# translate_tool
**Repository Path**: simonyuan2019/translate_tool
## Basic Information
- **Project Name**: translate_tool
- **Description**: 基于大模型 API 的在线翻译工具，支持网页版和 Chrome 扩展，全球 30+ 语言互译，支持划词翻译
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No
## Statistics
- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-06-20
- **Last Updated**: 2026-09-06
## Categories & Tags
**Categories**: Uncategorized
**Tags**: None
## README
# AI Tool Box · AI Smart Translation & Productivity Toolkit
> An online translation tool powered by LLM APIs, available as a web app and Chrome extension, supporting 30+ languages with text selection translation.
**Language**: [中文](README.md) | English
---
## ✨ Features
### Web Version
- **30+ Languages** — Chinese, English, Japanese, Korean, French, German, Spanish, Russian, Arabic, and more
- **Flexible API** — Compatible with any OpenAI Chat Completions API provider (OpenAI, DeepSeek, Qwen, etc.)
- **Deep Context-Aware Translation** — AI performs a 5-step deep analysis before translating (domain identification → text type assessment → tone analysis → audience profiling → intent understanding) for significantly more accurate translations
- **Original Format Preservation** — Fully preserves Markdown, HTML, code blocks, and other original formats with auto-formatting after translation
- **Chrome Extension Fullscreen Mode** — New "fullscreen button" in popup top-right corner, opens full translation page in new tab (no height limit)
- **Dynamic Popup Height** — Max height set to `screen.availHeight`, can drag to screen bottom
- **6 Minimal Premium Themes with One-Click Switching** — Low-saturation Apple-style palette: light (Paper / Mist / Cream) × dark (Graphite / Slate / Nocturne), neutral layered backgrounds, restrained accent colors, Apple-style text ladder, hairline borders + soft diffuse shadows; floating panel in the bottom-right corner, default Graphite
- **Branded Title Typography** — Display font Sora (async loaded, falls back to Syne/Plus Jakarta Sans), gradient text + glow, clamp-based responsive sizing
- **Typing Effect** — Translation results appear character by character for a smooth experience
- **Translation History** — Auto-saves up to 20 recent translations with one-click recall
- **Quick Actions** — Swap languages, paste from clipboard, clear, and copy results
- **Keyboard Shortcut** — `Ctrl + Enter` to translate instantly
- **Privacy First** — All settings and history stored locally in browser localStorage
- **Responsive Design** — Works seamlessly on desktop and mobile devices
- **Cross-surface record sync** — tasks, cards and every history are shared live between the web app, the extension popup and the side panel (changes on one surface appear instantly on the others and survive refreshes); the AI config syncs everywhere from a single save
- **History export** — Smart Translation history, Work Report records, Email Summary history, English Learning history and Hot News Radar cards can each be exported to a local JSON file with one click
- **Work Report** — Built-in work report generator with AI one-click summary, history management, and date filtering
- **Task List** — Built-in task manager with add/complete/delete, priority levels, progress tracking, Markdown batch import/export (with checkbox syntax), Apple Reminders one-click import (URL Scheme + AppleScript file fallback), Google Calendar sync, and .ics calendar download
- **English Learning Assistant** — Built-in English learning module with word study, full-text translation + difficult-word extraction for sentences/passages (up to 20 words, one card per word with a speaker button for one-click pronunciation), AI definitions, text-to-speech, learning history (restored exactly as studied), and note export
- **Email Summary** — New 5th tab: paste email threads or upload local files (.txt/.md/.eml/.pdf); AI produces a professional four-section detailed summary (subject & background / timeline table / technical key points / risks & caveats) plus a responsibility-based To Do List (P0/P1/P2 priorities), with 30-language output, HTML/Markdown download, and history review & editing
- **PDF Parsing** — Built-in pdf.js local parsing (up to 200MB) with automatic handling of huge files: reading progress indicator, early stop at text budget / page cap, per-page memory release
- **Auto-Truncation for Oversized Content** — Over 60,000 characters, automatically keeps head & tail, omits and marks the middle — no manual splitting needed
- **AI Parse** — New 6th tab: Classic mode (paste notes/requirements → AI extracts a task list, tick items and batch-create into the todo list, with priorities/tags/sub-steps) + Analysis mode (describe requirements or upload attachments → AI generates a structured analysis summary, auto-detects .eml email threads); results can be copied / downloaded as Markdown / HTML
- **AI Prompts** — New 7th tab: enter a rough requirement → generate an expert-level structured prompt (with "📋 Prompt", "⚠ Assumptions", "💡 Usage tips"); copy the prompt body or everything, download
- **Hot News Radar** — New 8th tab: create multiple hot-news cards, each with its own prompt (used by AI for categorization); AI fetches real-time hot lists from Weibo / Zhihu / Baidu / Douyin / Bilibili / IT之家 / 36Kr and picks the Top 10 most relevant entries per card; cards show rank colors, source tags, clickable source links, heat values, and a refresh button (re-fetch + AI re-curation)
- **Page Reuse Architecture** — Work Report, Task List, English Learning, Email Summary, AI Parse, AI Prompts, and Hot News Radar tabs embed standalone pages via iframe, sharing the same codebase with Chrome extension
- **Zero Dependencies** — Pure HTML + CSS + JavaScript, no installation required
### Chrome Extension
- **Text Selection Translation** — Select text on any webpage, a translation icon appears automatically
- **Popup Translation Panel** — Click the toolbar icon for quick text translation
- **Right-click Menu** — Select text and right-click to choose "AI Tool Box Translate"
- **6 Minimal Premium Themes + Textured UI** — Shares the theme.css system with the web version: restrained low-saturation palette, glassmorphism cards, subtle noise background and depth shadows
- **Deep Context-Aware Translation** — AI performs 5-step context analysis (domain/text type/tone/audience/intent) for precise translations
- **Resizable Popup** — Drag any edge or corner to freely resize the popup (320–800px wide, 300–780px tall), size auto-saved
- **Uninterrupted Translation** — Translation continues in background Service Worker even if popup closes; results auto-restored on reopen
- **Translation History** — Auto-saves up to 20 recent translations with individual deletion and clear-all
