datetime | None = None
-    ) -> None:
+    @property
+    def transcript_mode_enabled(self) -> bool:
+        """Return True when transcript mode is currently active."""
+        return self._transcript_mode_enabled
+
+    def toggle_transcript_mode(self) -> None:
+        """Flip transcript mode on/off."""
+        self.set_transcript_mode(not self._transcript_mode_enabled)
+
+    def set_transcript_mode(self, enabled: bool) -> None:
+        """Set transcript mode explicitly."""
+        if enabled == self._transcript_mode_enabled:
+            return
+
+        self._transcript_mode_enabled = enabled
+        self.apply_verbosity(enabled)
+
+        if enabled:
+            self._summary_hint_printed_once = False
+            self._transcript_hint_printed_once = False
+            self._transcript_header_printed = False
+            self._transcript_enabled_message_printed = False
+            self._stop_live_display()
+            self._clear_console_safe()
+            self._print_transcript_enabled_message()
+            self._render_transcript_backfill()
+        else:
+            self._transcript_hint_printed_once = False
+            self._transcript_header_printed = False
+            self._transcript_enabled_message_printed = False
+            self._clear_console_safe()
+        self._render_summary_static_sections()
+        summary_notice = (
+            "[dim]Returning to the summary view. Streaming will continue here.[/dim]"
+            if not self.state.finalizing_ui
+            else "[dim]Returning to the summary view.[/dim]"
+        )
+        self.console.print(summary_notice)
+        self._render_summary_after_transcript_toggle()
+        if not self.state.finalizing_ui:
+            self._print_summary_hint(force=True)
+
+    def _clear_console_safe(self) -> None:
+        """Best-effort console clear that ignores platform quirks."""
+        try:
+            self.console.clear()
+        except Exception:
+            pass
+
+    def _print_transcript_hint(self) -> None:
+        """Render the transcript toggle hint, keeping it near the bottom."""
+        if not self._transcript_mode_enabled:
+            return
+        try:
+            self.console.print(self._transcript_hint_message)
+        except Exception:
+            pass
+        else:
+            self._transcript_hint_printed_once = True
+
+    def _print_transcript_enabled_message(self) -> None:
+        if self._transcript_enabled_message_printed:
+            return
+        try:
+            self.console.print("[dim]Transcript mode enabled — streaming raw transcript events.[/dim]")
+        except Exception:
+            pass
+        else:
+            self._transcript_enabled_message_printed = True
+
+    def _ensure_transcript_header(self) -> None:
+        if self._transcript_header_printed:
+            return
+        try:
+            self.console.rule("Transcript Events")
+        except Exception:
+            self._transcript_header_printed = True
+            return
+        self._transcript_header_printed = True
+
+    def _print_summary_hint(self, force: bool = False) -> None:
+        """Show the summary-mode toggle hint."""
+        controller = getattr(self, "transcript_controller", None)
+        if controller and not getattr(controller, "enabled", False):
+            if not force:
+                self._summary_hint_printed_once = True
+            return
+        if not force and self._summary_hint_printed_once:
+            return
+        try:
+            self.console.print(self._summary_hint_message)
+        except Exception:
+            return
+        self._summary_hint_printed_once = True
+
+    def _render_transcript_backfill(self) -> None:
+        """Render any captured events that haven't been shown in transcript mode."""
+        pending = self.state.events[self._transcript_render_cursor :]
+        self._ensure_transcript_header()
+        if not pending:
+            self._print_transcript_hint()
+            return
+
+        baseline = self.state.streaming_started_event_ts
+        for ev in pending:
+            received_ts = _coerce_received_at(ev.get("received_at"))
+            render_debug_event(
+                ev,
+                self.console,
+                received_ts=received_ts,
+                baseline_ts=baseline,
+            )
+
+        self._transcript_render_cursor = len(self.state.events)
+        self._print_transcript_hint()
+
+    def _capture_event(self, ev: dict[str, Any], received_at: datetime | None = None) -> None:
         """Capture a deep copy of SSE events for transcript replay."""
         try:
             captured = json.loads(json.dumps(ev))
@@ -557,6 +1219,8 @@ class RichStreamRenderer:
                     captured["received_at"] = repr(received_at)
 
         self.state.events.append(captured)
+        if self._transcript_mode_enabled:
+            self._transcript_render_cursor = len(self.state.events)
 
     def get_aggregated_output(self) -> str:
         """Return the concatenated assistant output collected so far."""
@@ -566,16 +1230,7 @@ class RichStreamRenderer:
         """Return captured SSE events."""
         return list(self.state.events)
 
-    def _maybe_insert_thinking_gap(
-        self, task_id: str | None, context_id: str | None
-    ) -> None:
-        """Insert thinking gap if needed."""
-        # Implementation would track thinking states
-        pass
-
-    def _ensure_tool_panel(
-        self, name: str, args: Any, task_id: str, context_id: str
-    ) -> str:
+    def _ensure_tool_panel(self, name: str, args: Any, task_id: str, context_id: str) -> str:
         """Ensure a tool panel exists and return its ID."""
         formatted_title = format_tool_title(name)
         is_delegation = is_delegation_tool(name)
@@ -595,15 +1250,10 @@ class RichStreamRenderer:
             # Add Args section once
             if args:
                 try:
-                    args_content = (
-                        "**Args:**\n```json\n"
-                        + json.dumps(args, indent=2)
-                        + "\n```\n\n"
-                    )
+                    args_content = "**Args:**\n```json\n" + json.dumps(args, indent=2) + "\n```\n\n"
                 except Exception:
                     args_content = f"**Args:**\n{args}\n\n"
                 self.tool_panels[tool_sid]["chunks"].append(args_content)
-            self.tool_order.append(tool_sid)
 
