kon-coding-agent 0.3.3__tar.gz → 0.3.5__tar.gz

kon_coding_agent-0.3.5/.github/workflows/test.yml

@@ -0,0 +1,33 @@
+name: Test
+
+on: [push, pull_request]
+
+permissions:
+  contents: read
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.12","3.13", "3.14"]
+    steps:
+    - uses: actions/checkout@v3
+    - name: Set up Python ${{ matrix.python-version }}
+      uses: actions/setup-python@v4
+      with:
+        python-version: ${{ matrix.python-version }}
+        cache: pip
+        cache-dependency-path: pyproject.toml
+    - name: Cache models
+      uses: actions/cache@v3
+      with:
+        path: ~/.cache
+        key: ${{ runner.os }}-torch-
+    - name: Install dependencies
+      run: |
+        pip install . --group dev
+    - name: Run tests
+      run: |
+        python -m pytest -s
+

{kon_coding_agent-0.3.3 → kon_coding_agent-0.3.5}/.kon/skills/kon-release-publish/SKILL.md

@@ -33,38 +33,44 @@ Use this skill when the user asks to cut a new Kon version, tag it, publish to P
    - `git status --short --branch` must be clean (or confirm with user)
    - `git tag --list` and `git log --oneline <prev_tag>..HEAD` to summarize changes
 
-2. **Version bump**
+2. **Update CHANGELOG.md**
+   - Replace the `## [Unreleased]` section's `- No changes yet.` with a new versioned heading: `## <version> - YYYY-MM-DD`
+   - Use `git log --oneline <prev_tag>..HEAD` to categorize changes into `### Added`, `### Changed`, `### Fixed` sections
+   - Credit external contributors with `- @username`
+   - Commit message: `docs: update changelog for <version>`
+
+3. **Version bump**
    - Update version in all 3 files above
 
-3. **Quality gates**
+4. **Quality gates**
    - `uv run ruff format .`
    - `uv run ruff check .`
    - `uv run pyright .`
    - `uv run pytest`
 
-4. **Commit**
+5. **Commit**
    - Commit message: `build: bump version to <version>`
 
-5. **Tag**
+6. **Tag**
    - Annotated tag: `git tag -a v<version> -m "v<version> ..."`
-   - Include concise “changes since previous tag” bullets
+   - Include concise "changes since previous tag" bullets
 
-6. **Push**
+7. **Push**
    - `git push origin main`
    - `git push origin v<version>`
 
-7. **Build + verify artifacts**
+8. **Build + verify artifacts**
    - `rm -rf dist && uv build`
    - `uv run python -m twine check dist/*`
 
-8. **Publish to PyPI**
+9. **Publish to PyPI**
    - Prefer token file if present (example `~/.pypi-token`):
    - `TWINE_USERNAME=__token__ TWINE_PASSWORD="$(< ~/.pypi-token)" uv run python -m twine upload dist/*`
    - Verify:
      - `https://pypi.org/project/kon-coding-agent/<version>/`
      - `https://pypi.org/pypi/kon-coding-agent/json` reports latest version
 
-9. **Create GitHub release**
+10. **Create GitHub release**
    - If token exists at `~/.github-token`, call Releases API:
    - `POST /repos/<owner>/<repo>/releases` with:
      - `tag_name: v<version>`
@@ -83,6 +89,7 @@ Use this skill when the user asks to cut a new Kon version, tag it, publish to P
 
 ## Output checklist to report
 
+- Changelog updated for `<version>`
 - Version bumped in all files
 - Checks passed
 - Commit hash

{kon_coding_agent-0.3.3 → kon_coding_agent-0.3.5}/AGENTS.md

@@ -4,6 +4,7 @@
 
 - Don't add trivial docstrings. Only add docstrings when explaining complex functionality.
 - This project uses `uv`. Run `uv run ruff format .` after editing or creating any files.
+- If generating and running a Python script, use `uv run python` instead of `python`.
 
 ## Testing
 

{kon_coding_agent-0.3.3 → kon_coding_agent-0.3.5}/CHANGELOG.md

@@ -6,6 +6,70 @@ All notable changes to this project will be documented in this file.
 
 - No changes yet.
 