- **Original Format Preservation** — Supports Markdown, HTML format input with auto-formatting
- **Records & config shared across surfaces** — the popup, side panel and web version share all records and config: data created on any surface (including every module's history) syncs to the others in real time, and histories can be exported to local JSON files
- **Toggle Switch** — Enable/disable text selection translation in settings
- **Language Preference Memory** — Automatically saves source and target language choices
- **Email Summary Entry** — New envelope icon in popup header opens the Email Summary page in a new tab (full parity with web version, including PDF upload and 30-language output)
- **Side Panel** — Chrome 114+ dedicated side panel bundling all 8 modules (Smart Translation / Work Report / Task List / English Learning / Email Summary / AI Parse / AI Prompts / Hot News Radar) with one-click tab switching and lazy-loaded iframes; open it via the popup's Side Panel button, the `Alt+Shift+L` shortcut, or the "Open AI Tool Box in Side Panel" context-menu item
- **AI Parse / AI Prompts** — Two new Side Panel tabs (lazy-loaded); the popup also gains two entry buttons that open the standalone pages in a new tab
- **Hot News Radar** — New Side Panel tab (lazy-loaded); the popup gains an entry button; cards and results sync across pages via chrome.storage
## 📸 Preview
## 🚀 Quick Start
### How to Use
1. Open `index.html` in any modern browser
2. Click the **"API 设置" (API Settings)** button in the top-right corner
3. Fill in the configuration:
| Field | Description | Example |
|-------|-------------|---------|
| **Base URL** | LLM API endpoint | `https://api.openai.com/v1` |
| **API Key** | Your API key | `sk-xxxxxxxxxxxxxxxx` |
| **Model** | Model name | `gpt-4o` / `deepseek-v4-pro` |
4. Click **"保存配置" (Save)**
5. Select source and target languages, enter text, and click **"开始翻译" (Translate)**
### Supported API Providers
| Provider | Base URL | Model Examples |
|----------|----------|----------------|
| OpenAI | `https://api.openai.com/v1` | `gpt-4o`, `gpt-4o-mini` |
| DeepSeek | `https://api.deepseek.com/v1` | `deepseek-v4-pro` |
| Qwen (Alibaba) | `https://dashscope.aliyuncs.com/compatible-mode/v1` | `qwen-plus` |
| Zhipu AI | `https://open.bigmodel.cn/api/paas/v4` | `glm-4-flash` |
| Moonshot | `https://api.moonshot.cn/v1` | `moonshot-v1-8k` |
> Any service compatible with the OpenAI `/chat/completions` endpoint will work.
## ⌨️ Keyboard Shortcuts
| Shortcut | Action |
|----------|--------|
| `Ctrl` + `Enter` | Start translation |
## 🛠️ Tech Stack
- Pure HTML + CSS + JavaScript (zero framework dependencies)
- Google Fonts (Inter + Noto Sans SC)
- OpenAI-compatible Chat Completions API
## 📁 Project Structure
```
translation_tool/
├── index.html # Web app (translation main page, 8 tabs: Translate / Report / Todos / English / Email / AI Parse / AI Prompts / Hot News Radar)
├── workreport.html # Work report page (standalone, shared with extension)
├── workreport.js # Work report core logic (IIFE encapsulated)
├── english_learning.html # English learning assistant page (standalone, shared)
├── todolist.html # Task list page (standalone, with Markdown/AppleScript features)
├── todolist.js # Task list core logic (IIFE encapsulated, cross-page AI Parse task sync)
├── ai-service.js # AI Parse / AI Prompts shared service (config read-write / OpenAI-compatible chat / task extraction / prompt engineering)
├── ai_parse.html # AI Parse page (classic task extraction + analysis mode structured summary)
├── ai_parse.js # AI Parse page logic
├── ai_prompts.html # AI Prompts page (enter a requirement → generate a structured prompt)
├── ai_prompts.js # AI Prompts page logic
├── hotnews.html # Hot News Radar page (card-based Top 10 hot news, AI-curated by prompt)
├── hotnews.js # Hot News Radar logic (hot-list aggregation + AI curation + card management)
├── install_url_scheme.sh # Apple Reminders URL Scheme bridge installer
├── theme.css # Shared theme system (6 minimal premium theme variables + per-page UI layers + glass cards/noise/ambient glow styles)
├── theme.js # Theme switcher / data-mode light-dark flag / iframe theme sync / legacy theme migration / MD preview binding
├── markdown.js # Markdown renderer
├── md-editor.js # Markdown editor component
├── email_summary.html # Email summary page (standalone, shared with extension)
├── email_summary.js # Email summary core logic (SKILL prompt / AI calls / PDF parsing / history)
├── pdf.min.js # pdf.js 3.11.174 (PDF text extraction, bundled locally)
├── pdf.worker.min.js # pdf.js worker
├── ai_summary_prompt.md # Work report AI summary prompt documentation
├── preview.png # Web version screenshot
├── chrome_extension/ # Chrome browser extension
│ ├── manifest.json # Extension config
│ ├── popup.html # Popup UI
│ ├── popup.css # Popup styles
│ ├── popup.js # Popup logic
│ ├── sidepanel.html # Side Panel UI (Chrome 114+, 8 tabs)
│ ├── sidepanel.css # Side Panel styles (single-layer, theme-token driven)
│ ├── sidepanel.js # Side Panel logic (tab switching / lazy load / config sync)
│ ├── ai-service.js # AI Parse / AI Prompts shared service (shared with web version)
│ ├── ai_parse.html # AI Parse page (Chrome extension version)
│ ├── ai_parse.js # AI Parse page logic
│ ├── ai_prompts.html # AI Prompts page (Chrome extension version)
│ ├── ai_prompts.js # AI Prompts page logic
│ ├── fullpage.html # Fullscreen translation page
│ ├── fullpage.js # Fullscreen page logic
│ ├── content.js # Text selection translation
│ ├── content.css # Tooltip styles
│ ├── background.js # Service worker
│ ├── workreport.html # Chrome extension work report page
│ ├── workreport.js # Chrome extension work report logic
│ ├── english_learning.html # English learning assistant page
│ ├── english_learning.js # English learning logic (external JS, CSP compliant)
│ ├── todolist.html # Task list page
│ ├── todolist.js # Task list logic
│ ├── email_summary.html # Email summary page
│ ├── email_summary.js # Email summary logic
│ ├── pdf.min.js # pdf.js (PDF parsing, bundled locally for MV3 CSP)
│ ├── pdf.worker.min.js # pdf.js worker
│ ├── install_url_scheme.sh # URL Scheme bridge installer
│ ├── native_host.py # Chrome Native Messaging host (optional)
│ ├── native_host_manifest.json # Native Messaging manifest template
│ ├── install_native_host.sh # Native Messaging install script
│ ├── icons/ # Extension icons
│ └── _locales/ # i18n files
├── vibe_images/ # Icon source files
├── README.md # Chinese documentation
└── README_EN.md # English documentation
```
## 🧩 Chrome Extension
In addition to the web version, this project includes a **Chrome browser extension** with **text selection translation**.