         return tool_sid
 
@@ -614,8 +1264,13 @@ class RichStreamRenderer:
         tool_name: str,
         tool_args: Any,
         _tool_sid: str,
+        *,
+        tracked_step: Step | None = None,
     ) -> Step | None:
         """Start or get a step for a tool."""
+        if tracked_step is not None:
+            return tracked_step
+
         if is_delegation_tool(tool_name):
             st = self.steps.start_or_get(
                 task_id=task_id,
@@ -635,9 +1290,7 @@ class RichStreamRenderer:
 
         # Record server start time for this step if available
         if st and self.stream_processor.server_elapsed_time is not None:
-            self._step_server_start_times[st.step_id] = (
-                self.stream_processor.server_elapsed_time
-            )
+            self._step_server_start_times[st.step_id] = self.stream_processor.server_elapsed_time
 
         return st
 
@@ -651,26 +1304,18 @@ class RichStreamRenderer:
         """Process additional tool calls to avoid duplicates."""
         for call_name, call_args, _ in tool_calls_info or []:
             if call_name and call_name != tool_name:
-                self._process_single_tool_call(
-                    call_name, call_args, task_id, context_id
-                )
+                self._process_single_tool_call(call_name, call_args, task_id, context_id)
 
-    def _process_single_tool_call(
-        self, call_name: str, call_args: Any, task_id: str, context_id: str
-    ) -> None:
+    def _process_single_tool_call(self, call_name: str, call_args: Any, task_id: str, context_id: str) -> None:
         """Process a single additional tool call."""
         self._ensure_tool_panel(call_name, call_args, task_id, context_id)
 
         st2 = self._create_step_for_tool_call(call_name, call_args, task_id, context_id)
 
         if self.stream_processor.server_elapsed_time is not None and st2:
-            self._step_server_start_times[st2.step_id] = (
-                self.stream_processor.server_elapsed_time
-            )
+            self._step_server_start_times[st2.step_id] = self.stream_processor.server_elapsed_time
 
-    def _create_step_for_tool_call(
-        self, call_name: str, call_args: Any, task_id: str, context_id: str
-    ) -> Any:
+    def _create_step_for_tool_call(self, call_name: str, call_args: Any, task_id: str, context_id: str) -> Any:
         """Create appropriate step for tool call."""
         if is_delegation_tool(call_name):
             return self.steps.start_or_get(
@@ -689,9 +1334,7 @@ class RichStreamRenderer:
                 args=call_args,
             )
 
-    def _detect_tool_completion(
-        self, metadata: dict, content: str
-    ) -> tuple[bool, str | None, Any]:
+    def _detect_tool_completion(self, metadata: dict, content: str) -> tuple[bool, str | None, Any]:
         """Detect if a tool has completed and return completion info."""
         tool_info = metadata.get("tool_info", {}) if isinstance(metadata, dict) else {}
 
@@ -701,18 +1344,14 @@ class RichStreamRenderer:
             # content like "Completed google_serper"
             tname = content.replace("Completed ", "").strip()
             if tname:
-                output = (
-                    tool_info.get("output") if tool_info.get("name") == tname else None
-                )
+                output = tool_info.get("output") if tool_info.get("name") == tname else None
                 return True, tname, output
         elif metadata.get("status") == "finished" and tool_info.get("name"):
             return True, tool_info.get("name"), tool_info.get("output")
 
         return False, None, None
 
-    def _get_tool_session_id(
-        self, finished_tool_name: str, task_id: str, context_id: str
-    ) -> str:
+    def _get_tool_session_id(self, finished_tool_name: str, task_id: str, context_id: str) -> str:
         """Generate tool session ID."""
         return f"tool_{finished_tool_name}_{task_id}_{context_id}"
 
@@ -742,7 +1381,7 @@ class RichStreamRenderer:
             meta["duration_seconds"] = dur
             meta["server_finished_at"] = (
                 self.stream_processor.server_elapsed_time
-                if isinstance(self.stream_processor.server_elapsed_time, int | float)
+                if isinstance(self.stream_processor.server_elapsed_time, (int, float))
                 else None
             )
             meta["finished_at"] = monotonic()
@@ -752,9 +1391,7 @@ class RichStreamRenderer:
     ) -> None:
         """Add tool output to panel metadata."""
         if finished_tool_output is not None:
-            meta["chunks"].append(
-                self._format_output_block(finished_tool_output, finished_tool_name)
-            )
+            meta["chunks"].append(self._format_output_block(finished_tool_output, finished_tool_name))
             meta["output"] = finished_tool_output
 
     def _mark_panel_as_finished(self, meta: dict[str, Any], tool_sid: str) -> None:
@@ -784,9 +1421,7 @@ class RichStreamRenderer:
         self._mark_panel_as_finished(meta, tool_sid)
         self._add_tool_output_to_panel(meta, finished_tool_output, finished_tool_name)
 
