release: 0.5.7

.vscode/extensions.json

@@ -1,3 +1,3 @@
 {
-  "recommendations": ["dbaeumer.vscode-eslint", "lokalise.i18n-ally", "esbenp.prettier-vscode"]
+  "recommendations": ["dbaeumer.vscode-eslint", "lokalise.i18n-ally", "esbenp.prettier-vscode", "TypeScriptTeam.native-preview"]
 }

.vscode/settings.json

@@ -5,5 +5,7 @@
   "i18n-ally.keystyle": "nested",
   "i18n-ally.sourceLanguage": "zh-CN",
   "i18n-ally.namespace": true,
-  "i18n-ally.pathMatcher": "{locale}/{namespaces}.json"
+  "i18n-ally.pathMatcher": "{locale}/{namespaces}.json",
+  "unocss.disable": true,
+  "typescript.experimental.useTsgo": true
 }

docs/architecture/tool-system.md

@@ -29,7 +29,7 @@ graph TB
 
         AgentToolMgr[AgentToolManager]
         FsHandler[AgentFileSystemHandler]
-        Browser[Yo Browser Tools]
+        YoBrowser[Yo Browser CDP]
     end
 
     subgraph "外部服务"
@@ -48,10 +48,10 @@ graph TB
     McpClient --> MCPServers
 
     AgentToolMgr --> FsHandler
-    AgentToolMgr --> Browser
+    AgentToolMgr --> YoBrowser
 
     FsHandler --> Files
-    Browser --> Web
+    YoBrowser --> Web
 
     classDef router fill:#e3f2fd
     classDef mcp fill:#fff3e0
@@ -60,7 +60,7 @@ graph TB
 
