kon-coding-agent 0.3.8__tar.gz → 0.3.9__tar.gz

{kon_coding_agent-0.3.8 → kon_coding_agent-0.3.9}/CHANGELOG.md

@@ -6,6 +6,41 @@ All notable changes to this project will be documented in this file.
 
 - No changes yet.
 
+## 0.3.9 - 2026-05-20
+
+### Added
+
+- Added inline image display from tool results in chat UI.
+- Added tree view for handoff navigation.
+- Added ability to edit queued messages.
+- Added Kanagawa Dragon theme.
+- Added lazy provider loading for faster startup - @Meltedd.
+
+### Changed
+
+- Swapped thinking block keybindings.
+- Grouped preferences under settings.
+- Refreshed startup resource display.
+- Smoother streaming experience.
+- Updated permission mode symbols — single tick for auto, stop icon for prompt.
+- Lowercase slash command descriptions, `L` prefix for queue items.
+- Updated README with ASCII art title, new screenshot, and refined styling.
+
+### Fixed
+
+- Fixed Codex provider event handling, transport, and tracking - @Meltedd.
+- Fixed tool expansion crash on startup - @0xku.
+- Fixed Enter key not submitting permission prompts.
+- Fixed manual shell output expand behavior.
+- Fixed streaming cursor removal.
+- Fixed tool output top padding missing blank line.
+- Fixed tree selector display and empty tree state alignment.
+- Fixed input cursor visibility in light themes.
+- Fixed floating list popup styling.
+- Fixed selected color alignment across themes and solarized-light dim color.
+- Fixed legacy Shift+Enter mapping.
+- Fixed Windows startup warning by delaying textual image import - @sukhbinder.
+
 ## 0.3.8 - 2026-05-08
 
 ### Added

{kon_coding_agent-0.3.8 → kon_coding_agent-0.3.9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: kon-coding-agent
-Version: 0.3.8
+Version: 0.3.9
 Summary: Minimal coding agent
 License-File: LICENSE
 Requires-Python: >=3.12
@@ -15,11 +15,16 @@ Requires-Dist: pillow>=12.1.1
 Requires-Dist: pydantic>=2.12.5
 Requires-Dist: readability-lxml>=0.8.4
 Requires-Dist: rich>=14.3.2
+Requires-Dist: textual-image[textual]>=0.12.0
 Requires-Dist: textual>=8.0.0
 Description-Content-Type: text/markdown
 
-<h1 align="center">Kon</h1>
-<p align="center">A minimal terminal coding agent with a small core, strong defaults, and user-owned context.</p>
+<pre align="center">
+░█░█░█▀█░█▀█
+░█▀▄░█░█░█░█
+░▀░▀░▀▀▀░▀░▀
+</pre>
+<p align="center">Minimal coding agent harness</p>
 <p align="center">
   <a href="https://pypi.org/project/kon-coding-agent/"><img alt="PyPI" src="https://img.shields.io/pypi/v/kon-coding-agent?style=flat-square" /></a>
   <a href="https://www.python.org/downloads/release/python-3120/"><img alt="Python" src="https://img.shields.io/badge/python-3.12%2B-blue?style=flat-square" /></a>
@@ -27,7 +32,7 @@ Description-Content-Type: text/markdown
 </p>
 
 <p align="center">
-  <img src="docs/images/kon-screenshot.png" alt="Kon terminal UI screenshot" width="490" />
+  <img src="docs/images/kon-screenshot.png" alt="Kon terminal UI screenshot" width="700" />
 </p>
 
 Kon is a minimal coding agent focused on a tiny core prompt, a small built-in toolset, and project-specific context layered on top only when you want it. The default system prompt stays **under 270 tokens**, and even including the built-in tool descriptions and parameter schemas, the fixed harness stays at about **~1,000 tokens**. The core experience is built around just **6 default tools** plus **2 optional web tools**.