### Extension Features
- **Popup Translation Panel** — Click the toolbar icon for quick text translation
- **Text Selection Translation** — Select text on any webpage, a translation icon appears automatically
- **Right-click Menu** — Select text and right-click to choose "LinguaFlow Translate"
- **6 Minimal Premium Themes + Textured UI** — Shares the theme system with the web version: restrained low-saturation palette, glassmorphism cards, subtle noise background and depth shadows
- **Resizable Popup** — Drag any edge or corner to resize (320–800px wide, 300–780px tall), auto-saved
- **Uninterrupted Translation** — Background Service Worker continues translating even after popup closes; results auto-restored
- **Translation History** — Auto-saves up to 20 entries with individual deletion and clear-all
- **Deep Context-Aware Translation** — 5-step analysis workflow for precise, natural translations
- **30+ Languages** — Same language support as the web version
- **Typing Effect** — Translation results appear character by character
- **Toggle Switch** — Enable/disable text selection translation in settings
### Installation
1. Copy the `chrome_extension` folder to your local machine
2. Open Chrome and navigate to `chrome://extensions/`
3. Enable **Developer mode** in the top-right corner
4. Click **"Load unpacked"**
5. Select the `chrome_extension` folder
6. Click the AI Tool Box icon in the toolbar and configure your API to get started
### How to Use Text Selection Translation
1. Select text on any webpage with your mouse
2. A purple translation icon appears near the selection
3. Click the icon to see the translation in a floating tooltip
4. Copy the result with one click
### How to Use the Side Panel (Chrome 114+)
1. Click the AI Tool Box toolbar icon, then click the **Side Panel** button in the popup header (or press `Alt+Shift+L`, or right-click → "Open AI Tool Box in Side Panel")
2. The side panel opens on the right edge of the window with 8 tabs: **Smart Translation / Work Report / Task List / English Learning / Email Summary / AI Parse / AI Prompts / Hot News Radar**
3. Click any tab to switch modules instantly — each module is lazily loaded on first open to keep startup fast
4. Configure your API in the side panel's **Settings** panel (gear icon) — the config syncs to the popup, fullscreen page and every module in real time
5. Click the circular theme button in the bottom-right corner to switch between the 6 themes directly inside the side panel — changes sync to every module in real time
## 📋 Browser Compatibility
- Chrome 90+
- Edge 90+
- Firefox 88+
- Safari 15+
## 📝 Changelog
### v0.25.10 (2026-09-06)
- **Fix: selection-translation icon still appears after turning the toggle off** — the bug was in storage, not the toggle logic: besides the popup's "save settings" (which writes the full config object), the side panel, AI Parse/Prompts/Hot News (`AiService.saveConfig`) and the web settings page (relayed `content.js → background`) all wrote a **partial object replacing `chrome.storage.local.config` wholesale**, silently dropping `enableSelectTranslate`. The selection feature defaults to on (check is `!== false`), so the icon "resurrected" whenever the field vanished (`sourceLang/targetLang` were lost the same way). Fix: every partial save path now does **read → merge → write** (`chrome_extension/background.js`, `sidepanel.js`, both `ai-service.js` copies).
- **Fix: selection translation "Failed to fetch" on strict-CSP sites** — the content script used to `fetch()` the API directly in the page context, where requests are subject to the page's CSP `connect-src`; on document/enterprise sites with strict CSP the request was blocked by the browser (translation worked in the popup with the same config). Fix: selection requests now go through the existing background proxy bridge (`linguaflow:proxyFetch`, same channel as the web app, not bound by page CSP). URL normalization, reasoning-model gating and the 400 temperature-retry are all kept; network/CSP/dead-backend failures now surface actionable Chinese diagnostics, and API 4xx responses still pass through the provider error message.
### v0.25.9 (2026-09-05)
- **English Learning passage mode redesigned** — sentence/passage input (>4 words) now asks the model for `{translation, words}`: the result area shows a **full Chinese translation** card first, then full learning cards for **up to 20** difficult words (was 15, with no translation); every word card header gets a speaker button that reads the word aloud (reuses the pronunciation panel's voice/rate settings, wired via event delegation for CSP compliance).
- **Learning history now matches the study view** — fixes history restore showing the raw AI JSON string in passage mode: history entries store the **structured parsed result** (instead of raw AI text) and restore through the same `displayResult` pipeline used during study; legacy entries degrade to a Markdown fallback render. The "save today's content" export (md/html) is now generated from structured data (the old regex-over-rendered-HTML extraction had drifted from the render structure and produced empty notes); passage exports include the full translation; 4 dead helpers serving only the old export were removed.
- **Fix: "Clear all" no longer wipes learning history** — the learning-content clear button also deleted the `learningHistory` key, erasing all history; removed. Clearing now writes an empty state and relays it via `elRelayRecord` (removeItem-only left stale chrome.storage data that "resurrected" on reload in the extension); the same fix applies to the history section's "clear all". The learning-content "clear all" no longer shows a confirmation dialog per request.
### v0.25.8 (2026-09-05)
- **Hot News Radar switches its search engine to Tavily** — the card refresh pipeline is rebuilt in three steps: (1) AI extracts 3-5 core search keywords (incl. synonyms/related terms) from the card prompt, cached 30 min per prompt; (2) each keyword is searched via Tavily news API (2-day window, cached 5 min per keyword; CORS-open endpoint — the web page connects directly, no extension needed); (3) AI scores candidates on timeliness (prefer within 24h) / heat (frequency & discussion volume) / impact (scope), grouped into up to 10 themes, each item with Chinese title, 40-char summary, source and 0-100 heat score (weights 40/30/30). Cards now show theme groups, keyword chips, summaries and publish times.
