Gemini authentication with browser-based OAuth login and model selection dropdowns

.files/test_fixes.md

@@ -0,0 +1,185 @@
+# Test Isolation Fixes
+
+Two bugs. Two files touched.
+
+---
+
+## 1. `packages/modifiers-napi/src/__tests__/index.test.ts`
+
+**Problem:** `mock.module('bun:ffi', ...)` is declared at module scope with no cleanup.
+Bun reuses workers across files in a full suite run, so the mock bleeds into other
+files that import anything touching `bun:ffi`. When `index.test.ts` runs _after_
+another file has already loaded `bun:ffi` into the module registry, the module-level
+mock declaration loses the race and the test gets the wrong module — or nothing at all.
+
+**Fix:** Add `afterAll(() => mock.restore())`.
+
+```typescript
+// packages/modifiers-napi/src/__tests__/index.test.ts
+
+import { afterAll, beforeEach, describe, expect, mock, test } from 'bun:test'
+
+let ffiShouldThrow = false
+let nativeFlags = 0
+let dlopenCalls = 0
+
+mock.module('bun:ffi', () => ({
+  FFIType: {
+    i32: 0,
+    u64: 0,
+  },
+  dlopen: () => {
+    dlopenCalls++
+    if (ffiShouldThrow) {
+      throw new Error('ffi load failed')
+    }
+    return {
+      symbols: {
+        CGEventSourceFlagsState: () => nativeFlags,
+      },
+    }
+  },
+}))
+
+// ✅ NEW: restore the bun:ffi mock after this file finishes
+// so it doesn't bleed into other files running in the same worker
+afterAll(() => {
+  mock.restore()
+})
+
+beforeEach(() => {
+  ffiShouldThrow = false
+  nativeFlags = 0
+  dlopenCalls = 0
+})
+
+// ... rest of file unchanged
+```
+
+---
+
+## 2. `src/utils/staticRender.tsx`
+
+**Problem:** `await instance.waitUntilExit()` hangs indefinitely when Ink's reconciler
+commit phase doesn't complete. `waitUntilExit()` only resolves when `exit()` is called
+inside `useLayoutEffect` — but if `@anthropic/ink`'s `wrappedRender` has stale
+process-level state from a previous test file (singleton reconciler, leaked
+`process.on('exit')` handlers, a cached fiber root), React never commits, `useLayoutEffect`
+never fires, and the whole render silently hangs until the 10s test timeout kills it.
+
+This only surfaces in the full suite because Bun shares workers across files sequentially.
+Running the file alone gives it a clean worker with no prior Ink state.
+
+**Fix:** Race `waitUntilExit()` against a 3s safety timeout. On timeout, call
+`instance.unmount()` to force Ink cleanup and throw a descriptive error instead of
+a silent 10s hang.
+
+Also: explicitly remove the stream `data` listener and destroy the stream after render.
+Without this, the PassThrough stream stays open and its listener accumulates garbage
+across calls in the same process lifetime.
+
+```typescript
+// src/utils/staticRender.tsx
+
+import * as React from 'react';
+import { useLayoutEffect } from 'react';
+import { PassThrough } from 'stream';
+import stripAnsi from 'strip-ansi';
+import { wrappedRender as render, useApp } from '@anthropic/ink';
+
+function RenderOnceAndExit({ children }: { children: React.ReactNode }): React.ReactNode {
+  const { exit } = useApp();
+
+  useLayoutEffect(() => {
+    const timer = setTimeout(exit, 0);
+    return () => clearTimeout(timer);
+  }, [exit]);
+
+  return <>{children}</>;
+}
+
+const SYNC_START = '\x1B[?2026h';
+const SYNC_END = '\x1B[?2026l';
+
+function extractFirstFrame(output: string): string {
+  const startIndex = output.indexOf(SYNC_START);
+  if (startIndex === -1) return output;
+
+  const contentStart = startIndex + SYNC_START.length;
+  const endIndex = output.indexOf(SYNC_END, contentStart);
+  if (endIndex === -1) return output;
+
+  return output.slice(contentStart, endIndex);
+}
+
+export async function renderToAnsiString(node: React.ReactNode, columns?: number): Promise<string> {
+  let output = '';
+
+  const stream = new PassThrough();
+  if (columns !== undefined) {
+    (stream as unknown as { columns: number }).columns = columns;
+  }
+
+  // ✅ NEW: named handler so we can remove it after render
+  const dataHandler = (chunk: Buffer) => {
+    output += chunk.toString();
+  };
+  stream.on('data', dataHandler);
+
+  const instance = await render(<RenderOnceAndExit>{node}</RenderOnceAndExit>, {
+    stdout: stream as unknown as NodeJS.WriteStream,
+    patchConsole: false,
+  });
+
+  // ✅ NEW: race waitUntilExit against a 3s hard limit.
+  // If Ink's reconciler is stuck (stale worker state from a prior test file),
+  // this surfaces a real error instead of silently hanging for 10s.
+  await Promise.race([
+    instance.waitUntilExit(),
+    new Promise<void>((_, reject) =>
+      setTimeout(() => {
+        instance.unmount();
+        reject(
+          new Error(
+            '[staticRender] Ink render did not exit within 3s — wrappedRender may have stale process state from a prior test file',
+          ),
+        );
+      }, 3000),
+    ),
+  ]);
+
+  // ✅ NEW: clean up the stream so it doesn't accumulate across calls
+  stream.off('data', dataHandler);
+  stream.destroy();
+
+  return extractFirstFrame(output);
+}
+
+export async function renderToString(node: React.ReactNode, columns?: number): Promise<string> {
+  const output = await renderToAnsiString(node, columns);
+  return stripAnsi(output);
+}
+```
+
+---
+
+## If AutofixProgress tests still hang after this
+
+The 3s safety net will stop the silent timeout and give you a real error message.
+If you're still seeing the hang, the root is inside `wrappedRender` from `@anthropic/ink`.
+Look for any of these in that file:
+
+- A module-level singleton (cached app instance, fiber root, reconciler)
+- `process.on('exit')` / `process.on('SIGTERM')` handlers added on each `render()` call but never removed
+- A global `stdout` patch applied once and assumed fresh on every call
+
+Share that file and the exact line can be pinpointed.
+
+---
+
+## Summary
+
+| File               | Change                                                  | Why                                              |
+| ------------------ | ------------------------------------------------------- | ------------------------------------------------ |
+| `index.test.ts`    | Add `afterAll(() => mock.restore())`                    | Stops `bun:ffi` mock bleeding into other workers |
+| `staticRender.tsx` | Race `waitUntilExit()` with 3s timeout + stream cleanup | Surfaces real errors instead of silent 10s hangs |