{kon_coding_agent-0.3.8 → kon_coding_agent-0.3.9}/README.md

@@ -1,5 +1,9 @@
-<h1 align="center">Kon</h1>
-<p align="center">A minimal terminal coding agent with a small core, strong defaults, and user-owned context.</p>
+<pre align="center">
+░█░█░█▀█░█▀█
+░█▀▄░█░█░█░█
+░▀░▀░▀▀▀░▀░▀
+</pre>
+<p align="center">Minimal coding agent harness</p>
 <p align="center">
   <a href="https://pypi.org/project/kon-coding-agent/"><img alt="PyPI" src="https://img.shields.io/pypi/v/kon-coding-agent?style=flat-square" /></a>
   <a href="https://www.python.org/downloads/release/python-3120/"><img alt="Python" src="https://img.shields.io/badge/python-3.12%2B-blue?style=flat-square" /></a>
@@ -7,7 +11,7 @@
 </p>
 
 <p align="center">
-  <img src="docs/images/kon-screenshot.png" alt="Kon terminal UI screenshot" width="490" />
+  <img src="docs/images/kon-screenshot.png" alt="Kon terminal UI screenshot" width="700" />
 </p>
 
 Kon is a minimal coding agent focused on a tiny core prompt, a small built-in toolset, and project-specific context layered on top only when you want it. The default system prompt stays **under 270 tokens**, and even including the built-in tool descriptions and parameter schemas, the fixed harness stays at about **~1,000 tokens**. The core experience is built around just **6 default tools** plus **2 optional web tools**.

{kon_coding_agent-0.3.8 → kon_coding_agent-0.3.9}/pyproject.toml

@@ -14,7 +14,7 @@ default = true
 
 [project]
 name = "kon-coding-agent"
-version = "0.3.8"
+version = "0.3.9"
 description = "Minimal coding agent"
 readme = "README.md"
 requires-python = ">=3.12"
@@ -31,6 +31,7 @@ dependencies = [
     "readability-lxml>=0.8.4",
     "rich>=14.3.2",
     "textual>=8.0.0",
+    "textual-image[textual]>=0.12.0",
 ]
 
 [dependency-groups]

{kon_coding_agent-0.3.8 → kon_coding_agent-0.3.9}/src/kon/core/types.py

@@ -66,6 +66,7 @@ class ToolCallDelta(BaseModel):
     type: Literal["tool_call_delta"] = "tool_call_delta"
     index: int  # Correlates with ToolCallStart.index
     arguments_delta: str
+    replace: bool = False
 
 
 class StreamDone(BaseModel):

{kon_coding_agent-0.3.8 → kon_coding_agent-0.3.9}/src/kon/llm/__init__.py

@@ -10,6 +10,7 @@ from .models import (
 from .oauth import clear_credentials as clear_copilot_credentials
 from .oauth import (
     clear_openai_credentials,
+    is_copilot_logged_in,
     is_openai_logged_in,
     load_openai_credentials,
     openai_login,
@@ -18,33 +19,15 @@ from .oauth import get_valid_openai_token as get_openai_token
 from .oauth import get_valid_token as get_copilot_token
 from .oauth import load_credentials as load_copilot_credentials
 from .oauth import login as copilot_login
-from .providers import (
-    API_TYPE_TO_PROVIDER_CLASS,
-    PROVIDER_API_BY_NAME,
-    CopilotAnthropicProvider,
-    CopilotProvider,
-    CopilotResponsesProvider,
-    OpenAICodexResponsesProvider,
-    OpenAICompletionsProvider,
-    OpenAIResponsesProvider,
-    is_copilot_logged_in,
-    resolve_provider_api_type,
-)
+from .providers import PROVIDER_API_BY_NAME, get_provider_class, resolve_provider_api_type
 
 __all__ = [
-    "API_TYPE_TO_PROVIDER_CLASS",
     "DEFAULT_THINKING_LEVELS",
     "PROVIDER_API_BY_NAME",
     "ApiType",
     "BaseProvider",
-    "CopilotAnthropicProvider",
-    "CopilotProvider",
-    "CopilotResponsesProvider",
     "LLMStream",
     "Model",
-    "OpenAICodexResponsesProvider",
-    "OpenAICompletionsProvider",
-    "OpenAIResponsesProvider",
     "ProviderConfig",
     "clear_copilot_credentials",
     "clear_openai_credentials",
@@ -55,6 +38,7 @@ __all__ = [
     "get_model",
     "get_models_by_provider",
     "get_openai_token",
+    "get_provider_class",
     "is_copilot_logged_in",
     "is_openai_logged_in",
     "load_copilot_credentials",