-    def _get_step_duration(
-        self, finished_tool_name: str, task_id: str, context_id: str
-    ) -> float | None:
+    def _get_step_duration(self, finished_tool_name: str, task_id: str, context_id: str) -> float | None:
         """Get step duration from tool panels."""
         tool_sid = f"tool_{finished_tool_name}_{task_id}_{context_id}"
         return self.tool_panels.get(tool_sid, {}).get("duration_seconds")
@@ -833,8 +1468,13 @@ class RichStreamRenderer:
         finished_tool_output: Any,
         task_id: str,
         context_id: str,
+        *,
+        tracked_step: Step | None = None,
     ) -> None:
         """Finish the corresponding step for a completed tool."""
+        if tracked_step is not None:
+            return
+
         step_duration = self._get_step_duration(finished_tool_name, task_id, context_id)
 
         if is_delegation_tool(finished_tool_name):
@@ -856,9 +1496,7 @@ class RichStreamRenderer:
 
     def _should_create_snapshot(self, tool_sid: str) -> bool:
         """Check if a snapshot should be created."""
-        return self.cfg.append_finished_snapshots and not self.tool_panels.get(
-            tool_sid, {}
-        ).get("snapshot_printed")
+        return self.cfg.append_finished_snapshots and not self.tool_panels.get(tool_sid, {}).get("snapshot_printed")
 
     def _get_snapshot_title(self, meta: dict[str, Any], finished_tool_name: str) -> str:
         """Get the title for the snapshot."""
@@ -866,7 +1504,7 @@ class RichStreamRenderer:
 
         # Add elapsed time to title
         dur = meta.get("duration_seconds")
-        if isinstance(dur, int | float):
+        if isinstance(dur, (int, float)):
             elapsed_str = self._format_snapshot_duration(dur)
             adjusted_title = f"{adjusted_title}  · {elapsed_str}"
 
@@ -903,15 +1541,13 @@ class RichStreamRenderer:
 
         return body_text
 
-    def _create_snapshot_panel(
-        self, adjusted_title: str, body_text: str, finished_tool_name: str
-    ) -> Any:
+    def _create_snapshot_panel(self, adjusted_title: str, body_text: str, finished_tool_name: str) -> Any:
         """Create the snapshot panel."""
         return create_tool_panel(
             title=adjusted_title,
             content=body_text or "(no output)",
             status="finished",
-            theme=self.cfg.theme,
+            theme=DEFAULT_RENDERER_THEME,
             is_delegation=is_delegation_tool(finished_tool_name),
         )
 
@@ -920,9 +1556,7 @@ class RichStreamRenderer:
         self.console.print(snapshot_panel)
         self.tool_panels[tool_sid]["snapshot_printed"] = True
 
-    def _create_tool_snapshot(
-        self, finished_tool_name: str, task_id: str, context_id: str
-    ) -> None:
+    def _create_tool_snapshot(self, finished_tool_name: str, task_id: str, context_id: str) -> None:
         """Create and print a snapshot for a finished tool."""
         tool_sid = f"tool_{finished_tool_name}_{task_id}_{context_id}"
 
@@ -936,9 +1570,7 @@ class RichStreamRenderer:
         body_text = "".join(meta.get("chunks") or [])
         body_text = self._clamp_snapshot_body(body_text)
 
-        snapshot_panel = self._create_snapshot_panel(
-            adjusted_title, body_text, finished_tool_name
-        )
+        snapshot_panel = self._create_snapshot_panel(adjusted_title, body_text, finished_tool_name)
 
         self._print_and_mark_snapshot(tool_sid, snapshot_panel)
 
@@ -949,24 +1581,29 @@ class RichStreamRenderer:
         tool_args: Any,
         _tool_out: Any,
         tool_calls_info: list[tuple[str, Any, Any]],
+        *,
+        tracked_step: Step | None = None,
     ) -> None:
         """Handle agent step event."""
         metadata = event.get("metadata", {})
-        task_id = event.get("task_id")
-        context_id = event.get("context_id")
+        task_id = event.get("task_id") or metadata.get("task_id")
+        context_id = event.get("context_id") or metadata.get("context_id")
         content = event.get("content", "")
 
         # Create steps and panels for the primary tool
         if tool_name:
-            tool_sid = self._ensure_tool_panel(
-                tool_name, tool_args, task_id, context_id
+            tool_sid = self._ensure_tool_panel(tool_name, tool_args, task_id, context_id)
+            self._start_tool_step(
+                task_id,
+                context_id,
+                tool_name,
+                tool_args,
+                tool_sid,
+                tracked_step=tracked_step,
             )
-            self._start_tool_step(task_id, context_id, tool_name, tool_args, tool_sid)
 
         # Handle additional tool calls
-        self._process_additional_tool_calls(
-            tool_calls_info, tool_name, task_id, context_id
-        )
+        self._process_additional_tool_calls(tool_calls_info, tool_name, task_id, context_id)
 
         # Check for tool completion
         (
@@ -976,11 +1613,13 @@ class RichStreamRenderer:
         ) = self._detect_tool_completion(metadata, content)
 
         if is_tool_finished and finished_tool_name:
-            self._finish_tool_panel(
-                finished_tool_name, finished_tool_output, task_id, context_id
-            )
+            self._finish_tool_panel(finished_tool_name, finished_tool_output, task_id, context_id)
             self._finish_tool_step(
-                finished_tool_name, finished_tool_output, task_id, context_id
+                finished_tool_name,
+                finished_tool_output,
+                task_id,
+                context_id,
+                tracked_step=tracked_step,
             )
             self._create_tool_snapshot(finished_tool_name, task_id, context_id)
 