.gitignore

@@ -12,10 +12,12 @@ coverage
 src/utils/vendor/
 
 # AI tool runtime directories
-.agents/
-.claude/
-.omx/
-.docs/task/
+.agents/*
+.claude/*
+.omx/*
+.docs/task/*
+.files/OAuth.json
+
 # Binary / screenshot files (root only)
 /*.png
 *.bmp

README.md

@@ -26,6 +26,7 @@
 | **Poor Mode**               | 穷鬼模式，关闭记忆提取和键入建议,大幅度减少并发请求                                                                          | /poor 可以开关                                                                                                                            |
 | **Channels 频道通知**       | MCP 服务器推送外部消息到会话（飞书/Slack/Discord/微信等），`--channels plugin:name@marketplace` 启用                         | [文档](https://ccb.agent-aura.top/docs/features/channels)                                                                                 |
 | **自定义模型供应商**        | OpenAI/Anthropic/Gemini/Grok 兼容  (`/login`)                                                                                          | [文档](https://ccb.agent-aura.top/docs/features/all-features-guide)                                                                        |
+| **本地 LLM (Ollama/Local)** | 支持 Ollama, LM Studio, Jan.ai, LocalAI。支持在 `/login` 中一键拉取模型、检查硬件状态、本地优先运行。                               | /login 选择 Local LLM                                                                                                                     |
 | Voice Mode                  | 语音输入，支持豆包语言输入（`/voice doubao`）                                                                   | [文档](https://ccb.agent-aura.top/docs/features/voice-mode)                                                                               |
 | Computer Use                | 屏幕截图、键鼠控制                                                                                                           | [文档](https://ccb.agent-aura.top/docs/features/computer-use)                                                                             |
 | Chrome Use                  | 浏览器自动化、表单填写、数据抓取                                                                                             | [自托管](https://ccb.agent-aura.top/docs/features/chrome-use-mcp) [原生版](https://ccb.agent-aura.top/docs/features/claude-in-chrome-mcp) |
@@ -144,10 +145,17 @@ bun run build
 
