superqode 2.4.34__tar.gz → 2.4.35__tar.gz

{superqode-2.4.34 → superqode-2.4.35}/CHANGELOG.md

@@ -7,6 +7,41 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [2.4.35] - 2026-09-30
+
+### Fixed
+
+- Remove numbered turn banners from the transcript, keeping the clean user
+  heading and prompt rail.
+- Label shell tool actions as Run, and keep failed tools compact with a red
+  cross and a clickable numbered detail view. Show failure cards only in
+  verbose mode while retaining recorded command, output, and exit metadata.
+- Move the moving three-dot working indicator into the empty prompt box. Keep
+  drafts editable, pause its timer while typing or unfocused, and restore the
+  usual prompt hint after work or when an approval needs an answer.
+  Use a bright theme accent for the label and random theme colors for the dots.
+- Show the keyboard selection on Free, Budget, and other BYOK model rows,
+  keeping the highlighted model visible when navigating across groups.
+- Preserve tool-call IDs across Core harness turns, saved sessions, and context
+  compaction. Repair matching legacy tool results before sending history so
+  strict providers such as OpenRouter do not reject subsequent prompts.
+- Show branded top and bottom animations for three seconds when a run starts,
+  then in one-second bursts every five seconds, keeping live status visible
+  between bursts. Pause animations when the terminal loses focus and stop burst
+  timers when the run ends.
+- Refresh catalogs once at startup instead of hourly, and load LiteLLM only
+  when needed by a selected connection instead of preloading on every launch.
+- Run file searches, previews, filesystem autocomplete, and Git diff collection
+  away from the TUI event loop. Reuse short-lived completion metadata, discard
+  stale results, and keep typing responsive while slow filesystem work finishes.
+- Stream directory-search files and retain only the first 50 preview lines
+  instead of allocating whole files for those operations.
+- Keep local warmup enabled, clearly label questions waiting for model readiness,
+  and retain queued questions when warmup fails. Use one Ollama availability check
+  and disable reasoning only for its tiny warmup request.
+- Pass Ollama context and GPU options through LiteLLM without a nested options
+  object that the server ignores; preserve explicit caller settings.
+
 ## [2.4.34] - 2026-09-30
 
 ### Fixed

{superqode-2.4.34 → superqode-2.4.35}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: superqode
-Version: 2.4.34
+Version: 2.4.35
 Summary: The harness layer for coding agents
 Author-email: Shashi Jagtap <info@super-agentic.ai>
 Maintainer-email: Shashi Jagtap <info@super-agentic.ai>

{superqode-2.4.34 → superqode-2.4.35}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "superqode"
-version = "2.4.34"
+version = "2.4.35"
 description = "The harness layer for coding agents"
 readme = "README.md"
 license = "Apache-2.0"

{superqode-2.4.34 → superqode-2.4.35}/src/superqode/__init__.py

@@ -38,7 +38,7 @@ __all__ = [
     "sidebar",
 ]
 
-__version__ = "2.4.34"
+__version__ = "2.4.35"
 
 # Stable, lightweight public extension surface.  Importing this package does
 # not discover or execute third-party extensions; discovery happens only when