@@ -1030,9 +1669,7 @@ class RichStreamRenderer:
 
     def _get_analysis_progress_info(self) -> dict[str, Any]:
         total_steps = len(self.steps.order)
-        completed_steps = sum(
-            1 for sid in self.steps.order if is_step_finished(self.steps.by_id[sid])
-        )
+        completed_steps = sum(1 for sid in self.steps.order if is_step_finished(self.steps.by_id[sid]))
         current_step = None
         for sid in self.steps.order:
             if not is_step_finished(self.steps.by_id[sid]):
@@ -1040,13 +1677,11 @@ class RichStreamRenderer:
                 break
         # Prefer server elapsed time when available
         elapsed = 0.0
-        if isinstance(self.stream_processor.server_elapsed_time, int | float):
+        if isinstance(self.stream_processor.server_elapsed_time, (int, float)):
             elapsed = float(self.stream_processor.server_elapsed_time)
         elif self._started_at is not None:
             elapsed = monotonic() - self._started_at
-        progress_percent = (
-            int((completed_steps / total_steps) * 100) if total_steps else 0
-        )
+        progress_percent = int((completed_steps / total_steps) * 100) if total_steps else 0
         return {
             "total_steps": total_steps,
             "completed_steps": completed_steps,
@@ -1100,29 +1735,42 @@ class RichStreamRenderer:
     def _format_step_status(self, step: Step) -> str:
         """Format step status with elapsed time or duration."""
         if is_step_finished(step):
-            if step.duration_ms is None:
-                return LESS_THAN_1MS
-            elif step.duration_ms >= 1000:
-                return f"[{step.duration_ms / 1000:.2f}s]"
-            elif step.duration_ms > 0:
-                return f"[{step.duration_ms}ms]"
-            return LESS_THAN_1MS
+            return self._format_finished_badge(step)
         else:
             # Calculate elapsed time for running steps
             elapsed = self._calculate_step_elapsed_time(step)
-            if elapsed >= 1:
+            if elapsed >= 0.1:
                 return f"[{elapsed:.2f}s]"
-            ms = int(elapsed * 1000)
-            return f"[{ms}ms]" if ms > 0 else LESS_THAN_1MS
+            ms = int(round(elapsed * 1000))
+            if ms <= 0:
+                return ""
+            return f"[{ms}ms]"
+
+    def _format_finished_badge(self, step: Step) -> str:
+        """Compose duration badge for finished steps including source tagging."""
+        if getattr(step, "duration_unknown", False) is True:
+            payload = "??s"
+        else:
+            duration_ms = step.duration_ms
+            if duration_ms is None:
+                payload = "<1ms"
+            elif duration_ms < 0:
+                payload = "<1ms"
+            elif duration_ms >= 100:
+                payload = f"{duration_ms / 1000:.2f}s"
+            elif duration_ms > 0:
+                payload = f"{duration_ms}ms"
+            else:
+                payload = "<1ms"
+
+        return f"[{payload}]"
 
     def _calculate_step_elapsed_time(self, step: Step) -> float:
         """Calculate elapsed time for a running step."""
         server_elapsed = self.stream_processor.server_elapsed_time
         server_start = self._step_server_start_times.get(step.step_id)
 
-        if isinstance(server_elapsed, int | float) and isinstance(
-            server_start, int | float
-        ):
+        if isinstance(server_elapsed, (int, float)) and isinstance(server_start, (int, float)):
             return max(0.0, float(server_elapsed) - float(server_start))
 
         try:
@@ -1136,6 +1784,21 @@ class RichStreamRenderer:
             return step.name
         return "thinking..." if step.kind == "agent" else f"{step.kind} step"
 
+    def _resolve_step_label(self, step: Step) -> str:
+        """Return the display label for a step with sensible fallbacks."""
+        raw_label = getattr(step, "display_label", None)
+        label = raw_label.strip() if isinstance(raw_label, str) else ""
+        if label:
+            return normalise_display_label(label)
+
+        if not (step.name or "").strip():
+            return UNKNOWN_STEP_DETAIL
+
+        icon = self._get_step_icon(step.kind)
+        base_name = self._get_step_display_name(step)
+        fallback = " ".join(part for part in (icon, base_name) if part).strip()
+        return normalise_display_label(fallback)
+
     def _check_parallel_tools(self) -> dict[tuple[str | None, str | None], list]:
         """Check for parallel running tools."""
         running_by_ctx: dict[tuple[str | None, str | None], list] = {}
@@ -1158,69 +1821,379 @@ class RichStreamRenderer:
     def _compose_step_renderable(
         self,
         step: Step,
-        running_by_ctx: dict[tuple[str | None, str | None], list],
+        branch_state: tuple[bool, ...],
     ) -> Any:
-        """Compose a single renderable for the steps panel."""
-        finished = is_step_finished(step)
-        status_br = self._format_step_status(step)
-        display_name = self._get_step_display_name(step)
-
-        if (
-            not finished
-            and step.kind == "tool"
-            and self._is_parallel_tool(step, running_by_ctx)
-        ):
-            status_br = status_br.replace("]", " 🔄]")
+        """Compose a single renderable for the hierarchical steps panel."""
+        prefix = build_connector_prefix(branch_state)
+        text_line = self._build_step_text_line(step, prefix)
+        renderables = self._wrap_step_text(step, text_line)
 