+## 0.3.5 - 2026-04-18
+
+### Added
+
+- Standalone session HTML export with self-contained styling.
+- Configurable request timeout for API calls - @jspruit.
+- GitHub CI tests for Python 3.12–3.13 - @sukhbinder.
+- Width-aware popup lists and queue display.
+
+### Changed
+
+- Highlight color applied to the second column in floating lists.
+- Batch scroll during streaming, cache query_one lookups, pause spinner timer when idle.
+- Diff line length capped at 200 characters.
+
+### Fixed
+
+- Persist thinking level in session header and change all defaults to high.
+- Normalize OpenAI provider imports.
+- Widen resume popup labels.
+- Remove unused UI app import.
+
+### Performance
+
+- Added `gc.freeze()` and `PAUSE_GC_ON_SCROLL` to reduce GC stutters.
+
+## 0.3.4 - 2026-04-10
+
+### Fixed
+
+- Fixed Windows UTF-8 encoding errors in file operations - @sukhbinder.
+- Fixed local Gemma model thinking block compatibility.
+- Removed duplicate force-include for builtin skills in build config.
+
+### Changed
+
+- Updated local model documentation.
+
+## 0.3.3 - 2026-04-08
+
+### Added
+
+- Added queued agent steering between turns - @0xku.
+- Added bundled `/init` slash command for project scaffolding - @0xku.
+- Added help info for queue and steer queue commands.
+- Added GLM-5.1 support for zai provider.
+
+### Changed
+
+- Improved read tool directory listings.
+- Updated README with steer queue documentation.
+- Added `$` icon for bash, `%` for web tools, and `←` for edit tool.
+- Used muted color for shortcut key hints in exit/delete prompts.
+
+### Fixed
+
+- Let ESC interrupt retry backoff immediately - @0xku.
+- Fixed OpenAI login stdin leak by removing orphaned thread - @Meltedd.
+- Fixed OpenAI and Anthropic local compat with auth flags.
+- Fixed interrupt handling before handoff thread switch.
+- Fixed subprocess communication drain on cancellation/timeout.
+- Added zipfile path traversal validation.
+- Removed token throughput metrics.
+
 ## 0.3.2 - 2026-03-22
 
 ### Added

{kon_coding_agent-0.3.3 → kon_coding_agent-0.3.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: kon-coding-agent
-Version: 0.3.3
+Version: 0.3.5
 Summary: Minimal coding agent
 License-File: LICENSE
 Requires-Python: >=3.12

kon_coding_agent-0.3.5/docs/local-models.md

@@ -0,0 +1,80 @@
+# Local Models
+
+This document provides detailed information about running and configuring local models with Kon.
+
+## Tested Models
+
+> Tested on llama server build b8740
+
+| Model | Quantization | Context Length | TPS | System Specs |
+| ----- | -------------- | -------------- | --- | ------------ |
+| `zai-org/glm-4.7-flash` | Q4_K_M | 65,536 | N/A | i7-14700F × 28, 64GB RAM, 24GB VRAM (RTX 3090) |
+| `unsloth/Qwen3.5-27B-GGUF` | Q4_K_M | 32,768 | ~30 | i7-14700F × 28, 64GB RAM, 24GB VRAM (RTX 3090) |
+| `unsloth/gemma-4-26B-A4B-it-GGUF` | UD-Q4_K_M | 32,768 | ~100 | i7-14700F × 28, 64GB RAM, 24GB VRAM (RTX 3090) |
+
+Run Qwen3.5 27B on an RTX 3090 with a 32k context window using llama-server:
+
+```bash
+/path-to-llama-server/llama-server \
+  --model /path-to-model/Qwen3.5-27B-Q4_K_M.gguf \
+  --port 5000 \
+  --ctx-size 32768 \
+  --gpu-layers all \
+  --threads 8 \
+  --threads-batch 8 \
+  --batch-size 1024 \
+  --ubatch-size 512 \
+  --flash-attn on
+```
+
+On this machine, that setup generates at roughly 30 tokens per second.
+
+Then start Kon for a one-off local session:
+
+```bash
+kon --model unsloth/Qwen3.5-27B-GGUF --provider openai \
+  --base-url http://localhost:5000/v1 \
+  --openai-compat-auth none
+```
+
+Run Gemma 4 26B A4B on the same machine using llama-server:
+
+```bash
+/path-to-llama-server/llama-server \
+  --model /path-to-model/gemma-4-26B-A4B-it-UD-Q4_K_M.gguf \
+  --port 5000 \
+  --ctx-size 32768 \
+  --gpu-layers all \
+  --threads 8 \
+  --threads-batch 8 \
+  --batch-size 1024 \
+  --ubatch-size 512 \
+  --flash-attn on \
+  --temperature 1.5
+```
+
+Then start Kon against that local server:
+
+```bash
+kon --model unsloth/gemma-4-26B-A4B-it-GGUF --provider openai \
+  --base-url http://localhost:5000/v1 \
+  --openai-compat-auth none
+```
+
+To avoid passing provider, model, and auth flags every time you start Kon, you can define your local setup in `~/.kon/config.toml`. This also allows you to tune compaction to trigger at a specific point relative to your model's context window.
+
+If this is your default setup, put it in `~/.kon/config.toml` instead:
+
+```toml
+[llm]
+default_provider = "openai"
+default_model = "unsloth/gemma-4-26B-A4B-it-GGUF"
+default_base_url = "http://localhost:5000/v1"
+
+[llm.auth]
+openai_compat = "none" # or "auto"
+
+[compaction]
+# Set this close to your model's context size (e.g., 30000 for a 32k window)
+buffer_tokens = 27768 # 32768 - 5000 (safety margin)
+```