{superqode-2.4.34 → superqode-2.4.35}/src/superqode/agent/loop.py

@@ -696,6 +696,13 @@ def repair_dangling_tool_calls(messages: List["AgentMessage"]) -> List["AgentMes
     i, n = 0, len(messages)
     while i < n:
         msg = messages[i]
+        if msg.role == "tool":
+            # A bounded history/compacted tail can begin after its assistant
+            # call was removed. Keep the output as context, never as an orphan
+            # protocol tool message that a strict provider will reject.
+            out.append(replace(msg, role="user", tool_call_id=None, name=None))
+            i += 1
+            continue
         out.append(msg)
         tool_calls = msg.tool_calls if msg.role == "assistant" else None
         if not tool_calls:
@@ -708,14 +715,42 @@ def repair_dangling_tool_calls(messages: List["AgentMessage"]) -> List["AgentMes
         while j < n and messages[j].role == "tool":
             following.append(messages[j])
             j += 1
-        out.extend(following)
-
-        answered_ids = {m.tool_call_id for m in following if m.tool_call_id}
-        # Calls not satisfied by a matching id.
-        remaining = [tc for tc in tool_calls if not (tc.get("id") and tc.get("id") in answered_ids)]
-        # Tool results without an id answer the remaining calls positionally.
-        idless = sum(1 for m in following if not m.tool_call_id)
-        unanswered = remaining[idless:]
+        # Give incomplete assistant call records IDs without mutating the
+        # stored history, and use the same IDs for their repaired results.
+        tool_calls = [
+            tc
+            if isinstance(tc.get("id"), str) and tc["id"].strip()
+            else dict(tc, id=str(uuid.uuid4()))
+            for tc in tool_calls
+        ]
+        if tool_calls != msg.tool_calls:
+            out[-1] = replace(msg, tool_calls=tool_calls)
+        answered_ids = {
+            m.tool_call_id
+            for m in following
+            if isinstance(m.tool_call_id, str) and m.tool_call_id.strip()
+        }
+        remaining = [tc for tc in tool_calls if tc["id"] not in answered_ids]
+        call_ids = {tc["id"] for tc in tool_calls}
+        seen = set()
+        orphaned = []
+        for result in following:
+            identifier = result.tool_call_id
+            if not isinstance(identifier, str) or not identifier.strip():
+                matches = [
+                    tc for tc in remaining if (tc.get("function") or {}).get("name") == result.name
+                ]
+                call = matches[0] if matches else (remaining[0] if remaining else None)
+                if call is not None:
+                    remaining.remove(call)
+                    result = replace(result, tool_call_id=call["id"])
+                    identifier = call["id"]
+            if isinstance(identifier, str) and identifier in call_ids and identifier not in seen:
+                out.append(result)
+                seen.add(identifier)
+            else:
+                orphaned.append(replace(result, role="user", tool_call_id=None, name=None))
+        unanswered = [tc for tc in tool_calls if tc["id"] not in seen]
 
         for tc in unanswered:
             tc_id = tc.get("id") or str(uuid.uuid4())
@@ -732,6 +767,7 @@ def repair_dangling_tool_calls(messages: List["AgentMessage"]) -> List["AgentMes
                     name=name,
                 )
             )
+        out.extend(orphaned)
         i = j
     return out
 
@@ -1290,6 +1326,7 @@ class AgentLoop:
                     role=message.role,
                     content=message.content,
                     tool_calls=message.tool_calls,
+                    tool_call_id=message.tool_call_id,
                     name=message.tool_name,
                 )
             )
@@ -1391,6 +1428,9 @@ class AgentLoop:
                         "role": m.role,
                         "content": _content_for_counting(m.content),
                         "tool_calls": m.tool_calls,
+                        "tool_call_id": m.tool_call_id,
+                        "name": m.name,
+                        "reasoning_content": m.reasoning_content,
                     }
                     for m in messages
                 ]
@@ -1983,6 +2023,9 @@ class AgentLoop:
                 "role": m.role,
                 "content": _content_for_counting(m.content),
                 "tool_calls": m.tool_calls,
+                "tool_call_id": m.tool_call_id,
+                "name": m.name,
+                "reasoning_content": m.reasoning_content,
                 "tool_result": m.content if m.role == "tool" else None,
             }
             for m in messages
@@ -2028,6 +2071,9 @@ class AgentLoop:
                     "role": m.role,
                     "content": _content_for_counting(m.content),
                     "tool_calls": m.tool_calls,
+                    "tool_call_id": m.tool_call_id,
+                    "name": m.name,
+                    "reasoning_content": m.reasoning_content,
                     "tool_result": m.content if m.role == "tool" else None,
                 }
                 for m in messages
@@ -2077,6 +2123,9 @@ class AgentLoop:
                     "role": m.role,
                     "content": _content_for_counting(m.content),
                     "tool_calls": m.tool_calls,
+                    "tool_call_id": m.tool_call_id,
+                    "name": m.name,
+                    "reasoning_content": m.reasoning_content,
                     "tool_result": m.content if m.role == "tool" else None,
                 }
                 for m in messages
@@ -2120,6 +2169,9 @@ class AgentLoop:
                 "role": m.role,
                 "content": _content_for_counting(m.content),
                 "tool_calls": m.tool_calls,
+                "tool_call_id": m.tool_call_id,
+                "name": m.name,
+                "reasoning_content": m.reasoning_content,
                 "tool_result": m.content if m.role == "tool" else None,
             }
             for m in body
@@ -2152,7 +2204,14 @@ class AgentLoop:
             # Fallback: mechanical prune-from-front (existing path).
             pruned_dicts = self.context_manager.prune_history(msg_dicts)
             result_messages = [
-                AgentMessage(role=d["role"], content=d["content"], tool_calls=d.get("tool_calls"))
+                AgentMessage(
+                    role=d["role"],
+                    content=d["content"],
+                    tool_calls=d.get("tool_calls"),
+                    tool_call_id=d.get("tool_call_id"),
+                    name=d.get("name"),
+                    reasoning_content=d.get("reasoning_content"),
+                )
                 for d in pruned_dicts
             ]
 