-        icon = self._get_step_icon(step.kind)
-        text_line = Text(style="dim")
-        text_line.append(icon)
-        text_line.append(" ")
-        text_line.append(display_name)
-        if status_br:
+        args_renderable = self._build_args_renderable(step, prefix)
+        if args_renderable is not None:
+            renderables.append(args_renderable)
+
+        return self._collapse_renderables(renderables)
+
+    def _build_step_text_line(
+        self,
+        step: Step,
+        prefix: str,
+    ) -> Text:
+        """Create the textual portion of a step renderable."""
+        text_line = Text()
+        text_line.append(prefix, style="dim")
+        text_line.append(self._resolve_step_label(step))
+
+        status_badge = self._format_step_status(step)
+        self._append_status_badge(text_line, step, status_badge)
+        self._append_state_glyph(text_line, step)
+        return text_line
+
+    def _append_status_badge(self, text_line: Text, step: Step, status_badge: str) -> None:
+        """Append the formatted status badge when available."""
+        glyph_key = getattr(step, "status_icon", None)
+        glyph = glyph_for_status(glyph_key)
+
+        if status_badge:
+            text_line.append(" ")
+            text_line.append(status_badge, style="cyan")
+
+        if glyph:
             text_line.append(" ")
-            text_line.append(status_br)
-        if finished:
-            text_line.append(" ✓")
+            style = self._status_icon_style(glyph_key)
+            if style:
+                text_line.append(glyph, style=style)
+            else:
+                text_line.append(glyph)
+
+    def _append_state_glyph(self, text_line: Text, step: Step) -> None:
+        """Append glyph/failure markers in a single place."""
+        failure_reason = (step.failure_reason or "").strip()
+        if failure_reason:
+            text_line.append(f" {failure_reason}")
+
+    @staticmethod
+    def _status_icon_style(icon_key: str | None) -> str | None:
+        """Return style for a given status icon."""
+        if not icon_key:
+            return None
+        return STATUS_ICON_STYLES.get(icon_key)
+
+    def _wrap_step_text(self, step: Step, text_line: Text) -> list[Any]:
+        """Return the base text, optionally decorated with a trailing spinner."""
+        if getattr(step, "status", None) == "running":
+            spinner = self._step_spinners.get(step.step_id)
+            if spinner is None:
+                spinner = Spinner("dots", style="dim")
+                self._step_spinners[step.step_id] = spinner
+            return [TrailingSpinnerLine(text_line, spinner)]
+
+        self._step_spinners.pop(step.step_id, None)
+        return [text_line]
+
+    def _collapse_renderables(self, renderables: list[Any]) -> Any:
+        """Collapse a list of renderables into a single object."""
+        if not renderables:
+            return None
 
-        if finished:
-            return text_line
+        if len(renderables) == 1:
+            return renderables[0]
 