{kon_coding_agent-0.3.8 → kon_coding_agent-0.3.9}/src/kon/llm/oauth/__init__.py

@@ -5,6 +5,7 @@ from .copilot import (
     get_base_url_from_token,
     get_copilot_auth_path,
     get_valid_token,
+    is_copilot_logged_in,
     load_credentials,
     login,
 )
@@ -29,6 +30,7 @@ __all__ = [
     "get_openai_auth_path",
     "get_valid_openai_token",
     "get_valid_token",
+    "is_copilot_logged_in",
     "is_openai_logged_in",
     "load_credentials",
     "load_openai_credentials",

{kon_coding_agent-0.3.8 → kon_coding_agent-0.3.9}/src/kon/llm/oauth/copilot.py

@@ -86,6 +86,10 @@ def clear_credentials() -> None:
         path.unlink()
 
 
+def is_copilot_logged_in() -> bool:
+    return load_credentials() is not None
+
+
 def _get_urls(domain: str) -> dict[str, str]:
     return {
         "device_code": f"https://{domain}/login/device/code",

kon_coding_agent-0.3.9/src/kon/llm/providers/__init__.py

@@ -0,0 +1,61 @@
+from ..base import BaseProvider
+from ..models import ApiType
+
+PROVIDER_API_BY_NAME: dict[str, ApiType] = {
+    "openai": ApiType.OPENAI_COMPLETIONS,
+    "zhipu": ApiType.OPENAI_COMPLETIONS,
+    "deepseek": ApiType.OPENAI_COMPLETIONS,
+    "github-copilot": ApiType.GITHUB_COPILOT,
+    "openai-responses": ApiType.OPENAI_RESPONSES,
+    "openai-codex": ApiType.OPENAI_CODEX_RESPONSES,
+    "azure-ai-foundry": ApiType.AZURE_AI_FOUNDRY,
+}
+
+
+def resolve_provider_api_type(provider: str | None) -> ApiType:
+    if provider is None:
+        return ApiType.OPENAI_COMPLETIONS
+
+    api_type = PROVIDER_API_BY_NAME.get(provider)
+    if api_type is None:
+        valid = ", ".join(sorted(PROVIDER_API_BY_NAME))
+        raise ValueError(f"Unknown provider '{provider}'. Valid providers: {valid}")
+
+    return api_type
+
+
+def get_provider_class(api_type: ApiType) -> type[BaseProvider]:
+    match api_type:
+        case ApiType.GITHUB_COPILOT:
+            from .copilot import CopilotProvider
+
+            return CopilotProvider
+        case ApiType.GITHUB_COPILOT_RESPONSES:
+            from .copilot import CopilotResponsesProvider
+
+            return CopilotResponsesProvider
+        case ApiType.OPENAI_RESPONSES:
+            from .openai_responses import OpenAIResponsesProvider
+
+            return OpenAIResponsesProvider
+        case ApiType.OPENAI_CODEX_RESPONSES:
+            from .openai_codex_responses import OpenAICodexResponsesProvider
+
+            return OpenAICodexResponsesProvider
+        case ApiType.ANTHROPIC_COPILOT:
+            from .copilot_anthropic import CopilotAnthropicProvider
+
+            return CopilotAnthropicProvider
+        case ApiType.AZURE_AI_FOUNDRY:
+            from .azure_ai_foundry import AzureAIFoundryProvider
+
+            return AzureAIFoundryProvider
+        case ApiType.OPENAI_COMPLETIONS:
+            from .openai_completions import OpenAICompletionsProvider
+
+            return OpenAICompletionsProvider
+
+    raise ValueError(f"Unsupported API type: {api_type.value}")
+
+
+__all__ = ["PROVIDER_API_BY_NAME", "get_provider_class", "resolve_provider_api_type"]

{kon_coding_agent-0.3.8 → kon_coding_agent-0.3.9}/src/kon/llm/providers/copilot.py

@@ -142,7 +142,3 @@ class CopilotResponsesProvider(OpenAIResponsesProvider):
             temperature=temperature,
             max_tokens=max_tokens,
         )
-
-
-def is_copilot_logged_in() -> bool:
-    return load_credentials() is not None

{kon_coding_agent-0.3.8 → kon_coding_agent-0.3.9}/src/kon/llm/providers/openai_codex_responses.py