     class ToolP,Mapper router
     class McpP,ServerMgr,ToolMgr,McpClient mcp
-    class AgentToolMgr,FsHandler,Browser agent
+    class AgentToolMgr,FsHandler,YoBrowser agent
     class MCPServers,Files,Web external
 ```
 
@@ -622,23 +622,20 @@ class AgentFileSystemHandler {
 3. **边界检查**：防止 `../` 越界访问
 4. **正则验证**：`grep_search` 和 `text_replace` 使用 `validateRegexPattern` 防 ReDoS
 
-### Browser 工具
+### YoBrowser CDP 工具
 
-```typescript
-// 通过 Yo Browser Presenter 调用
-async callBrowserTool(toolName: string, args: any): Promise<string> {
-  switch (toolName) {
-    case 'browser_navigate':
-      return await this.yoBrowserPresenter.navigate(args.url)
-    case 'browser_scrape':
-      return await this.yoBrowserPresenter.scrape(args.url)
-    case 'browser_screenshot':
-      return await this.yoBrowserPresenter.screenshot(args.url)
-    default:
-      throw new Error(`未知的 Browser 工具: ${toolName}`)
-  }
-}
-```
+YoBrowser 提供基于 Chrome DevTools Protocol (CDP) 的最小工具集，在 agent 模式下直接可用。
+
+**可用工具**：
+- `yo_browser_tab_list` - 列出所有浏览器 tabs
+- `yo_browser_tab_new` - 创建新 tab
+- `yo_browser_tab_activate` - 激活指定 tab
+- `yo_browser_tab_close` - 关闭 tab
+- `yo_browser_cdp_send` - 发送 CDP 命令
+
+**安全约束**：
+- `local://` URL 禁止 CDP attach（在 `BrowserTab.ensureSession()` 中检查）
+- 所有 CDP 命令通过 `webContents.debugger.sendCommand()` 执行
 
 ## 🔐 权限系统
 

docs/archives/workspace-agent-refactoring-summary.md

@@ -196,7 +196,7 @@ graph TB
 
 - MCP 工具：保持原始命名
 - Agent FileSystem 工具：不加前缀（`read_file` 等）
-- Yo Browser：保留 `browser_` 前缀
+- Yo Browser：使用 `yo_browser_` 前缀
 
 ### 工具路由机制
 

docs/specs/agent-provider-simplification/plan.md

@@ -0,0 +1,68 @@
+# Plan: Agent Provider Simplification (ACP-only)
+
+## Summary
+
+Replace the “agent provider” abstraction and detection logic with a single explicit rule: **ACP is the only agent provider and is identified by `providerId === 'acp'`.**
+
+## Current Call Flow (relevant parts)
+
+- Main:
+  - `ProviderInstanceManager.createProviderInstance()` already special-cases `provider.id === 'acp'`.
+  - `ProviderInstanceManager.isAgentProvider()` uses `instanceof BaseAgentProvider` and (if instance not created) a constructor prototype check (`isAgentConstructor`).
+  - `LLMProviderPresenter.isAgentProvider()` exposes this to the renderer via `ILlmProviderPresenter`.
+- Renderer:
+  - `src/renderer/src/stores/modelStore.ts` calls `llmproviderPresenter.isAgentProvider(providerId)` over IPC to choose between:
+    - `agentModelStore.refreshAgentModels(providerId)` (ACP path)
+    - `refreshStandardModels + refreshCustomModels` (standard path)
+  - Other renderer logic already treats ACP as special via `provider.id === 'acp'`.
+
+## Proposed Changes
+
+### 1) Remove agent-provider classification API
+
+- Remove `isAgentProvider(providerId: string)` from:
+  - `src/shared/types/presenters/llmprovider.presenter.d.ts`
+  - `src/shared/types/presenters/legacy.presenters.d.ts`
+  - `src/main/presenter/llmProviderPresenter/index.ts`
+  - `src/main/presenter/llmProviderPresenter/managers/providerInstanceManager.ts`
+
+Rationale: It is only used by the renderer for ACP gating, and ACP can be identified locally by ID.
+
+### 2) Replace renderer gating with an explicit ACP check
+
+- In `src/renderer/src/stores/modelStore.ts`:
+  - Remove the async IPC call `llmP.isAgentProvider(providerId)`.
+  - Replace with a local predicate: `providerId === 'acp'`.
+  - Keep the existing ACP refresh path using `agentModelStore.refreshAgentModels('acp')` (no behavioral change).
+
+### 3) Remove `BaseAgentProvider` (optional but preferred)
+
+Because `BaseAgentProvider` is only used by `AcpProvider`, delete the base class and:
+
+- Make `AcpProvider` extend `BaseLLMProvider` directly.
+- Move `cleanup()` logic into `AcpProvider` (or delegate to `AcpSessionManager` / `AcpProcessManager`).
+- Ensure `cleanup()` is safe to call multiple times and during shutdown.
+
+Notes:
+- `acpCleanupHook` currently awaits `cleanup()` even though `BaseAgentProvider.cleanup()` is `void`. Consider standardizing ACP cleanup to `Promise<void>` to match usage.
+
+## Compatibility / Migration
+
+- No user data migration.
+- Provider ID `acp` remains unchanged and is treated as a stable internal contract.
+- Any internal IPC typing generation must be updated to reflect removal of `isAgentProvider`.
+
+## Test Strategy
+
+Add minimal tests focusing on the only behavioral dependency (renderer model refresh selection):
+
+- Renderer unit test for `modelStore.refreshProviderModels()`:
+  - When `providerId === 'acp'`, it uses `agentModelStore.refreshAgentModels`.
+  - When `providerId !== 'acp'`, it uses standard refresh path.
+
+Main-process unit tests are optional; the change is mostly removal and ACP-id checks.