-        spinner = Spinner("dots", text=text_line, style="dim")
-        return Align.left(spinner)
+        return Group(*renderables)
+
+    def _build_args_renderable(self, step: Step, prefix: str) -> Text | Group | None:
+        """Build a dimmed argument line for tool or agent steps."""
+        if step.kind not in {"tool", "delegate", "agent"}:
+            return None
+        if step.kind == "agent" and step.parent_id:
+            return None
+        formatted_args = self._format_step_args(step)
+        if not formatted_args:
+            return None
+        if isinstance(formatted_args, list):
+            return self._build_arg_list(prefix, formatted_args)
+
+        args_text = Text()
+        args_text.append(prefix, style="dim")
+        args_text.append(" " * 5)
+        args_text.append(formatted_args, style="dim")
+        return args_text
+
+    def _build_arg_list(self, prefix: str, formatted_args: list[str | tuple[int, str]]) -> Group | None:
+        """Render multi-line argument entries preserving indentation."""
+        arg_lines: list[Text] = []
+        for indent_level, text_value in self._iter_arg_entries(formatted_args):
+            arg_text = Text()
+            arg_text.append(prefix, style="dim")
+            arg_text.append(" " * 5)
+            arg_text.append(" " * (indent_level * 2))
+            arg_text.append(text_value, style="dim")
+            arg_lines.append(arg_text)
+        if not arg_lines:
+            return None
+        return Group(*arg_lines)
+
+    @staticmethod
+    def _iter_arg_entries(
+        formatted_args: list[str | tuple[int, str]],
+    ) -> Iterable[tuple[int, str]]:
+        """Yield normalized indentation/value pairs for argument entries."""
+        for value in formatted_args:
+            if isinstance(value, tuple) and len(value) == 2:
+                indent_level, text_value = value
+                yield indent_level, str(text_value)
+            else:
+                yield 0, str(value)
+
+    def _format_step_args(self, step: Step) -> str | list[str] | list[tuple[int, str]] | None:
+        """Return a printable representation of tool arguments."""
+        args = getattr(step, "args", None)
+        if args is None:
+            return None
+
+        if isinstance(args, dict):
+            return self._format_dict_args(args, step=step)
+
+        if isinstance(args, (list, tuple)):
+            return self._safe_pretty_args(list(args))
+
+        if isinstance(args, (str, int, float)):
+            return self._stringify_args(args)
+
+        return None
+
+    def _format_dict_args(self, args: dict[str, Any], *, step: Step) -> str | list[str] | list[tuple[int, str]] | None:
+        """Format dictionary arguments with guardrails."""
+        if not args:
+            return None
+
+        masked_args = self._redact_arg_payload(args)
+
+        if self._should_collapse_single_query(step):
+            single_query = self._extract_single_query_arg(masked_args)
+            if single_query:
+                return single_query
+
+        return self._format_dict_arg_lines(masked_args)
+
+    @staticmethod
+    def _extract_single_query_arg(args: dict[str, Any]) -> str | None:
+        """Return a trimmed query argument when it is the only entry."""
+        if len(args) != 1:
+            return None
+        key, value = next(iter(args.items()))
+        if key != "query" or not isinstance(value, str):
+            return None
+        stripped = value.strip()
+        return stripped or None
+
+    @staticmethod
+    def _redact_arg_payload(args: dict[str, Any]) -> dict[str, Any]:
+        """Apply best-effort masking before rendering arguments."""
+        try:
+            cleaned = redact_sensitive(args)
+            return cleaned if isinstance(cleaned, dict) else args
+        except Exception:
+            return args
+
+    @staticmethod
+    def _should_collapse_single_query(step: Step) -> bool:
+        """Return True when we should display raw query text."""
+        if step.kind == "agent":
+            return True
+        if step.kind == "delegate":
+            return True
+        return False
+
+    def _format_dict_arg_lines(self, args: dict[str, Any]) -> list[tuple[int, str]] | None:
+        """Render dictionary arguments as nested YAML-style lines."""
+        lines: list[tuple[int, str]] = []
+        for raw_key, value in args.items():
+            key = str(raw_key)
+            lines.extend(self._format_nested_entry(key, value, indent=0))
+        return lines or None
+
+    def _format_nested_entry(self, key: str, value: Any, indent: int) -> list[tuple[int, str]]:
+        """Format a mapping entry recursively."""
+        lines: list[tuple[int, str]] = []
+
+        if isinstance(value, dict):
+            if value:
+                lines.append((indent, f"{key}:"))
+                lines.extend(self._format_nested_mapping(value, indent + 1))
+            else:
+                lines.append((indent, f"{key}: {{}}"))
+            return lines
+
+        if isinstance(value, (list, tuple, set)):
+            seq_lines = self._format_sequence_entries(list(value), indent + 1)
+            if seq_lines:
+                lines.append((indent, f"{key}:"))
+                lines.extend(seq_lines)
+            else:
+                lines.append((indent, f"{key}: []"))
+            return lines
+
+        formatted_value = self._format_arg_value(value)
+        if formatted_value is not None:
+            lines.append((indent, f"{key}: {formatted_value}"))
+        return lines
+
+    def _format_nested_mapping(self, mapping: dict[str, Any], indent: int) -> list[tuple[int, str]]:
+        """Format nested dictionary values."""
+        nested_lines: list[tuple[int, str]] = []
+        for raw_key, value in mapping.items():
+            key = str(raw_key)
+            nested_lines.extend(self._format_nested_entry(key, value, indent))
+        return nested_lines
+
+    def _format_sequence_entries(self, sequence: list[Any], indent: int) -> list[tuple[int, str]]:
+        """Format list/tuple/set values with YAML-style bullets."""
+        if not sequence:
+            return []
+
+        lines: list[tuple[int, str]] = []
+        for item in sequence:
+            lines.extend(self._format_sequence_item(item, indent))
+        return lines
+
+    def _format_sequence_item(self, item: Any, indent: int) -> list[tuple[int, str]]:
+        """Format a single list entry."""
+        if isinstance(item, dict):
+            return self._format_dict_sequence_item(item, indent)
+
+        if isinstance(item, (list, tuple, set)):
+            return self._format_nested_sequence_item(list(item), indent)
+
+        formatted = self._format_arg_value(item)
+        if formatted is not None:
+            return [(indent, f"- {formatted}")]
+        return []
+
+    def _format_dict_sequence_item(self, mapping: dict[str, Any], indent: int) -> list[tuple[int, str]]:
+        """Format a dictionary entry within a list."""
+        child_lines = self._format_nested_mapping(mapping, indent + 1)
+        if child_lines:
+            return self._prepend_sequence_prefix(child_lines, indent)
+        return [(indent, "- {}")]
+
+    def _format_nested_sequence_item(self, sequence: list[Any], indent: int) -> list[tuple[int, str]]:
+        """Format a nested sequence entry within a list."""
+        child_lines = self._format_sequence_entries(sequence, indent + 1)
+        if child_lines:
+            return self._prepend_sequence_prefix(child_lines, indent)
+        return [(indent, "- []")]
+
+    @staticmethod
+    def _prepend_sequence_prefix(child_lines: list[tuple[int, str]], indent: int) -> list[tuple[int, str]]:
+        """Attach a sequence bullet to the first child line."""
+        _, first_text = child_lines[0]
+        prefixed: list[tuple[int, str]] = [(indent, f"- {first_text}")]
+        prefixed.extend(child_lines[1:])
+        return prefixed
+
+    def _format_arg_value(self, value: Any) -> str | None:
+        """Format a single argument value with per-value truncation."""
+        if value is None:
+            return "null"
+        if isinstance(value, (bool, int, float)):
+            return json.dumps(value, ensure_ascii=False)
+        if isinstance(value, str):
+            return self._format_string_arg_value(value)
+        return _truncate_display(str(value), limit=ARGS_VALUE_MAX_LEN)
+
+    @staticmethod
+    def _format_string_arg_value(value: str) -> str:
+        """Return a trimmed, quoted representation of a string argument."""
+        sanitised = value.replace("\n", " ").strip()
+        sanitised = sanitised.replace('"', '\\"')
+        trimmed = _truncate_display(sanitised, limit=ARGS_VALUE_MAX_LEN)
+        return f'"{trimmed}"'
+
+    @staticmethod
+    def _safe_pretty_args(args: dict[str, Any]) -> str | None:
+        """Defensively format argument dictionaries."""
+        try:
+            return pretty_args(args, max_len=160)
+        except Exception:
+            return str(args)
+
+    @staticmethod
+    def _stringify_args(args: Any) -> str | None:
+        """Format non-dictionary argument payloads."""
+        text = str(args).strip()
+        if not text:
+            return None
+        return _truncate_display(text)
 