@@ -2792,7 +2851,9 @@ class AgentLoop:
                         )
                     )
                     if self._session_manager:
-                        self._session_manager.add_tool_result(tool_name, result.to_message())
+                        self._session_manager.add_tool_result(
+                            tool_name, result.to_message(), tool_call_id=tool_call_id
+                        )
                     image_msg = self._image_followup_message(result)
                     if image_msg is not None:
                         messages.append(image_msg)
@@ -3344,7 +3405,9 @@ class AgentLoop:
                         )
                     )
                     if self._session_manager:
-                        self._session_manager.add_tool_result(tool_name, result.to_message())
+                        self._session_manager.add_tool_result(
+                            tool_name, result.to_message(), tool_call_id=tool_call_id
+                        )
                     image_msg = self._image_followup_message(result)
                     if image_msg is not None:
                         messages.append(image_msg)

{superqode-2.4.34 → superqode-2.4.35}/src/superqode/agent/session_manager.py

@@ -59,6 +59,7 @@ class SessionMessage:
     tool_calls: Optional[List[Dict]] = None
     tool_name: Optional[str] = None
     tool_result: Optional[str] = None
+    tool_call_id: Optional[str] = None
 
 
 def _metadata_from_dict(data: Dict[str, Any]) -> SessionMetadata:
@@ -257,12 +258,14 @@ class SessionStore:
         session_id: str,
         tool_name: str,
         result: str,
+        tool_call_id: Optional[str] = None,
     ):
-        """Append a tool result message."""
+        """Append a tool result message, preserving its assistant-call link."""
         message = SessionMessage(
             role="tool",
             content=result,
             tool_name=tool_name,
+            tool_call_id=tool_call_id,
         )
         self.append_message(session_id, message)
 
@@ -487,11 +490,11 @@ class SessionManager:
             SessionMessage(role="assistant", content=content, tool_calls=tool_calls),
         )
 
-    def add_tool_result(self, tool_name: str, result: str):
+    def add_tool_result(self, tool_name: str, result: str, tool_call_id: Optional[str] = None):
         """Add tool result to current session."""
         if not self._current_session_id:
             raise RuntimeError("No active session. Call start_session first.")
-        self.store.append_tool_result(self._current_session_id, tool_name, result)
+        self.store.append_tool_result(self._current_session_id, tool_name, result, tool_call_id)
 
     def get_messages(self, limit: Optional[int] = None) -> List[SessionMessage]:
         """Get messages from current session."""

