Patch Changes

• b577d68 Thanks @liruifengv! - Add first-class OpenAI Codex OAuth support and provider runtime integration.

  Highlights

  • Add a new openai-codex-responses provider implementation for Codex-compatible Responses streaming.
  • Add provider auth infrastructure in main process:
    • secure local session storage
    • auth strategy abstraction
    • provider auth service with refresh and status push events
  • Expose provider auth IPC APIs and preload bridge methods for connect/disconnect/status checks.
  • Add settings UI updates for OAuth-based providers, including connect/disconnect flows and status display.
  • Extend shared schemas/types/constants to model provider auth configuration and status.
  • Add/expand tests for Codex provider streaming and auth service behavior.

• 81e9ef8 Thanks @liruifengv! - Add a built-in AskUserQuestion tool for blocking agent clarifications during a run.

  Highlights

  • Add the new AskUserQuestion tool so the agent can pause for one user answer and then continue the same run automatically.
  • Add main-process question request lifecycle handling, including in-memory pending state, IPC handlers, push events, and abort cleanup.
  • Add renderer question UI and session-scoped question state so users can answer or dismiss requests without sending a normal chat message.
  • Show AskUserQuestion progress and resolved answers in tool execution history.
  • Update prompts, shared types, and tests to support the new question flow end to end.

Patch Changes

• 9dab963 Thanks @liruifengv! - Add tool approval modes and in-chat permission requests.

  • add ask, auto-edit, and yolo approval modes for sessions, plus a global default approval mode in settings
  • gate tool execution through the main-process approval flow and surface pending approvals in the chat input area
  • show tool-specific permission details, including diff previews for Write and Edit requests

• 33bfeb5 Thanks @liruifengv! - Add unit test coverage for the built-in tool stack.

  • cover ToolRegistry registration, validation, and error handling paths
  • add file-system tool tests for Read, Write, Edit, and Glob
  • add mocked tests for Grep, WebFetch, and WebSearch behaviors

• a3914b1 Thanks @liruifengv! - Add unit tests for main-process execution, persistence, and state bridging.

  • cover Bash tool execution paths including abort, timeout, exit codes, and truncated output
  • add tests for Persistence, SessionStore, EventAdapter, and push bridging behavior
  • add ConfigStore tests for caching, deep merges, atomic writes, and API key resolution

Patch Changes

• 2c787ba Thanks @liruifengv! - Tighten chat message layout so assistant content stays within the message column.

  • constrain assistant turns and tool groups to the chat content width
  • prevent long tool summaries and markdown content from stretching the message area

• 487b55f Thanks @liruifengv! - Fix thinking blocks so completed reasoning collapses automatically in chat.

  • treat only the actively streaming last thinking block as expanded
  • collapse completed thinking content automatically after streaming ends

• ac2df1c Thanks @liruifengv! - Fix token usage accounting and display for assistant responses.

  • Align OpenAI Completions and Gemini token mapping with the current app behavior
  • Fix assistant turn token usage display so merged responses show the intended request usage
  • Fix context window usage to avoid flashing to zero while streaming and use total context occupancy when available

Patch Changes

• 552317a Thanks @liruifengv! - Fix macOS window lifecycle so reopening from the Dock no longer reinitializes the app or falls back to onboarding after the main window is closed.

  • Keep IPC handlers registered until the app actually quits, so reopening the app window on macOS does not break settings.get, session.list, or push event delivery.
  • Hide the main window on macOS when the close button is clicked instead of destroying it, so reopening from the Dock restores the existing window state without a full renderer reload.

  修复 macOS 窗口生命周期问题，避免用户关闭主窗口后从 Dock 重新打开时应用被重新初始化，或错误回到 onboarding 页面。

  • 将 IPC handler 的清理时机延后到应用真正退出时，避免在 macOS 重新打开窗口后 settings.get、session.list 和 push 事件失效。
  • 在 macOS 下点击关闭按钮时隐藏主窗口而不是销毁窗口，这样从 Dock 重新打开时会恢复已有窗口状态，而不是重新加载 renderer。

Minor Changes

• #13 efbb9ac Thanks @liruifengv! - Refactor Amon into a provider-agnostic agent platform and ship a major workflow upgrade across runtime, skills, and UI.

  BREAKING CHANGES

  • Removed the Claude Agent SDK and replaced it with Amon's own agent core and provider-agnostic runtime.

  • The settings and provider configuration schema changed substantially. Existing configs are not fully compatible with the new architecture: legacy fields are migrated on a best-effort basis, but deprecated fields and older provider-specific options are no longer supported and may require manual reconfiguration.

  • Introduce a new three-layer architecture with a standalone AI streaming layer, a framework-agnostic agent core, and Electron-specific integration for session state, IPC, and push events.