     def _render_steps_text(self) -> Any:
         """Render the steps panel content."""
         if not (self.steps.order or self.steps.children):
-            return Text("No steps yet", style="dim")
+            return _NO_STEPS_TEXT.copy()
+
+        nodes = list(self.steps.iter_tree())
+        if not nodes:
+            return _NO_STEPS_TEXT.copy()
+
+        window = self._summary_window_size()
+        display_nodes, header_notice, footer_notice = clamp_step_nodes(
+            nodes,
+            window=window,
+            get_label=self._get_step_label,
+            get_parent=self._get_step_parent,
+        )
+        step_renderables = self._build_step_renderables(display_nodes)
 
-        running_by_ctx = self._check_parallel_tools()
-        renderables: list[Any] = []
-        for sid in self.steps.order:
-            line = self._compose_step_renderable(self.steps.by_id[sid], running_by_ctx)
-            renderables.append(line)
+        if not step_renderables and not header_notice and not footer_notice:
+            return _NO_STEPS_TEXT.copy()
 
-        if not renderables:
-            return Text("No steps yet", style="dim")
+        return self._assemble_step_renderables(step_renderables, header_notice, footer_notice)
 
-        return Group(*renderables)
+    def _get_step_label(self, step_id: str) -> str:
+        """Get label for a step by ID."""
+        step = self.steps.by_id.get(step_id)
+        if step:
+            return self._resolve_step_label(step)
+        return UNKNOWN_STEP_DETAIL
 
-    def _should_skip_finished_panel(self, sid: str, status: str) -> bool:
-        """Check if a finished panel should be skipped."""
-        if status != "finished":
-            return False
+    def _get_step_parent(self, step_id: str) -> str | None:
+        """Get parent ID for a step by ID."""
+        step = self.steps.by_id.get(step_id)
+        return step.parent_id if step else None
 
-        if getattr(self.cfg, "append_finished_snapshots", False):
-            return True
+    def _summary_window_size(self) -> int:
+        """Return the active window size for step display."""
+        if self.state.finalizing_ui:
+            return 0
+        return int(self.cfg.summary_display_window or 0)
 
-        return (
-            not self.state.finalizing_ui
-            and sid not in self.stream_processor.current_event_finished_panels
-        )
+    def _assemble_step_renderables(self, step_renderables: list[Any], header_notice: Any, footer_notice: Any) -> Any:
+        """Assemble step renderables with header and footer into final output."""
+        renderables: list[Any] = []
+        if header_notice is not None:
+            renderables.append(header_notice)
+        renderables.extend(step_renderables)
+        if footer_notice is not None:
+            renderables.append(footer_notice)
 
-    def _update_final_duration(
-        self, duration: float | None, *, overwrite: bool = False
-    ) -> None:
+        if len(renderables) == 1:
+            return renderables[0]
+
+        return Group(*renderables)
+
+    def _build_step_renderables(self, display_nodes: list[tuple[str, tuple[bool, ...]]]) -> list[Any]:
+        """Convert step nodes to renderables for the steps panel."""
+        renderables: list[Any] = []
+        for step_id, branch_state in display_nodes:
+            step = self.steps.by_id.get(step_id)
+            if not step:
+                continue
+            renderable = self._compose_step_renderable(step, branch_state)
+            if renderable is not None:
+                renderables.append(renderable)
+        return renderables
+
+    def _update_final_duration(self, duration: float | None, *, overwrite: bool = False) -> None:
         """Store formatted duration for eventual final panels."""
         if duration is None:
             return
@@ -1240,20 +2213,7 @@ class RichStreamRenderer:
 
         self.state.final_duration_seconds = duration_val
         self.state.final_duration_text = self._format_elapsed_time(duration_val)
-
-    def _calculate_elapsed_time(self, meta: dict[str, Any]) -> str:
-        """Calculate elapsed time string for running tools."""
-        server_elapsed = self.stream_processor.server_elapsed_time
-        server_start = meta.get("server_started_at")
-
-        if isinstance(server_elapsed, int | float) and isinstance(
-            server_start, int | float
-        ):
-            elapsed = max(0.0, float(server_elapsed) - float(server_start))
-        else:
-            elapsed = max(0.0, monotonic() - (meta.get("started_at") or 0.0))
-
-        return self._format_elapsed_time(elapsed)
+        self._apply_root_duration(duration_val)
 
     def _format_elapsed_time(self, elapsed: float) -> str:
         """Format elapsed time as a readable string."""
@@ -1264,110 +2224,10 @@ class RichStreamRenderer:
         else:
             return "<1ms"
 