{superqode-2.4.34 → superqode-2.4.35}/src/superqode/app/inputs.py

@@ -2,10 +2,18 @@
 
 from __future__ import annotations
 
+from time import monotonic
+from random import sample
+
+from rich.text import Text
 from textual import events
 from textual.binding import Binding
+from textual.content import Content
+from textual.strip import Strip
 from textual.widgets import Input, TextArea
 
+from superqode.app.constants import THEME
+
 
 class SelectionAwareInput(TextArea):
     """
@@ -24,6 +32,71 @@ class SelectionAwareInput(TextArea):
     DEFAULT_PLACEHOLDER = (
         "Get started with :connect · Browse with mouse · Run shell commands with >"
     )
+    WORKING_DOT_FRAMES = ("●··", "·●·", "··●")
+
+    def set_working(self, working: bool) -> None:
+        """Show activity in an empty composer without changing its draft."""
+        if working != getattr(self, "_is_working", False):
+            if working:
+                self._ready_placeholder = self.placeholder
+                self._working_started_at = monotonic()
+            else:
+                if self.placeholder == getattr(self, "_working_placeholder", None):
+                    self.placeholder = self._ready_placeholder
+            self._is_working = working
+        self._sync_working_animation()
+
+    def _update_working_placeholder(self) -> None:
+        if not getattr(self, "_is_working", False) or self.text:
+            return
+        elapsed = max(0.0, monotonic() - self._working_started_at)
+        frame = self.WORKING_DOT_FRAMES[int(elapsed * 2) % len(self.WORKING_DOT_FRAMES)]
+        self._working_placeholder = f"Agent working {frame}"
+        accents = list(
+            dict.fromkeys(THEME[key] for key in ("purple", "pink", "gold", "cyan", "success"))
+        )
+        colors = sample(accents, min(3, len(accents)))
+        content = Text("Agent working ", style=f"bold {THEME['purple']}")
+        for index, dot in enumerate(frame):
+            content.append(dot, style=f"bold {colors[index % len(colors)]}")
+        self._working_content = content
+        self.placeholder = self._working_placeholder
+
+    def render_line(self, y: int) -> Strip:
+        # TextArea applies its muted placeholder style over rich text. Render
+        # only the working hint here so theme accents retain their brightness.
+        if (
+            getattr(self, "_is_working", False)
+            and not self.text
+            and self.placeholder == getattr(self, "_working_placeholder", None)
+        ):
+            lines = Content.from_text(self._working_content).wrap(max(1, self.content_size.width))
+            if 0 <= y < len(lines):
+                content = lines[y]
+                return Strip(content.render_segments(self.visual_style), content.cell_length)
+        return super().render_line(y)
+
+    def _sync_working_animation(self) -> None:
+        active = (
+            getattr(self, "_is_working", False)
+            and not self.text
+            and getattr(self.app, "_wave_window_focused", True)
+        )
+        timer = getattr(self, "_working_timer", None)
+        if not active:
+            if timer is not None:
+                timer.stop()
+                self._working_timer = None
+            return
+        self._update_working_placeholder()
+        if timer is None:
+            self._working_timer = self.set_interval(0.5, self._update_working_placeholder)
+
+    def on_unmount(self) -> None:
+        timer = getattr(self, "_working_timer", None)
+        if timer is not None:
+            timer.stop()
+            self._working_timer = None
 
     # A prompt box should behave like an ordinary text field. TextArea's defaults
     # are surprising here: Ctrl+A is line-start and Ctrl+U only deletes to the
@@ -66,6 +139,8 @@ class SelectionAwareInput(TextArea):
         kwargs.setdefault("highlight_cursor_line", False)
         kwargs.setdefault("tab_behavior", "focus")
         super().__init__(*args, **kwargs)
+        self._working_timer = None
+        self._is_working = False
         self.suggester = suggester
         self._history_draft: Optional[str] = None
         self._navigating_history: bool = False
@@ -126,6 +201,7 @@ class SelectionAwareInput(TextArea):
 
     def on_text_area_changed(self, event: TextArea.Changed) -> None:
         self._resize_to_content()
+        self._sync_working_animation()
         schedule = getattr(self.app, "_schedule_draft_save", None)
         if callable(schedule):
             schedule()
@@ -288,6 +364,11 @@ class SelectionAwareInput(TextArea):
                 event.prevent_default()
                 return
 