 ### 👤 新人配置 /login
 
-首次运行后，在 REPL 中输入 `/login` 命令进入登录配置界面，选择 **Anthropic Compatible** 即可对接第三方 API 兼容服务（无需 Anthropic 官方账号）。
-选择 OpenAI 和 Gemini 对应的栏目都是支持相应协议的
+首次运行后，在 REPL 中输入 `/login` 命令进入登录配置界面：
 
-需要填写的字段：
+1. **Anthropic Compatible**: 对接第三方 API 兼容服务（OpenRouter、AWS Bedrock 代理等）。
+2. **OpenAI / Gemini / Grok**: 对应各自协议的云端服务。
+   - **Gemini (Google Auth)**: 支持交互式浏览器登录。
+     1. 在 Google Cloud Console 的 **APIs & Services > OAuth consent screen** 中配置 OAuth 客户端（User Type 设为 External）。
+     2. 下载 credentials JSON 文件并保存到项目根目录的 `.files/OAuth.json`。     3. 在 `/login` 配置界面中留空 API Key 并按 Enter，程序将自动拉起浏览器完成授权并自动拉取模型列表。
+3. **Local LLM**: **(推荐)** 使用本地运行的模型。
+   - 支持 **Ollama**, **LM Studio**, **Jan.ai**, **LocalAI**。
+   - **Ollama 深度集成**: 可直接在 CLI 中查看已安装模型，或输入模型名（如 `llama3.1`）一键拉取（Pull）。支持模型列表交互切换和硬件状态自动检测。
+   - 自动检测本地运行状态和默认端口。
 
 | 📌 字段      | 📝 说明       | 💡 示例                      |
 | ------------ | ------------- | ---------------------------- |
@@ -157,9 +165,22 @@ bun run build
 | Sonnet Model | 均衡模型 ID   | `claude-sonnet-4-6`          |
 | Opus Model   | 高性能模型 ID | `claude-opus-4-6`            |
 
-- ⌨️ **Tab / Shift+Tab** 切换字段，**Enter** 确认并跳到下一个，最后一个字段按 Enter 保存
+### 🩺 系统诊断 /doctor
 
-> ℹ️ 支持所有 Anthropic API 兼容服务（如 OpenRouter、AWS Bedrock 代理等），只要接口兼容 Messages API 即可。
+如果你在使用过程中遇到环境问题（尤其是本地模型运行缓慢或无法连接），可以使用 `/doctor` 命令进行全方位诊断：
+
+- **硬件负载**: 自动显示当前 CPU 型号、核心数、剩余内存 (RAM) 以及系统架构。
+- **本地环境**: 检查 Ollama 等本地 Runner 是否正在运行，并列出所有可用模型。
+- **配置校验**: 检查环境变量（如 `LOCAL_BASE_URL`）和权限设置。
+- **故障排查**: 识别多个重复安装的版本、过期的版本锁或权限不足的更新。
+
+## 环境变量 (Environment Variables)
+
+除了交互式 `/login` 配置外，你也可以通过环境变量直接配置：
+
+- `LOCAL_BASE_URL`: 本地 LLM 运行地址 (例如 `http://localhost:11434`)。
+- `LOCAL_MODEL`: 本地 LLM 模型名称 (例如 `llama3.1`)。用于 `local` provider 时覆盖默认模型。
+- `GEMINI_BASE_URL`: Gemini API 的自定义基础地址。
 
 ## Feature Flags