{kon_coding_agent-0.3.3 → kon_coding_agent-0.3.5}/pyproject.toml

@@ -14,7 +14,7 @@ default = true
 
 [project]
 name = "kon-coding-agent"
-version = "0.3.3"
+version = "0.3.5"
 description = "Minimal coding agent"
 readme = "README.md"
 requires-python = ">=3.12"

{kon_coding_agent-0.3.3 → kon_coding_agent-0.3.5}/src/kon/config.py

@@ -76,6 +76,7 @@ class LLMConfig(BaseModel):
     default_thinking_level: str
     system_prompt: SystemPromptConfig
     tool_call_idle_timeout_seconds: float = 180
+    request_timeout_seconds: float = 600
     auth: AuthConfig = AuthConfig()
 
 

{kon_coding_agent-0.3.3 → kon_coding_agent-0.3.5}/src/kon/defaults/config.toml

@@ -11,6 +11,9 @@ default_thinking_level = "high"
 # Abort a tool call if it stays idle for this long during a turn.
 # Helps prevent stalled tool executions from hanging the agent loop.
 tool_call_idle_timeout_seconds = 180
+# HTTP request timeout for LLM API calls (in seconds).
+# Local models (e.g. llama.cpp) may need a higher value for long compaction requests.
+request_timeout_seconds = 600
 
 [llm.auth]
 # Auth policy for OpenAI-compatible and Anthropic-compatible endpoints.

{kon_coding_agent-0.3.3 → kon_coding_agent-0.3.5}/src/kon/llm/base.py

@@ -77,7 +77,7 @@ class ProviderConfig:
     model: str = ""
     max_tokens: int = 8192
     temperature: float | None = None
-    thinking_level: str = "medium"
+    thinking_level: str = "high"
     provider: str | None = None
     session_id: str | None = None
     openai_compat_auth_mode: AuthMode = "auto"

{kon_coding_agent-0.3.3 → kon_coding_agent-0.3.5}/src/kon/llm/providers/openai_completions.py

@@ -10,6 +10,8 @@ from openai.types.chat import (
     ChatCompletionToolParam,
 )
 
+from kon import config as kon_config
+
 from ...core.types import (
     AssistantMessage,
     ImageContent,
@@ -30,7 +32,7 @@ from ...core.types import (
     Usage,
     UserMessage,
 )
-from ..base import BaseProvider, LLMStream, ProviderConfig, resolve_api_key
+from ..base import BaseProvider, LLMStream, ProviderConfig, is_local_base_url, resolve_api_key
 from .openai_compat import supports_developer_role
 from .sanitize import sanitize_surrogates
 