+        if event.key == "escape" and getattr(app, "_completion_ready_value", None) is None:
+            hide_completion = getattr(app, "_hide_prompt_completion_panel", None)
+            if callable(hide_completion):
+                hide_completion()
+
         if event.key in {"escape", "ctrl+["} and getattr(self, "_navigating_history", False):
             draft = (
                 getattr(self, "_history_draft", None)

{superqode-2.4.34 → superqode-2.4.35}/src/superqode/app/mixins/agent_run.py

@@ -134,6 +134,77 @@ class AgentRunMixin:
         except Exception:
             pass
 
+    WAVE_INITIAL_BURST_SECONDS = 3.0
+    WAVE_BURST_SECONDS = 1.0
+    WAVE_BURST_INTERVAL = 5.0
+
+    def _set_wave_motion(self, active: bool) -> None:
+        for selector in ("#thinking-wave", "#thinking-wave-bottom"):
+            try:
+                wave = self.query_one(selector)
+                wave.is_active = active
+                if getattr(self, "_wave_burst_running", False):
+                    wave.add_class("visible")
+                else:
+                    wave.remove_class("visible")
+            except Exception:
+                pass
+
+    def _begin_wave_bursts(self) -> None:
+        if getattr(self, "_wave_burst_running", False):
+            return  # Streaming and preparation share one run's schedule.
+        if not getattr(self, "_wave_window_focused", True):
+            self._wave_resume_on_focus = True
+            try:
+                self.query_one("#streaming-thinking").auto_refresh = None
+            except Exception:
+                pass
+            return
+        self._wave_burst_running = True
+        self._wave_generation = getattr(self, "_wave_generation", 0) + 1
+        generation = self._wave_generation
+        self._show_wave_burst(generation, duration=self.WAVE_INITIAL_BURST_SECONDS)
+        self._wave_repeat_timer = self.set_interval(
+            self.WAVE_BURST_INTERVAL, lambda: self._show_wave_burst(generation)
+        )
+
+    def _show_wave_burst(
+        self, generation: int | None = None, *, duration: float | None = None
+    ) -> None:
+        if generation is not None and generation != getattr(self, "_wave_generation", 0):
+            return
+        if not getattr(self, "is_busy", False):
+            self._stop_wave_bursts()
+            return
+        if not getattr(self, "_wave_burst_running", False):
+            return
+        timer = getattr(self, "_wave_end_timer", None)
+        if timer is not None:
+            timer.stop()
+        self._set_wave_motion(True)
+        generation = getattr(self, "_wave_generation", 0)
+        self._wave_end_timer = self.set_timer(
+            self.WAVE_BURST_SECONDS if duration is None else duration,
+            lambda: self._end_wave_burst(generation),
+        )
+
+    def _end_wave_burst(self, generation: int | None = None) -> None:
+        if generation is not None and generation != getattr(self, "_wave_generation", 0):
+            return
+        self._wave_end_timer = None
+        self._set_wave_motion(False)
+
+    def _stop_wave_bursts(self) -> None:
+        self._wave_generation = getattr(self, "_wave_generation", 0) + 1
+        self._wave_burst_running = False
+        self._wave_resume_on_focus = False
+        for attribute in ("_wave_repeat_timer", "_wave_end_timer"):
+            timer = getattr(self, attribute, None)
+            if timer is not None:
+                timer.stop()
+            setattr(self, attribute, None)
+        self._set_wave_motion(False)
+
     def _start_stream_animation(self, log: ConversationLog):
         """Start animation during agent streaming."""
         self._stream_animation_frame = 0
@@ -153,24 +224,12 @@ class AgentRunMixin:
         except Exception:
             pass
 