-    def _calculate_finished_duration(self, meta: dict[str, Any]) -> str | None:
-        """Calculate duration string for finished tools."""
-        dur = meta.get("duration_seconds")
-        if isinstance(dur, int | float):
-            return self._format_elapsed_time(dur)
-
-        try:
-            server_now = self.stream_processor.server_elapsed_time
-            server_start = meta.get("server_started_at")
-            if isinstance(server_now, int | float) and isinstance(
-                server_start, int | float
-            ):
-                dur = max(0.0, float(server_now) - float(server_start))
-            elif meta.get("started_at") is not None:
-                dur = max(0.0, float(monotonic() - meta.get("started_at")))
-        except Exception:
-            dur = None
-
-        return self._format_elapsed_time(dur) if isinstance(dur, int | float) else None
-
-    def _process_running_tool_panel(
-        self,
-        title: str,
-        meta: dict[str, Any],
-        body: str,
-        *,
-        include_spinner: bool = False,
-    ) -> tuple[str, str] | tuple[str, str, str | None]:
-        """Process a running tool panel."""
-        elapsed_str = self._calculate_elapsed_time(meta)
-        adjusted_title = f"{title}  · {elapsed_str}"
-        chip = f"⏱ {elapsed_str}"
-        spinner_message: str | None = None
-
-        if not body.strip():
-            body = ""
-            spinner_message = f"{title} running... {elapsed_str}"
-        else:
-            body = f"{body}\n\n{chip}"
-
-        if include_spinner:
-            return adjusted_title, body, spinner_message
-        return adjusted_title, body
-
-    def _process_finished_tool_panel(self, title: str, meta: dict[str, Any]) -> str:
-        """Process a finished tool panel."""
-        duration_str = self._calculate_finished_duration(meta)
-        return f"{title}  · {duration_str}" if duration_str else title
-
-    def _create_tool_panel_for_session(
-        self, sid: str, meta: dict[str, Any]
-    ) -> AIPPanel | None:
-        """Create a single tool panel for the session."""
-        title = meta.get("title") or "Tool"
-        status = meta.get("status") or "running"
-        chunks = meta.get("chunks") or []
-        is_delegation = bool(meta.get("is_delegation"))
-
-        if self._should_skip_finished_panel(sid, status):
-            return None
-
-        body = "".join(chunks)
-        adjusted_title = title
-
-        spinner_message: str | None = None
-
-        if status == "running":
-            adjusted_title, body, spinner_message = self._process_running_tool_panel(
-                title, meta, body, include_spinner=True
-            )
-        elif status == "finished":
-            adjusted_title = self._process_finished_tool_panel(title, meta)
-
-        return create_tool_panel(
-            title=adjusted_title,
-            content=body,
-            status=status,
-            theme=self.cfg.theme,
-            is_delegation=is_delegation,
-            spinner_message=spinner_message,
-        )
-
-    def _render_tool_panels(self) -> list[AIPPanel]:
-        """Render tool execution output panels."""
-        if not getattr(self.cfg, "show_delegate_tool_panels", False):
-            return []
-        panels: list[AIPPanel] = []
-        for sid in self.tool_order:
-            meta = self.tool_panels.get(sid) or {}
-            panel = self._create_tool_panel_for_session(sid, meta)
-            if panel:
-                panels.append(panel)
-
-        return panels
-
     def _format_dict_or_list_output(self, output_value: dict | list) -> str:
         """Format dict/list output as pretty JSON."""
         try:
-            return (
-                self.OUTPUT_PREFIX
-                + "```json\n"
-                + json.dumps(output_value, indent=2)
-                + "\n```\n"
-            )
+            return self.OUTPUT_PREFIX + "```json\n" + json.dumps(output_value, indent=2) + "\n```\n"
         except Exception:
             return self.OUTPUT_PREFIX + str(output_value) + "\n"
 
@@ -1391,12 +2251,7 @@ class RichStreamRenderer:
         """Format string that looks like JSON."""
         try:
             parsed = json.loads(output)
-            return (
-                self.OUTPUT_PREFIX
-                + "```json\n"
-                + json.dumps(parsed, indent=2)
-                + "\n```\n"
-            )
+            return self.OUTPUT_PREFIX + "```json\n" + json.dumps(parsed, indent=2) + "\n```\n"
         except Exception:
             return self.OUTPUT_PREFIX + output + "\n"
 
@@ -1406,9 +2261,7 @@ class RichStreamRenderer:
         s = self._clean_sub_agent_prefix(s, tool_name)
 
         # If looks like JSON, pretty print it
-        if (s.startswith("{") and s.endswith("}")) or (
-            s.startswith("[") and s.endswith("]")
-        ):
+        if (s.startswith("{") and s.endswith("}")) or (s.startswith("[") and s.endswith("]")):
             return self._format_json_string_output(s)
 
         return self.OUTPUT_PREFIX + s + "\n"
@@ -1422,7 +2275,7 @@ class RichStreamRenderer:
 
     def _format_output_block(self, output_value: Any, tool_name: str | None) -> str:
         """Format an output value for panel display."""
-        if isinstance(output_value, dict | list):
+        if isinstance(output_value, (dict, list)):
             return self._format_dict_or_list_output(output_value)
         elif isinstance(output_value, str):
             return self._format_string_output(output_value, tool_name)