@@ -41,12 +43,13 @@ class OpenAICompletionsCompat:
     supports_developer_role: bool = True
     supports_reasoning_effort: bool = True
     max_tokens_field: Literal["max_tokens", "max_completion_tokens"] = "max_completion_tokens"
-    thinking_format: Literal["openai", "zai", "qwen"] = "openai"
+    thinking_format: Literal["openai", "zai", "qwen", "llama_gemma"] = "openai"
 
 
-def _detect_compat(provider: str, base_url: str) -> OpenAICompletionsCompat:
+def _detect_compat(provider: str, base_url: str, model: str = "") -> OpenAICompletionsCompat:
     normalized_provider = provider.lower()
     normalized_base_url = base_url.lower()
+    normalized_model = model.lower()
     is_zai = (
         normalized_provider == "zai"
         or normalized_provider == "zhipu"
@@ -61,6 +64,13 @@ def _detect_compat(provider: str, base_url: str) -> OpenAICompletionsCompat:
             thinking_format="zai",
         )
 
+    if is_local_base_url(base_url) and "gemma" in normalized_model:
+        return OpenAICompletionsCompat(
+            supports_developer_role=supports_developer_role(provider, base_url),
+            supports_reasoning_effort=False,
+            thinking_format="llama_gemma",
+        )
+
     return OpenAICompletionsCompat(
         supports_developer_role=supports_developer_role(provider, base_url)
     )
@@ -91,8 +101,14 @@ class OpenAICompletionsProvider(BaseProvider):
                 "Set OPENAI_API_KEY or ZAI_API_KEY environment variable, "
                 'or configure llm.auth.openai_compat = "auto"/"none" for local endpoints.'
             )
-        self._client = AsyncOpenAI(api_key=api_key, base_url=config.base_url)
-        self._compat = _detect_compat(config.provider or "", config.base_url or "")
+        self._client = AsyncOpenAI(
+            api_key=api_key,
+            base_url=config.base_url,
+            timeout=kon_config.llm.request_timeout_seconds,
+        )
+        self._compat = _detect_compat(
+            config.provider or "", config.base_url or "", config.model or ""
+        )
 
     async def _stream_impl(
         self,
@@ -138,7 +154,7 @@ class OpenAICompletionsProvider(BaseProvider):
         if compat.thinking_format == "zai":
             if thinking_level and thinking_level != "none":
                 extra_body["thinking"] = {"type": "enabled"}
-        elif compat.thinking_format == "qwen":
+        elif compat.thinking_format in {"qwen", "llama_gemma"}:
             extra_body["enable_thinking"] = bool(thinking_level and thinking_level != "none")
         elif (
             self.supports_reasoning_effort
@@ -238,11 +254,16 @@ class OpenAICompletionsProvider(BaseProvider):
 
         if system_prompt:
             role = "developer" if (compat and compat.supports_developer_role) else "system"
+            prompt_content = sanitize_surrogates(system_prompt)
+            if (
+                compat
+                and compat.thinking_format == "llama_gemma"
+                and self.config.thinking_level != "none"
+                and not prompt_content.startswith("<|think|>")
+            ):
+                prompt_content = "<|think|>" + prompt_content
             result.append(
-                cast(
-                    ChatCompletionMessageParam,
-                    {"role": role, "content": sanitize_surrogates(system_prompt)},
-                )
+                cast(ChatCompletionMessageParam, {"role": role, "content": prompt_content})
             )
 
         pending_images: list[ImageContent] = []

{kon_coding_agent-0.3.3 → kon_coding_agent-0.3.5}/src/kon/llm/providers/openai_responses.py

@@ -4,6 +4,8 @@ from typing import Any
 
 from openai import APIStatusError, AsyncOpenAI, RateLimitError
 
+from kon import config as kon_config
+
 from ...core.types import (
     AssistantMessage,
     ImageContent,
@@ -51,6 +53,7 @@ class OpenAIResponsesProvider(BaseProvider):
                 api_key=self.config.api_key,
                 base_url=self.config.base_url,
                 default_headers=self._headers,
+                timeout=kon_config.llm.request_timeout_seconds,
             )
         return self._client