@@ -22,11 +22,12 @@ from ...core.types import (
     ToolDefinition,
     Usage,
 )
-from ..base import BaseProvider, LLMStream, ProviderConfig
+from ..base import BaseProvider, LLMStream
 from ..oauth.openai import get_valid_openai_token, load_openai_credentials
 
 _MAX_RETRIES = 3
 _BASE_DELAY_MS = 1000
+_CONNECT_TIMEOUT_SECONDS = 30
 _OPENAI_BETA_RESPONSES_WEBSOCKETS = "responses_websockets=2026-02-06"
 _WS_FALLBACK_SESSIONS: set[str] = set()
 
@@ -36,7 +37,9 @@ class CodexTransportError(Exception):
 
 
 class CodexNonTransportError(Exception):
-    pass
+    def __init__(self, message: str, *, status: int | None = None) -> None:
+        super().__init__(message)
+        self.status = status
 
 
 def _format_provider_error(error: Exception) -> str:
@@ -54,10 +57,6 @@ class OpenAICodexResponsesProvider(BaseProvider):
     name = "openai-codex"
     thinking_levels: list[str] = ["none", "minimal", "low", "medium", "high", "xhigh"]  # noqa: RUF012
 
-    def __init__(self, config: ProviderConfig):
-        super().__init__(config)
-        self._websocket_fallback = False
-
     async def _stream_impl(
         self,
         messages: list[Message],
@@ -207,9 +206,7 @@ class OpenAICodexResponsesProvider(BaseProvider):
         body = self._build_request_body(messages, system_prompt, tools, temperature)
         session_id = self.config.session_id
 
-        if session_id in _WS_FALLBACK_SESSIONS:
-            self._websocket_fallback = True
-        else:
+        if session_id not in _WS_FALLBACK_SESSIONS:
             emitted = False
             try:
                 websocket_events = self._stream_websocket_events(
@@ -222,11 +219,10 @@ class OpenAICodexResponsesProvider(BaseProvider):
             except CodexNonTransportError as e:
                 yield StreamError(error=_format_provider_error(e))
                 return
-            except Exception as e:
+            except (CodexTransportError, aiohttp.ClientError, OSError) as e:
                 if emitted:
                     yield StreamError(error=_format_provider_error(e))
                     return
-                self._websocket_fallback = True
                 if session_id:
                     _WS_FALLBACK_SESSIONS.add(session_id)
 
@@ -241,9 +237,30 @@ class OpenAICodexResponsesProvider(BaseProvider):
         self, events: AsyncIterator[dict[str, Any]], llm_stream: LLMStream
     ) -> AsyncIterator[StreamPart]:
         current_tool_calls: dict[str, dict[str, Any]] = {}
+        call_key_by_item_id: dict[str, str] = {}
         last_tool_call_id: str | None = None
         tool_call_index = 0
 
+        def _resolve_key(item_id: Any) -> str | None:
+            if isinstance(item_id, str) and item_id:
+                return call_key_by_item_id.get(item_id)
+            return last_tool_call_id
+
+        def _reconcile(call_data: dict[str, Any], final_args: str) -> ToolCallDelta | None:
+            current_args = call_data["arguments"]
+            if final_args.startswith(current_args):
+                missing = final_args[len(current_args) :]
+                if not missing:
+                    return None
+                call_data["arguments"] += missing
+                return ToolCallDelta(index=call_data["index"], arguments_delta=missing)
+            if final_args == current_args:
+                return None
+            call_data["arguments"] = final_args
+            return ToolCallDelta(
+                index=call_data["index"], arguments_delta=final_args, replace=True
+            )
+
         async for event in events:
             event_type = event.get("type")
             if not isinstance(event_type, str):
@@ -267,32 +284,65 @@ class OpenAICodexResponsesProvider(BaseProvider):
                     name = item.get("name")
                     if isinstance(call_id, str) and isinstance(name, str):
                         full_id = f"{call_id}|{item_id}" if isinstance(item_id, str) else call_id
+                        initial_args = item.get("arguments")
+                        initial_args_text = initial_args if isinstance(initial_args, str) else ""
                         current_tool_calls[full_id] = {
-                            "id": full_id,
-                            "name": name,
-                            "arguments": "",
+                            "arguments": initial_args_text,
                             "index": tool_call_index,
                         }
+                        if isinstance(item_id, str) and item_id:
+                            call_key_by_item_id[item_id] = full_id
                         last_tool_call_id = full_id
                         yield ToolCallStart(index=tool_call_index, id=full_id, name=name)
+                        if initial_args_text:
+                            yield ToolCallDelta(
+                                index=tool_call_index, arguments_delta=initial_args_text
+                            )
                         tool_call_index += 1
 