+
+## Rollout
+
+Single PR is acceptable if changes stay localized (types + modelStore + ACP provider base class cleanup).
+

docs/specs/agent-provider-simplification/spec.md

@@ -0,0 +1,43 @@
+# Agent Provider Simplification (ACP-only)
+
+## Background
+
+DeepChat currently distinguishes between:
+
+- **LLM providers**: network-backed providers that implement `BaseLLMProvider` (OpenAI/Anthropic/etc).
+- **Agent providers**: providers that manage local agent sessions/processes (currently only `acp` via `AcpProvider`).
+
+The codebase implements this distinction via a dedicated base class (`BaseAgentProvider`) and a runtime/type-detection API (`isAgentProvider`), which is then consumed from the renderer via IPC.
+
+## Problem
+
+- `BaseAgentProvider` is only used by `AcpProvider`, so the abstraction adds indirection without real reuse.
+- Provider type detection is over-engineered (`isAgentConstructor` + prototype checks) and duplicates existing ACP-specific branching.
+- The renderer calls `llmproviderPresenter.isAgentProvider(providerId)` over IPC, but the only “agent provider” is `providerId === 'acp'`. This creates unnecessary main↔renderer coupling and call complexity.
+
+## Goals
+
+- Treat **ACP as the only agent provider** and identify it **only by `providerId === 'acp'`**.
+- Remove the generic “agent provider type detection” path and the renderer IPC dependency for this decision.
+- Keep user-visible behavior unchanged:
+  - ACP agents still appear as selectable models when ACP is enabled.
+  - Non-ACP providers keep the standard model/custom-model refresh behavior.
+  - Shutdown and provider disable still clean up ACP resources.
+
+## Non-goals
+
+- Supporting multiple agent providers beyond ACP.
+- Redesigning ACP model derivation (agents-as-models) or session/workspace semantics.
+- Changing persisted provider IDs or stored settings schemas.
+
+## Acceptance Criteria
+
+- Renderer no longer calls `llmproviderPresenter.isAgentProvider(...)`; ACP decision is local (`providerId === 'acp'`).
+- Main process no longer needs `isAgentConstructor` / prototype-based provider classification.
+- No remaining runtime dependency on `BaseAgentProvider` for correctness (ACP cleanup remains correct).
+- `pnpm run typecheck`, `pnpm test`, `pnpm run lint` pass.
+
+## Open Questions
+
+- None.
+

docs/specs/agent-provider-simplification/tasks.md

@@ -0,0 +1,26 @@
+# Tasks: Agent Provider Simplification (ACP-only)
+
+1. Update renderer to stop using IPC for agent-provider detection
+   - Remove `llmproviderPresenter.isAgentProvider` usage from `src/renderer/src/stores/modelStore.ts`.
+   - Gate ACP behavior by `providerId === 'acp'`.
+
+2. Remove `isAgentProvider` from the presenter contract
+   - Remove from `src/shared/types/presenters/llmprovider.presenter.d.ts`.
+   - Remove from `src/shared/types/presenters/legacy.presenters.d.ts`.
+   - Remove implementation from `src/main/presenter/llmProviderPresenter/index.ts`.
+
+3. Remove main-side agent-provider classification implementation
+   - Delete `ProviderInstanceManager.isAgentProvider()` and `isAgentConstructor()` in `src/main/presenter/llmProviderPresenter/managers/providerInstanceManager.ts`.
+   - Ensure no other code path depends on `BaseAgentProvider` type checks.
+
+4. Remove `BaseAgentProvider` abstraction (preferred)
+   - Delete `src/main/presenter/llmProviderPresenter/baseAgentProvider.ts`.
+   - Update `src/main/presenter/llmProviderPresenter/providers/acpProvider.ts` to extend `BaseLLMProvider` directly.
+   - Keep/adjust ACP cleanup semantics (safe shutdown, provider disable, app quit).
+
+5. Add/adjust tests
+   - Add a Vitest suite under `test/renderer/**` validating model refresh selection for ACP vs non-ACP.
+
+6. Quality gates
+   - Run `pnpm run format`, `pnpm run lint`, `pnpm run typecheck`, and `pnpm test`.
+

docs/specs/chat-settings-control/plan.md

@@ -0,0 +1,122 @@
+# Plan: Control Settings via Chat
+
+## Key Decision: Skill-Based Context Control
+
+This feature MUST be described and delivered as a DeepChat skill so that additional instructions/context are only injected when the user actually requests to change DeepChat settings.
+
+- Skill Name (suggested): `deepchat-settings`
+- Activation: Activated via `skill_control` **ONLY** when the user request involves DeepChat settings/preferences.
+- Deactivation: Call `skill_control` after completing the setting change to keep context lean.
+
+## Tool Injection Control (No Skill, No Tools)
+
+Configuration-related tools MUST NOT appear in the LLM tool list (and MUST NOT be mentioned in the system prompt) unless the `deepchat-settings` skill is active.
+
+Implementation intent:
+
+- Define dedicated tools (MCP-format function definitions):