- **New "Web Search (Tavily)" settings block** — key input with tvly- prefix validation and a connectivity test button, stored as hn_tavily_key and wired into the cross-surface record sync (free key at tavily.com).
- **Hot-board aggregation & Bing layer retired** — the 60s/UApi boards and Bing RSS channel are removed; legacy card data still renders and upgrades on refresh.
### v0.25.7 (2026-09-03)
- **English Learning supports long passages** — fixes "pasting a full English email only produced one word card": the prompt template was designed for a single word, so the model picked just one word from the passage. Input length now routes automatically: ≤4 words keep the single word/phrase flow; longer input switches to **material mode** — extracts key vocabulary (sorted by importance, up to 15, preferring harder/technical/pivotal terms) and renders a full learning card per word (phonetics/POS/definitions/usage/examples/synonyms/antonyms/mnemonics) with an extraction summary and numbering. History shows a short label ("xxx + N words") while still restoring the full original text. English Learning module only (both copies in sync), zero impact on other modules.
### v0.25.6 (2026-09-03)
- **Change: Email Summary back to one-shot output** — per user feedback, reverts the v0.25.5 streaming line-by-line rendering: the summary is now a non-streaming request rendered once when complete; while waiting, only elapsed time is shown ("generating… Xs"), the Cancel button remains (works on direct connections). All v0.25.5 compatibility fixes (URL normalization, reasoning-model gating, 400 retry, SW keepalive, error diagnostics) are retained. The streaming infrastructure (proxyFetchStream and the bridge chunk protocol) is kept but dormant — no module uses it, zero impact on the other 7 modules.
### v0.25.5 (2026-09-03)
- **Fix: "API service unreachable" when summarizing very long PDFs on the web app** — with no-CORS gateways (e.g. Alibaba Token Plan), the 60k-character **non-streaming** generation often takes 10+ minutes: idle connections get cut by gateways/middleboxes, or the page-side 600s bridge timeout fires first. Email Summary now uses **SSE streaming**: bytes keep flowing (no more idle cuts), a 180s **idle timeout** replaces the total timeout, output renders live (chars received · elapsed) with a **Cancel** button; the bridge gained a chunked transport channel (long port, background streams the response), falling back to the legacy non-streaming path (600s) if the stream channel fails before any chunk; partial output is preserved on mid-stream interruption.
- **Fix: extension Email Summary unable to load files** — root cause: the English Learning storage adapter wrote chrome.storage **objects** into localStorage without serializing (stored as "[object Object]"), and its change listener covered every synced key, poisoning the localStorage shared by all extension pages; Email Summary unguarded top-level JSON.parse crashed the whole module, disabling every button including "Select local file". Fixed: the writer now serializes; history/config reads in Email Summary / Smart Translation / English Learning are fault-tolerant (dirty data degrades safely instead of crashing the page). **Note: click "Reload" on chrome://extensions after updating**; manifest version bumped 0.20.0 to 0.25.5 so the loaded copy is identifiable.
- **Model compatibility unified** — URL normalization and reasoning-model parameter adaptation now cover every remaining AI call site: Smart Translation (index.html / popup background / fullpage / selection translate), Work Report (both copies), English Learning (both copies). Uniform behavior: buildUrl normalization (full-width colons/slashes, trailing slashes, duplicated endpoint — multi-level paths like Zhipu / Token Plan are the usual victims); isReasoningModel gating (deepseek-reasoner / qwq / thinking models no longer receive temperature / max_tokens); automatic retry without temperature on 400 errors mentioning it.
- **Fixed**: English Learning "Test connection" used the non-existent response.textAsync() (always failed); extension Hotnews used the response wrapper text method as a property in 3 places (bridge / Bing search / model list paths); Google Fonts onload inline handlers in 5 extension pages were blocked by MV3 CSP (now switched via JS).
- **New: "Fetch model list" in the main settings panel** — requests {Base URL}/models to autocomplete model IDs (same pattern as Hotnews Radar); essential for dedicated gateways (Token Plan / Zhipu Coding Plan) that only accept specific model IDs; goes through the proxy bridge so no-CORS endpoints work on the web app too.
- **New tests**: tests/stream-and-url.test.mjs (URL normalization; SSE chunk splitting / CRLF / [DONE] / reasoning_content; bridge stream chunking / cancel / fallback), tests/pdf-parse.test.mjs (PDF extraction pipeline using the bundled pdf.js).