             elif event_type == "response.function_call_arguments.delta":
                 delta = event.get("delta")
-                if (
-                    isinstance(delta, str)
-                    and last_tool_call_id
-                    and last_tool_call_id in current_tool_calls
-                ):
-                    current_tool_calls[last_tool_call_id]["arguments"] += delta
-                    idx = int(current_tool_calls[last_tool_call_id]["index"])
-                    yield ToolCallDelta(index=idx, arguments_delta=delta)
+                if not isinstance(delta, str):
+                    continue
+                call_key = _resolve_key(event.get("item_id"))
+                if not call_key or call_key not in current_tool_calls:
+                    continue
+                call_data = current_tool_calls[call_key]
+                call_data["arguments"] += delta
+                yield ToolCallDelta(index=call_data["index"], arguments_delta=delta)
+
+            elif event_type == "response.function_call_arguments.done":
+                final_args = event.get("arguments")
+                if not isinstance(final_args, str):
+                    continue
+                call_key = _resolve_key(event.get("item_id"))
+                if not call_key or call_key not in current_tool_calls:
+                    continue
+                call_data = current_tool_calls[call_key]
+                delta_part = _reconcile(call_data, final_args)
+                if delta_part is not None:
+                    yield delta_part
+
+            elif event_type == "response.output_item.done":
+                item = event.get("item")
+                if not isinstance(item, dict) or item.get("type") != "function_call":
+                    continue
+                final_args = item.get("arguments")
+                if not isinstance(final_args, str):
+                    continue
+                call_key = _resolve_key(item.get("id"))
+                if not call_key or call_key not in current_tool_calls:
+                    continue
+                call_data = current_tool_calls[call_key]
+                delta_part = _reconcile(call_data, final_args)
+                if delta_part is not None:
+                    yield delta_part
 
             elif event_type in {"response.completed", "response.done", "response.incomplete"}:
                 response_obj = event.get("response")
                 if isinstance(response_obj, dict):
                     self._apply_response_metadata(response_obj, llm_stream)
-                    stop_reason = self._map_stop_reason(response_obj.get("status"))
+                    stop_reason = self._map_stop_reason(response_obj)
                     if current_tool_calls and stop_reason == StopReason.STOP:
                         stop_reason = StopReason.TOOL_USE
                     yield StreamDone(stop_reason=stop_reason)
@@ -324,14 +374,14 @@ class OpenAICodexResponsesProvider(BaseProvider):
         if isinstance(usage, dict):
             input_details = usage.get("input_tokens_details")
             cached = 0
-            cache_write = int(usage.get("cache_write_tokens") or 0)
+            cache_write_value = usage.get("cache_write_tokens")
             if isinstance(input_details, dict):
                 cached = int(input_details.get("cached_tokens") or 0)
-                cache_write = int(
-                    input_details.get("cache_write_tokens")
-                    or input_details.get("cache_creation_tokens")
-                    or cache_write
-                )
+                if "cache_write_tokens" in input_details:
+                    cache_write_value = input_details["cache_write_tokens"]
+                elif "cache_creation_tokens" in input_details:
+                    cache_write_value = input_details["cache_creation_tokens"]
+            cache_write = int(cache_write_value) if cache_write_value is not None else 0
             input_tokens = int(usage.get("input_tokens") or 0)
             non_cached_input = max(input_tokens - cached, 0)
             llm_stream._usage = Usage(
@@ -348,7 +398,9 @@ class OpenAICodexResponsesProvider(BaseProvider):
         self, body: dict[str, Any], headers: dict[str, str]
     ) -> AsyncIterator[dict[str, Any]]:
         last_error: str | None = None
-        timeout = aiohttp.ClientTimeout(total=kon_config.llm.request_timeout_seconds)
+        timeout = aiohttp.ClientTimeout(
+            sock_connect=_CONNECT_TIMEOUT_SECONDS, sock_read=kon_config.llm.request_timeout_seconds
+        )
         async with aiohttp.ClientSession(timeout=timeout) as session:
             response: aiohttp.ClientResponse | None = None
             for attempt in range(_MAX_RETRIES + 1):
@@ -361,7 +413,7 @@ class OpenAICodexResponsesProvider(BaseProvider):
                     delay = _BASE_DELAY_MS * (2**attempt) / 1000
                     await asyncio.sleep(delay)
                     continue
-                raise CodexNonTransportError(last_error)
+                raise CodexNonTransportError(last_error, status=response.status)
 