  • Add built-in support for Anthropic, OpenAI Completions, OpenAI Responses, and Google Gemini providers, with richer thinking-level controls, extra provider parameters, and web search support.

  • Expand the local agent workflow with bundled tools for bash, file read/write/edit, glob, grep, web fetch, and web search, plus project bootstrap context from AGENTS.md, SOUL.md, and BOOTSTRAP.md.

  • Overhaul the skills system with built-in skill packaging, skill management UI, recommended skills, and improved loading from bundled, user, and workspace sources.

  • Refresh the chat and settings experience with grouped assistant turns, context usage indicators, improved permission and confirmation flows, better provider configuration, and safer input handling for IME users.

  • Add broader automated test coverage and update the documentation to match the new architecture and product capabilities.

  将 Amon 重构为一个与 Provider 无关的 Agent 平台，并在运行时、Skills 和 UI 层面带来一轮重要的工作流升级。

  重大变更

  • 移除了 Claude Agent SDK，改为使用 Amon 自己实现的 Agent 核心和与 Provider 无关的运行时。

  • 设置和 Provider 配置结构发生了较大变化。现有配置与新架构并非完全兼容：旧字段会尽力迁移，但已废弃的字段以及旧的 provider 特定配置项已不再支持，部分场景可能需要手动重新配置。

  • 引入新的三层架构，包括独立的 AI 流式处理层、与框架无关的 Agent 核心，以及负责会话状态、IPC 和推送事件的 Electron 集成层。

  • 新增对 Anthropic、OpenAI Completions、OpenAI Responses 和 Google Gemini 的内置支持，并提供更丰富的 thinking level 控制、额外 provider 参数能力，以及 web search 支持。

  • 扩展本地 Agent 工作流，内置 bash、文件读写编辑、glob、grep、web fetch 和 web search 等工具，同时支持从 AGENTS.md、SOUL.md 和 BOOTSTRAP.md 加载项目启动上下文。

  • 重构 Skills 系统，提供内置 skill 打包能力、skill 管理界面、推荐 skills，并改进从内置、用户目录和工作区来源加载 skills 的机制。

  • 升级聊天和设置体验，包括分组展示 assistant turns、上下文使用量指示器、更完善的权限与确认流程、更好的 provider 配置体验，以及针对输入法用户更安全的输入处理。

  • 补充更全面的自动化测试覆盖，并同步更新文档，使其与新的架构和产品能力保持一致。

Patch Changes

• e97da9f Thanks @liruifengv! - Fix external image rendering in markdown and enable proper file download

  • Updated CSP img-src to allow https: and http: sources so external markdown images render correctly
  • Updated CSP connect-src to allow external fetch requests for image downloads
  • Added CORS bypass via session.webRequest.onHeadersReceived to prevent cross-origin fetch failures
  • Added will-download handler to show a save dialog instead of silently downloading or opening a new window

• e98f3ed Thanks @liruifengv! - Open external links in system default browser instead of Electron window

  • Added setWindowOpenHandler to intercept window.open() and redirect http/https links to the system browser via shell.openExternal
  • Disabled Streamdown's built-in link safety modal (linkSafety: { enabled: false }) since links are now handled by the OS

Patch Changes

• #11 c2fa99d Thanks @liruifengv! - feat: add i18n internationalization support for Chinese and English

  • Add i18next + react-i18next for renderer process and main process
  • Create translation files for 9 namespaces (common, chat, message, settings, sidebar, permission, onboarding, menu, validation)
  • Add language selector in General Settings (English / 中文), default to English
  • Replace all hardcoded UI strings (~200) across ~30 component files with t() calls
  • Internationalize Electron menus and native dialogs
  • Real-time language switching without app restart, synced across all windows

Patch Changes

• c240587 Thanks @liruifengv! - 更新供应商模型配置：

  • GLM: 新增 glm-5 模型
  • MiniMax: 新增 MiniMax-M2.5 模型
  • Kimi: 新增 Kimi 供应商，API 地址 https://api.kimi.com/coding，模型 kimi-for-coding
  • Claude 官方/自定义/OpenRouter: 新增 claude-opus-4-6 / anthropic/claude-opus-4.6 模型

0.2.0

Minor Changes

• 2829cd3 Thanks @liruifengv! - 支持使用 @ 提及文件

• eb5b5ae Thanks @liruifengv! - 增强计划审批和计划预览 UI/UX

Patch Changes

• 352b7d4 Thanks @liruifengv! - 提升 Write / Edit 工具的 UI

• #8 212146f Thanks @liruifengv! - 修复输入框 bug

• #7 1ae6af2 Thanks @liruifengv! - 使用状态机重构 SDK 流式消息系统

Patch Changes

• 1930e72 Thanks @liruifengv! - Add some preset provider

• 457fc81 Thanks @liruifengv! - Add tip for unsaved settings change