### v0.25.3 (2026-09-02)
- **Fix: synced records disappearing after page refresh** — the content script was injected at `document_idle` (page scripts could read a stale localStorage first), and records created on the other surface while this page was closed never reached it. Fixes:
- content script now injects at **`document_start`** (before any page script runs)
- On injection it **pre-pulls** all 15 synced record keys from chrome.storage into localStorage (the extension is the authoritative source), so page initialization always reads the latest data — records created cross-surface survive refreshes
- Also eliminates a data-loss path where the English Learning page's init copy could wipe shared records with empty/stale local data
- Smart Translation history normalization: entries from the extension popup get their flag/time fields backfilled by language code for consistent display
### v0.25.2 (2026-09-02)
- **Live three-way record sync completed (web / extension / side panel)** — the v0.25.0 pipeline already shared data; this release adds the missing live-refresh listeners everywhere:
- New unified `AiService.onRecordSync(key, cb)` API (web: content.js record-sync message + storage event fallback; extension: chrome.storage.onChanged)
- Wired into: Work Report (records + summaries lists), Email Summary (history list), English Learning (history list), AI Parse & AI Prompts (history lists, never clobbering an in-progress draft)
- Previously live: Task List, Hot News Radar, Smart Translation history — every record module now syncs live across all three surfaces
### v0.25.1 (2026-09-02)
- **Fix: record sync not working for pages inside iframes** — content scripts only inject into the top frame, so reverse-sync messages from iframe pages (Hot News Radar, Task List, etc.) went unanswered while the top-level Smart Translation synced fine. All 14 relay points now post to **window.top**
- **Export all histories to local files** — each module's history area gains an 「导出」 button that downloads a JSON file (ai-toolbox--.json): Smart Translation history, Work Report records+summaries, Email Summary history, English Learning history, Hot News Radar cards (Task List keeps its existing Markdown/ics export)
### v0.25.0 (2026-09-02)
- **Two-way record sync between web and extension** — user records (previously stored separately on each surface) are now fully shared: task list, Hot News Radar cards, Smart Translation history/draft, work report records & summaries, email summary history, English learning history & content, AI Parse / AI Prompts state
- Mechanism: web writes → content.js relays into chrome.storage (extension side syncs); extension writes → background broadcasts to every tab's content.js which writes the matching localStorage key (web side syncs, automatically triggering the pages' existing storage listeners for live UI refresh)
- Live refresh: task list, Hot News Radar and Smart Translation history update on-screen instantly; other modules are write-through synced (data shared, visible after a page refresh)
- The chrome.storage ↔ localStorage key mapping lives in one table in background.js (covering td_/wr_ prefixes and popup's history/draft naming); adding a new synced record is a one-line change
- Requires the extension installed and allowed to access the page (file:// pages need "Allow access to file URLs")
### v0.24.2 (2026-09-02)
- **Proxy bridge now covers every AI module** — fixes the Smart Translation tab of index.html failing with "无法连接 API（CORS）" on Token Plan: the bridge previously only covered ai-service callers (AI Parse / AI Prompts / Hot News Radar), while Smart Translation, Work Report, Email Summary and English Learning used their own fetch calls. All are now wired through:
- `AiService.proxyFetch` returns a fetch-Response-compatible object (ok/status/text()/json()); every page's AI request goes through it (web direct-first, auto fallback to the extension bridge)
- 7 pages wired: index.html, Work Report (×2), Email Summary (×2), English Learning (×2)
- DeepSeek and Alibaba Cloud Token Plan now work in **every AI module on both the web and the extension** (web requires the extension installed and allowed to access the page)
- english_learning inline/external JS copies kept verbatim-identical (verified)
### v0.24.1 (2026-09-02)
- **Fix card refresh error "(g || []).forEach is not a function"** — the v0.24.0 search-layer merge mistakenly passed the pool object (returned by fetchPool) into mergePools as if it were an array; call site fixed and mergePools now guards against non-array entries
### v0.24.0 (2026-09-02)
- **Hot News Radar upgraded to true whole-web search** — candidate sources expanded from "10 hot-list boards" to a two-layer structure:
- **Hot-list layer (14+ boards)**: adds Hupu / WeRead / Juejin / ThePaper (14 UApi boards + 6 from 60s, merged and deduplicated)
- **Search layer (per-card prompt search)**: each card refresh now also queries **Bing web search RSS** with the card's prompt (10 real whole-web results with direct links and page summaries), relayed through the extension bridge to bypass CORS; search results are merged with hot-list entries before AI curation
- The AI prompt distinguishes the two candidate classes: Bing search results are prompt-matched by construction and kept first; hot-list entries are filtered by semantic relevance; search-layer failure gracefully degrades to hot-list-only mode
- End-to-end verified: 169 hot-list entries across 14 boards + Bing search layer → qwen3.7-max curated 10 strongly relevant AI items with reasons
### v0.23.2 (2026-09-01)
- **Fix: Hot News Radar "AI 返回格式异常 / empty results"** — two root causes:
- The candidate pool lacked vertical boards: the old "first successful source wins" chain meant the pool only contained general hot lists (Weibo/Zhihu/Douyin/Toutiao) with almost no tech items, so tech prompts inevitably curated an empty array. The pool is now a **parallel multi-source merge**: 60s general boards + UApi full boards (IT之家/36Kr/Bilibili/Baidu etc., via direct / extension bridge / public proxies), deduplicated — 124 items across 10 boards in testing
- An empty result was misreported as an error: a strict model returning `[]` (no strongly relevant entries) is a normal outcome, now gracefully presented as "no strongly relevant entries found; try a broader prompt or refresh"; JSON parsing gains trailing-comma tolerance and raw AI output is logged to console when parsing fails
- End-to-end verified (qwen3.7-max + 124-item merged pool): the "AI & LLM" prompt curated 5 strongly relevant items with reasons — precision over volume works
### v0.23.1 (2026-09-01)
- **Fix: CORS bridge unavailable for pages inside iframes (e.g. the Hot News Radar tab of index.html)** — content scripts only inject into the top frame by default (manifest lacks all_frames), so bridge messages posted to the iframe's own window went unanswered. Bridge requests now go to **window.top** (where the content script lives) and content.js replies to **e.source** (the exact frame that asked). Verified end-to-end with an iframe + mock content script + real HTTPS request. The Token Plan endpoint is reachable again from the radar tab inside index.html