-        # Show the branded top scanning line.
-        try:
-            thinking_wave = self.query_one("#thinking-wave", TopScanningLine)
-            thinking_wave.is_active = True
-            thinking_wave.add_class("visible")
-        except Exception:
-            pass
-
-        try:
-            bottom_wave = self.query_one("#thinking-wave-bottom", BottomScanningLine)
-            bottom_wave.is_active = True
-            bottom_wave.add_class("visible")
-        except Exception:
-            pass
+        self._begin_wave_bursts()
 
     def _stop_stream_animation(self):
         """Stop the streaming animation."""
         self.is_busy = False
+        self._stop_wave_bursts()
 
         self._set_composer_working_state(False)
 
@@ -217,20 +276,7 @@ class AgentRunMixin:
         except Exception:
             pass
 
-        # Show scanning line animation at TOP
-        try:
-            thinking_wave = self.query_one("#thinking-wave", TopScanningLine)
-            thinking_wave.is_active = True
-            thinking_wave.add_class("visible")
-        except Exception:
-            pass
-
-        try:
-            bottom_wave = self.query_one("#thinking-wave-bottom", BottomScanningLine)
-            bottom_wave.is_active = True
-            bottom_wave.add_class("visible")
-        except Exception:
-            pass
+        self._begin_wave_bursts()
 
         self._set_composer_working_state(True)
 
@@ -241,6 +287,7 @@ class AgentRunMixin:
             show_done: If True, show "Done in X.Xs" message. Default False for streaming.
         """
         self.is_busy = False
+        self._stop_wave_bursts()
 
         # Calm mode: commit the end-of-turn action roll-up before tearing down.
         if self._is_calm_output() and getattr(self, "_calm_actions", 0) > 0:
@@ -672,7 +719,9 @@ class AgentRunMixin:
         try:
             root_path = Path(os.getcwd())
             pre_existing_modified = {
-                change.path for change in get_git_changes(root_path) if change.status in ("M", "A")
+                change.path
+                for change in await asyncio.to_thread(get_git_changes, root_path)
+                if change.status in ("M", "A")
             }
         except Exception:
             pre_existing_modified = set()
@@ -1173,7 +1222,7 @@ class AgentRunMixin:
                     # Merge tool-tracked writes with git changes detected after the run.
                     try:
                         root_path = Path(os.getcwd())
-                        git_changes = get_git_changes(root_path)
+                        git_changes = await asyncio.to_thread(get_git_changes, root_path)
                         git_files_modified = [
                             change.path
                             for change in git_changes
@@ -1186,7 +1235,7 @@ class AgentRunMixin:
                         pass
 
                     # Compute file diffs for detected files
-                    file_diffs = self._compute_file_diffs(files_modified)
+                    file_diffs = await asyncio.to_thread(self._compute_file_diffs, files_modified)
 
                     # Use ACP-style outcome display for both ACP and BYOK
                     outcome_agent = (
@@ -1263,7 +1312,7 @@ class AgentRunMixin:
                     # Compute file diffs even when no text response.
                     try:
                         root_path = Path(os.getcwd())
-                        git_changes = get_git_changes(root_path)
+                        git_changes = await asyncio.to_thread(get_git_changes, root_path)
                         git_files_modified = [
                             change.path
                             for change in git_changes
@@ -1275,7 +1324,7 @@ class AgentRunMixin:
                     except Exception:
                         pass
 
-                    file_diffs = self._compute_file_diffs(files_modified)
+                    file_diffs = await asyncio.to_thread(self._compute_file_diffs, files_modified)
 
                     # Show completion summary with file changes
                     self._show_completion_summary(

{superqode-2.4.34 → superqode-2.4.35}/src/superqode/app/mixins/clickable_commands.py

@@ -173,6 +173,10 @@ class ClickableCommandMixin:
 
     def _run_clicked_command(self, command: str) -> None:
         command = command.strip().lstrip(":")
+        parts = command.split()
+        if len(parts) == 2 and parts[0] == "tools" and parts[1].isdigit():
+            self._show_tools(parts[1], self._clicked_command_log())
+            return
         if command not in CLICKABLE_COMMANDS:
             return
         if command == "back":