+  - `deepchat_settings_toggle`
+  - `deepchat_settings_set_language`
+  - `deepchat_settings_set_theme`
+  - `deepchat_settings_set_font_size`
+  - `deepchat_settings_open`
+- **DO NOT** expose them through MCP server/tool list UI (avoid being auto-enabled into `enabledMcpTools`).
+- Only inject these tool definitions when:
+  - `deepchat-settings` is enabled for the current conversation, AND
+  - The skill's pre-metadata `allowedTools` includes the tool name.
+
+This requires conversation-scoped tool definition construction:
+
+- Extend tool definition construction context to include `conversationId`.
+- Retrieve `skillsAllowedTools` for that conversation (via `SkillPresenter.getActiveSkillsAllowedTools`).
+- Only conditionally append `deepchat_settings_*` tool definitions when allowed.
+
+## Step 1: Safe Settings Application API (Main Process)
+
+### Entry Point
+
+Implement a narrow, validated application surface in the main process (presenter method or agent tool handler) for:
+
+- Accepting `unknown` input and validating it (Zod-style, similar to `AgentFileSystemHandler`).
+- Using an allowlist of setting IDs.
+- Applying changes by calling existing `ConfigPresenter` methods so existing event broadcasts remain correct.
+- Returning structured results to render confirmation/error messages.
+
+### Allowlisted Settings and Mapping
+
+Toggle settings:
+
+- `soundEnabled` -> `ConfigPresenter.setSoundEnabled(boolean)` (broadcasts: `CONFIG_EVENTS.SOUND_ENABLED_CHANGED`)
+- `copyWithCotEnabled` -> `ConfigPresenter.setCopyWithCotEnabled(boolean)` (broadcasts: `CONFIG_EVENTS.COPY_WITH_COT_CHANGED`)
+
+Enum settings:
+
+- `language` -> `ConfigPresenter.setLanguage(locale)` (broadcasts: `CONFIG_EVENTS.LANGUAGE_CHANGED`)
+- `theme` -> `ConfigPresenter.setTheme('dark' | 'light' | 'system')` (broadcasts: `CONFIG_EVENTS.THEME_CHANGED`)
+- `fontSizeLevel` -> `ConfigPresenter.setSetting('fontSizeLevel', level)` (broadcasts `CONFIG_EVENTS.FONT_SIZE_CHANGED` via special case)
+
+### Validation Rules
+
+- Strict allowlist; reject unknown IDs.
+- No implicit type conversion in Step 1.
+- Validation per setting:
+  - Booleans: must be boolean type
+  - Enum values: must match allowed set
+  - `fontSizeLevel`: must be integer within supported range (source of truth TBD; may align with `uiSettingsStore` constants)
+  - `language`: must be one of supported locales (reuse support list from config)
+
+### Defense in Depth: Require Skill Activity
+
+Even with controlled tool injection, maintain runtime checks:
+
+- If `deepchat-settings` is **NOT** enabled for the conversation, reject application and return error telling the model/user to activate it.
+- This ensures settings don't accidentally change due to unrelated agent behavior.
+
+## Step 2: Skill Definition (Natural Language Behavior)
+
+### Built-in Skill Artifact
+
+Add `resources/skills/deepchat-settings/SKILL.md`:
+
+- Pre-metadata `description` MUST explicitly state:
+  - This is ONLY for changing DeepChat application settings.
+  - Activate ONLY when user requests setting changes (settings/preferences/theme/language/font/sound/copy COT).
+  - Do NOT activate for OS settings or programming/code settings.
+- Body MUST define:
+  - Supported settings (allowlist) and canonical values.
+  - How to ask clarifying questions when ambiguous.
+  - When to refuse and instead open settings.
+  - Always deactivate after completing setting tasks.
+
+### Disallowed Settings -> Open Settings
+
+For requests involving MCP configuration, prompts, providers, API keys, etc.:
+
+- Do NOT apply via tools.
+- Provide precise instructions telling user where to change them.
+- Open settings window and navigate to relevant section if possible.
+
+Implementation options for opening/navigating settings:
+
+- Use `presenter.windowPresenter.createSettingsWindow()`.
+- Optionally `executeJavaScript` to set localStorage navigation hint that UI can read.
+- Or add dedicated IPC channel from main process -> settings renderer to navigate to tab/section.
+
+## Data Model
+
+Introduce shared request/response types (for Step 1 entry point + tools):
+
+- `ChatSettingId` (union of allowlisted IDs)
+- `ApplyChatSettingRequest` (discriminated union `{ id, value }`)
+- `ApplyChatSettingResult`
+  - `{ ok: true; id; value; previousValue?; appliedAt }`
+  - `{ ok: false; errorCode; message; details? }`
+
+## Testing Strategy
+
+- Main process (Vitest):
+  - Allowlist + validation (reject invalid values, no writes)
+  - Each supported setting maps to correct `ConfigPresenter` method
+  - Skill requirement enforcement works (tool rejects when skill inactive)
+- Renderer/UI (if any navigation hints added):
+  - Settings page navigation handler tests (optional)