- Hot News Radar error hints updated to mention installing the extension and enabling "Allow access to file URLs"
### v0.23.0 (2026-09-01)
- **Two-way config sync across all surfaces** — fixes the reverse gap: saving the API config in the web version (index.html settings / Hot News Radar config card) now relays through the extension content script into chrome.storage, so the popup and side panel sync instantly; previously only extension→web was synced. Saving on any surface now reaches every web page, the popup and the side panel in real time
- **CORS proxy bridge for the web version** — endpoints without CORS headers (e.g. Alibaba Cloud Token Plan) now work in the web version too: when a direct request fails, it is automatically relayed through the extension background (host_permissions bypasses CORS). AI calls, model-list fetching and hot-list retrieval all use this bridge (requires the extension installed and allowed to access the page; file:// pages need "Allow access to file URLs")
- **Merged Hot News Radar settings card** — "API config" and "create card" are now one "雷达设置" card (subsection titles + config status chip + divider), removing the cramped stacked cards; the config status is visible at a glance (model @ host when configured)
- **Stronger relevance filtering** — AI curation now prefers precision over volume: only entries strongly relevant to the prompt are kept, returning fewer than 10 when necessary instead of padding; each entry carries a short selection reason (shown under the title), and the card footer notes when fewer entries were returned
### v0.22.3 (2026-09-01)
- **Hot News Radar adapts to Alibaba Cloud Token Plan and similar gateways**:
- The API config card gains a "**Fetch model list**" button: it requests `{Base URL}/models` and turns the model input into an autocomplete dropdown (the Token Plan gateway only serves specific model IDs like `qwen3.6-flash` / `qwen3.7-plus` / `glm-5.2`; classic names such as `qwen-plus` return "Model not exist")
- Error messages now give targeted guidance for model-not-exist / invalid key / network failure (with CORS explanation)
- ⚠️ Known limitation: the Token Plan gateway does not enable browser CORS (no ACAO header, preflight 401) — **the web version cannot reach it directly; use it inside the Chrome extension** (which bypasses CORS via host_permissions). CORS-enabled endpoints like DeepSeek / Zhipu are unaffected in the web version
### v0.22.2 (2026-09-01)
- **Built-in API config for Hot News Radar** — when the page is opened standalone (e.g. file://) the extension config isn't reachable; the page now has a collapsible "API 配置" card at the top (Base URL / API Key / Model + save/clear) that writes through `AiService.saveConfig` (translate_config / chrome.storage), staying in sync with the popup and every other page; auto-expands as onboarding when unconfigured, and saving auto-retries cards that failed for lack of config
### v0.22.1 (2026-09-01)
- **Fix: hot-list fetch failing in the web version** — the original vvhan aggregation API became unreachable (CORS/down); the Hot News Radar data source was rebuilt as a multi-level fallback chain:
- Primary: **60s per-board hot lists** (Weibo / Zhihu / Douyin / Toutiao / IT之家 / 36Kr, CORS-enabled so both web and extension can fetch directly, with heat values and source links; the upstream rate-limits strictly, so fetching is sequential with 429 backoff retry)
- Secondary: **UApi hot lists** (more boards incl. Baidu / Bilibili / QQ News / Sspai: direct fetch inside the extension; the web version falls back through public CORS proxies)
- Last resort: **60s daily news** (real daily news, CORS-enabled and stable)
- The first source yielding a pool (minimum 20 items) wins; the error state only appears when every source fails, with per-source failures logged to the console
### v0.22.0 (2026-09-01)
- **New module: Hot News Radar (8th tab)** — card-based whole-web hot news monitoring:
- Create multiple hot-news cards, each with its own prompt used by AI for categorization (e.g. "AI & LLM related hot topics")
- Two-stage retrieval: fetch real-time hot lists from Weibo / Zhihu / Baidu / Douyin / Bilibili / Toutiao / IT之家 / 36Kr / Huxi etc. (free aggregation API, no key needed), then the configured LLM filters and ranks the Top 10 per the card's prompt — real data + AI curation, no fabricated news
- Card style mirrors hot-list products: rank color scale (1/2/3), source tags, clickable source links, right-aligned heat values (auto-formatted 万/亿), footer with last-updated time
- Per-card refresh button (re-fetch + AI re-curation, auto retry once on AI errors) plus "refresh all"; cards and results sync across pages via chrome.storage / localStorage
- 8th web tab, 8th Side Panel tab (lazy-loaded), popup header entry button; reuses ai-service.js config sync (usable right after saving the extension config)
- Internal: the hotnews page follows the v0.21 "single coherent stylesheet" architecture (6 themes light/dark adaptive, zero inline handlers, MV3 CSP compliant)
### v0.21.0 (2026-09-01)
- **Task List Dashboard Redesign** — stats-first multi-panel grid layout: 4 KPI stat cards on top (today progress ring / due today / completed / all tasks), quick-add card + task panel card + side sync-guide card; task items gain a priority accent bar, animated checkbox, chip-style metadata and hover lift; filter pills become a segmented control and sync buttons become unified ghost buttons with theme-tinted icons; the whole stylesheet is rewritten as a single coherent layer (old base layer + two UI refinement overlay layers removed), fully following the 6 themes in light & dark
- **English Learning Redesign (both copies unified)** — fixes the root-directory copy's hardcoded orange palette not following themes: all `--el-*` variables now map to global theme variables; provider presets become horizontal chips; the word header card uses the brand gradient (fixes a white-on-light contrast bug); the page CSS of both copies (web / extension) is now identical; the JS fork between the two copies is also resolved (the root copy's inline JS adopts the extension fixes: history rawContent re-rendering, fetch error handling, themed result-card SVG strokes) and the misplaced refinement CSS inside the export template is cleaned up
- **Glass Depth for English Learning** — sections and header now use the same glassmorphism craft as theme.css `.glass-card`: translucent glass background, backdrop blur, top-edge highlight, 4-layer diffuse shadow and gradient glow border, plus page ambient light — visual depth now matches the other pages
- **Unified Container Width** — the task list page container now uses the same `1400px + 24px` standard as the other pages
- **Side Panel Shell Rewrite** — `sidepanel.css` rewritten from 3 stacked overlay layers (581 lines) into a single coherent layer: all `--sp-*` tokens map to global theme variables, fixing "shell colors not following the theme" and a dark-mode scope bug where tabs used a hardcoded `#818CF8`; the header is reduced to glass + a single hairline; the tab bar becomes a single row of icon pills where the active item smoothly expands into a gradient pill with its label (fixes 7 tabs overflowing narrow panels)
- Internal: todolist / english_learning page styles now use a "single coherent stylesheet" architecture (replacing the v0.20 overlay approach); the two copies (root / chrome_extension) keep only the CSP-required differences
### v0.20.0 (2026-09-01)
- **Brand Upgrade** — the product is renamed **AI Tool Box**; all visible titles updated (main UI / extension popup / Side Panel / fullscreen page / sub-page titles, manifest name, context menus, selection tooltip) while internal identifiers stay unchanged for user-data compatibility
- **Title Font Upgrade** — Sora display font (async loaded), refined title typography (weight 800, tracking, clamp-based sizing)
- **Nocturne Theme Redesign** — reworked from "warm near-black violet" to "deep-sea indigo × electric blue" (primary #6E7BFF)
- **Work Report Page Refactor** — buttons/title bar unified to the brand gradient; language bar and summary area recolored
- **Task List Refactor** — gradient title, brand gradient buttons, card-style task items + layered shadows, glowing progress ring, refined empty state, narrow-container responsive breakpoints
- **English Learning Refactor** — the `--el-*` system bridged to global theme variables; dark banner replaced by a primary-color glow band with gradient title; full light/dark adaptation
- **Side Panel Redesign** — header ambient light + gradient logo/title, gradient pill segment control for page switching with narrow-panel overflow fix
- Internal: light/dark grouping now driven by `html[data-mode]`, legacy Catppuccin theme IDs auto-migrate (see PROJECT_HANDOFF §3.4/3.6)
### v0.19.7 – v0.19.8 (2026-08-29)
- Selection tooltip & context-menu detail fixes (v0.19.8)
- English learning page fixes: input-area right border restored when stretched, `resize: vertical` support (v0.19.7)
### v0.19.0 – v0.19.5 (2026-08-28 – 08-29)
- **New minimal premium theme system** — 6 low-saturation Apple-style themes (`lf-*` + `data-mode` light/dark grouping), replacing the 12 Catppuccin themes; legacy IDs auto-migrate
- English learning page: unified inner width with the AI Parse page, dark-theme white-edge fixes, themed scrollbars, pronunciation language selector expanded to 30 languages, pronunciation panel language selection restored
### v0.18.0 (2026-08-28)
- **Side Panel modern minimal redesign** — light gray-white / dark backgrounds, 12px rounded cards, subtle shadows, hairline borders, frosted-glass header, system fonts (Google Fonts dependency removed), dark mode adaptation
- **Config sync fix** — `ai-service.js` `initConfigSync` now also writes `localStorage('translate_config')` so AI Parse / AI Prompts pick up new API settings immediately
### v0.17.0 (2026-08-28)
- **AI Parse & AI Prompts** — integrated from the TaskFlow project
- AI Parse: Classic mode (paste notes → extract a task list) + Analysis mode (requirements → structured summary, email detection)
- AI Prompts: rough requirement → expert-level copyable prompt (with assumptions and usage tips)
- Web version gains two tabs (iframe embedded); Side Panel gains two tabs; popup gains entry buttons
- Shares the LLM API config (Base URL / API Key / Model) with instant sync across all pages
### v0.16.1 (2026-08-28)
- **Fix: theme could not be set in the Side Panel** — added the circular theme switcher button (bottom-right FAB) to the side panel, so all 12 Catppuccin themes can be switched directly inside the side panel and sync to the popup / full-page / every module in real time
### v0.16.0 (2026-08-28)
- **Chrome Side Panel** — A dedicated Chrome 114+ side panel bundling all 5 modules (Smart Translation / Work Report / Task List / English Learning / Email Summary) in one view with one-click tab switching and lazy-loaded iframes
- **Open via popup button** — New Side Panel button in the popup header
- **Keyboard shortcut** — `Alt+Shift+L` opens the side panel from any tab
- **Context menu** — Right-click → "Open LinguaFlow in Side Panel" works on any page
- **Shared architecture** — Reuses the existing sub-pages as iframes, inherits the Catppuccin theme system, and syncs API config (Base URL / API Key / Model) in real time with the popup
- **Requirements** — Chrome 114+ (`minimum_chrome_version`)
### v0.15.0 (2026-08-22)
- **Full-site Visual Refactor (Premium UI)** — High-end visual upgrade across all 5 pages (Translate / Work Report / Task List / English Learning / Email Summary) while fully preserving page structure, content, and business logic
- **Deep layered dark background** — Near-black layered gradient `#0A090D → #161323`, breaking away from flat black or plain solid colors; Glow Mesh ambient washes + subtle noise texture over core visual areas
- **Glassmorphism upgrade** — Cards use semi-transparent base `rgba(255,255,255,0.03~0.04)` + 1px highlight border + `backdrop-blur(30px) saturate(1.6)`, with a soft top-edge light reflection
- **Typography reshaping** — Google Fonts: Syne / Plus Jakarta Sans for headings, Inter for body; gradient title text, larger heading/body contrast, refined tracking & line-height for breathing room
- **Motion & interaction quality** — Fixed the missing `--transition` variable; unified 300ms `ease-out` micro-interactions site-wide; buttons/cards lift slightly (-2~-4px) on hover with edge glow + deepened outer shadow
- **De-finessed details** — Layered soft ambient shadows, unified icon style with consistent spacing & alignment
- **Responsive preserved** — Mobile tab-nav wraps/folds adaptively; latte light themes get a graceful light-glass treatment with no readability loss
- **Dark-theme de-glare** — Per user feedback, reduced overly bright/reflective buttons & cards in dark themes: dark-veil over the washed-out gradient on primary buttons, removed white top highlights, removed hover `brightness` boost (kept slight saturation only), softened the gloss sweep band, and toned down the card top-edge highlight
### v0.14.0 (2026-08-22)
- **Theme system rebuilt as Catppuccin-only (v2)** — Classic 6 themes removed, fully replaced by 12 official Catppuccin themes, default Mocha Blue
- **Latte (light)**: Blue #1e66f5 / Mauve #8839ef / Pink #ea76cb
- **Frappé (muted dark)**: Blue #8caaee / Mauve #ca9ee6 / Green #a6d189
- **Macchiato (dark)**: Blue #8aadf4 / Mauve #c6a0f6 / Teal #8bd5ca
- **Mocha (darkest)**: Blue #89b4fa / Mauve #cba6f7 / Green #a6e3a1
- Each theme an independent color world: base/mantle/crust layered backgrounds, subtext/overlay text ladder, sky/mauve/teal/peach warm-cool contrast accents; glass/shadows/gradients/alpha layers derived from the official palette; star/orb/noise intensity auto-adapted per flavor lightness
- **Texture polish v2** — All 12 themes upgraded: gradient canvas depth (replacing flat background), hue-tinted card shadows (replacing pure black), 5-layer multi-hue ambient washes (with secondary accents), tricolor starfield, hairline border accent segment, button hover gradient flow + glow, header dual glow, themed text selection, glowing focus ring, subtle scanline texture
- **Four-group theme panel** — Latte/Frappé/Macchiato/Mocha sections with scrollable adaptive-height panel; 12 swatches
- **Theme system synced** — theme.css/theme.js updated in both web version and Chrome extension; iframe sub-pages and Popup follow automatically
### v0.13.0 (2026-08-08)
- **Six-Theme Palette Rebuild (Color Worlds v2)** — Every theme reconstructed as an independent color world, retiring the homogenized soft pastels
- Hue-tinted neutrals: backgrounds/text/shadows all shift with the theme hue — no generic gray
- Deepened primaries (pigment feel); secondary accent (accent2) with deliberate hue spacing (e.g. Terracotta × misty slate-blue warm-cool contrast)
- Four-stop title gradients: tail blends into the secondary accent hue — each theme's headline glow is unique
- Six primaries: Celadon #2e9c8b / Ink #8b87f0 / Porcelain #4f63d8 / Terracotta #cc6742 / Sakura #d4698e / Deep Sea #4799e2; swatch palette updated in sync
- **Production Hardening** — Friendly error diagnostics on every AI surface (translation/work report/email summary/extension fullscreen/popup): 401/403 (invalid key), 404 (wrong URL/model), 429 (rate limit/balance), 5xx, CORS/offline all report the problem and a recovery path instead of raw errors
- **Destructive-Action Guards** — History clear-all gains a confirmation dialog; model placeholder corrected to the real `deepseek-chat`
- **Polish Pass** — Translation waiting state replaced with shimmer skeletons (no more blank clear); empty states de-italicized with lifted contrast; popup micro-text 9/10px raised to 11px; hardcoded error color switched to semantic `var(--red)`
- **Accessibility & Motion Restraint** — Button :focus-visible rings, prefers-reduced-motion disables loop animations, unified :active press feedback site-wide
### v0.12.0 (2026-08-07)
- **New Email Summary Module** — Added 5th tab to the web version; Chrome extension popup gains an envelope icon entry (opens in a new tab)
- AI summarizes email threads following the email-thread-summarizer SKILL spec: four-section detailed summary (subject & background / timeline table / technical key points / risks & caveats) plus a responsibility-based To Do List (our side / counterpart / joint verification, P0/P1/P2 priorities)
- Input methods: paste email threads manually, or upload local files (.txt/.md/.eml/.log/.pdf, etc.)
- Summary output language selectable from 30 world languages
- Results support copy and HTML/Markdown file download (fully preserving tables/quotes/task lists)
- Summary history auto-saved (up to 30 entries) with review, Markdown editing, and deletion
- **PDF Email File Support** — Built-in pdf.js 3.11.174 local parsing (both web and extension)
- File size cap 200MB with real-time reading progress
- Automatic handling of huge files: early stop at 120k-character text budget or 500-page cap with explicit annotation, per-page memory release
- Scanned/image-only PDFs auto-detected with OCR hint
- **Oversized Content Auto-Handling** — Content over 60,000 characters automatically keeps the first 60% + last 40%, omitting and marking the middle; the editor retains the full text — no manual splitting needed
- **Translation History Enhancement** — Translation history now displays both source text and translation; clicking a record restores both (web version + extension fullscreen page)
### v0.11.0 (2026-08-07)
- **Multi-Theme System Upgrade** — Added 6 themes with one-click switching (Ocean/Fresh/Dark/Light/Warm/Sakura) via floating button; iframe sub-pages auto-sync parent theme
- **Visual Style Redesign: Vibrant + Textured** — Medium-to-high saturation vibrant palette, glassmorphism cards (translucent + backdrop-blur + top sheen), SVG noise background, triple-layer radial ambient glow, multi-layer depth shadows, hairline gradient borders, gradient title glow
- **Extension & Main Site Unified** — chrome_extension pages aligned with root pages on 20px radius, container widths, and theme texture; popup upgraded with noise background + layered shadows + themed gradient button
- **Work Report HTML Export Fix** — AI summary HTML export now converts from raw Markdown (rawText), matching on-screen rendering exactly with headings/lists/bold support
- **Cross-Environment Consistency** — Markdown preview buttons use "bind when no inline onclick" mechanism, so extension pages behave identically under chrome-extension:// and file://
- **Restrained Animations** — Removed always-on neon animations (gradientShift/borderGlow), kept subtle fadeUp/orb-float effects
- **Popup Form Optimization** — Increased default min-height, fully expanded translation area, more generous default popup size
### v0.10.0 (2026-07-14)
- **Task List Markdown Support** — Added Markdown batch import and export
- Supports standard `- [ ]` / `- [x]` checkbox syntax parsing
- Auto-detects optional markers during import: `@date`, `@time`, `#priority`
- One-click export current task list as Markdown, copied to clipboard
- Auto-deduplication — tasks with same title + date won't be imported twice
- **Apple Reminders One-Click Import** — Click "🍎 提醒" to import reminders instantly
- **URL Scheme channel**: After one-time bridge install, clicking the button triggers AppleScript directly — reminders appear instantly (with macOS notification)
- **AppleScript file fallback**: Simultaneously downloads `.applescript` file as backup, openable in Script Editor
- **One-command bridge install**: Run `./install_url_scheme.sh` to register the `linguaflow-reminders://` protocol
- Auto-creates reminders with due dates and high/medium/low priority mapping
- **Task List UI Overhaul** — Full CSS/HTML refactor with visual upgrades
- CSS variables organized into logical sections with semantic comments
- Enlarged input bar paddings, dark-adapted date/time pickers, custom dropdown arrows
- Unified toolbar button sizing + enhanced colors (cyan ics / blue Google / purple MD / red Apple) with hover glow
- Task card hover micro-animation, completed-state green background, sync badge labels
- Refined filter pills, dark settings panel background, redesigned empty state
- All inline styles extracted to CSS classes
### v0.9.0 (2026-07-09)
- **New English Learning Assistant** — Added "English Learning" module to both web and Chrome extension
- AI Word Study: LLM-powered phonetics, definitions, examples, synonyms/antonyms, and memory tips
- Text-to-Speech: Web Speech API integration with multiple voice selection and speed control
- Learning History: auto-saves study records with individual deletion and clear-all
- Note Export: export today's learning content in Markdown and HTML formats
- Preset API Configs: DeepSeek / OpenAI / Ollama / SiliconFlow / Custom
- **Chrome Extension MV3 CSP Compliance** — Extracted English learning JS to external file `english_learning.js`, resolving MV3 default CSP restriction on inline `