             if response is None or response.status >= 400:
                 raise CodexNonTransportError(last_error or "Codex request failed after retries")
@@ -372,10 +424,15 @@ class OpenAICodexResponsesProvider(BaseProvider):
     async def _stream_websocket_events(
         self, body: dict[str, Any], headers: dict[str, str]
     ) -> AsyncIterator[dict[str, Any]]:
-        timeout = aiohttp.ClientTimeout(total=kon_config.llm.request_timeout_seconds)
+        timeout = aiohttp.ClientTimeout(
+            sock_connect=_CONNECT_TIMEOUT_SECONDS, sock_read=kon_config.llm.request_timeout_seconds
+        )
+        ws_timeout = aiohttp.ClientWSTimeout(ws_receive=kon_config.llm.request_timeout_seconds)  # type: ignore[call-arg]
         async with (
             aiohttp.ClientSession(timeout=timeout) as session,
-            session.ws_connect(self._resolve_websocket_url(), headers=headers) as ws,
+            session.ws_connect(
+                self._resolve_websocket_url(), headers=headers, heartbeat=20, timeout=ws_timeout
+            ) as ws,
         ):
             await ws.send_json({"type": "response.create", **body})
             saw_completion = False
@@ -405,12 +462,6 @@ class OpenAICodexResponsesProvider(BaseProvider):
                 elif msg.type == aiohttp.WSMsgType.ERROR:
                     error = ws.exception()
                     raise CodexTransportError(str(error) if error else "WebSocket error")
-                elif msg.type in {
-                    aiohttp.WSMsgType.CLOSE,
-                    aiohttp.WSMsgType.CLOSED,
-                    aiohttp.WSMsgType.CLOSING,
-                }:
-                    break
 
             if not saw_completion:
                 raise CodexTransportError("WebSocket stream closed before response.completed")
@@ -436,19 +487,23 @@ class OpenAICodexResponsesProvider(BaseProvider):
                 if isinstance(parsed, dict):
                     yield parsed
 
-    def _map_stop_reason(self, status: Any) -> StopReason:
+    def _map_stop_reason(self, response_obj: dict[str, Any]) -> StopReason:
+        status = response_obj.get("status")
         if status == "completed":
             return StopReason.STOP
         if status == "incomplete":
+            details = response_obj.get("incomplete_details")
+            reason = details.get("reason") if isinstance(details, dict) else None
+            if reason == "content_filter":
+                return StopReason.ERROR
             return StopReason.LENGTH
         if status in {"failed", "cancelled"}:
             return StopReason.ERROR
         return StopReason.STOP
 
     def should_retry_for_error(self, error: Exception) -> bool:
-        msg = str(error).lower()
-        return any(kw in msg for kw in ("429", "rate_limit", "server_error", "502", "503", "504"))
-
-
-def is_openai_logged_in() -> bool:
-    return load_openai_credentials() is not None
+        if isinstance(error, CodexTransportError):
+            return True
+        if isinstance(error, CodexNonTransportError) and error.status is not None:
+            return _is_retryable_status(error.status)
+        return False

{kon_coding_agent-0.3.8 → kon_coding_agent-0.3.9}/src/kon/loop.py

@@ -245,7 +245,7 @@ class Agent:
         # Get the latest assistant message that has usage.
         # The most recent assistant entry can be interrupted/error and have no usage.
         last_usage: Usage | None = None
-        for entry in reversed(self.session.entries):
+        for entry in reversed(self.session.active_entries):
             if isinstance(entry, MessageEntry) and isinstance(entry.message, AssistantMessage):
                 usage = entry.message.usage
                 if usage is None: