ChatGPTBox-dev/chatGPTBox

ChatGPTBox is a multi-browser extension that aggregates interactions with various AI providers, including ChatGPT, Claude, Bing, and Bard, into a unified interface. It provides context-aware assistance by analyzing content from supported websites like GitHub, YouTube, and Stack Overflow, offering features such as page summarization, text selection tools, and text-to-speech playback. The system manages persistent chat sessions through a dedicated side panel and popup, supporting both standard API endpoints and custom provider configurations.

42 changes.

The project rebranded as ChatGPTBox and executed a major architectural overhaul to Manifest V3, expanding support to multiple AI providers like Bing, Claude, and Bard. This period focused on building a robust multi-provider API layer, enhancing site-specific adapters for platforms such as GitHub and ArXiv, and introducing new UI features like independent chat panels and text-to-speech.

5 changes.

The project expanded its browser extension capabilities by adding site adapters for DuckDuckGo and Brave Search, alongside a new client for the Claude Web API. To support these features and ensure reliability, comprehensive Node.js unit tests were introduced with browser API shims. The period also focused on operational maturity through the implementation of automated scripts for releasing and submitting the extension to major web stores.

A new site adapter for ArXiv has been implemented to automatically extract the paper title, authors, and abstract from the page. This data is passed to a built-in prompt that instructs the research assistant to generate a structured summary focusing on key findings, methodology, and conclusions, ensuring the output maintains an academic tone.

src/content-script/site-adapters/arxiv · high confidence

A new site adapter for Brave Search has been introduced, allowing the extension to initialize and wait for the appropriate search result or sidebar containers on the Brave Search page before proceeding.

src/content-script/site-adapters/brave · high confidence

A new Bard client has been added to the services layer, enabling interaction with Google's Bard service. This implementation handles authentication via the \_\_Secure-1PSID cookie, retrieves necessary request parameters from the Gemini interface, and manages conversation state to send prompts and parse responses.

src/services/clients/bard · high confidence

A new site adapter for Juejin has been added, enabling the content script to automatically extract the article title, author, main content, and up to four comments from Juejin articles. This data is passed to a built-in prompt that asks an AI model to provide a summary of the article, its own opinion, and a summary of the selected comments, with the input text cropped to fit context limits.

src/content-script/site-adapters/juejin · high confidence

The Stack Overflow site adapter now extracts the question title, description, and up to two answers from the page, then formats them into a structured prompt for the AI model. This enables the extension to provide context-aware assistance specifically for Stack Overflow posts by leveraging the existing cropText utility to manage context length.

src/content-script/site-adapters/stackoverflow · high confidence

A new site adapter for WeChat Official Account articles (mp.weixin.qq) has been added to the content script. This adapter extracts the article title, author, and content, and automatically adjusts the layout of the QR code sidebar to improve readability. It then generates a structured prompt for an AI model to analyze and summarize the article, providing the source, summary, main conclusions, and an opinion.

src/content-script/site-adapters/weixin · high confidence

A new site adapter has been added for the Followin platform, enabling automatic content summarization. When a user visits an article or a long tweet on Followin, the extension extracts the title, author, and main content, then generates a concise summary featuring a conclusion and 3 to 5 key points in a markdown list.

src/content-script/site-adapters/followin · high confidence

The application now includes the Cairo font family, enabling proper rendering of Arabic and Latin text. This change adds CSS @font-face declarations for the Cairo font at 400 (normal) and 700 (bold) weights, covering Arabic, Latin, and Latin Extended Unicode ranges to ensure correct display of multilingual content.

src/fonts · high confidence

Introduced a new DeleteButton component that renders a trash icon and implements a two-step confirmation pattern. When the icon is clicked, it hides the icon and displays a 'Confirm' button; clicking 'Confirm' executes the provided callback, while losing focus or clicking 'Confirm' again cancels the action.

src/components/DeleteButton · high confidence

A new ReadButton component has been introduced, allowing users to convert text output into voice and listen to it via the browser's native Speech Synthesis API. The component includes logic to resolve the appropriate speech language based on configuration and to match available system voices, with specific fallbacks for English, Chinese, and Japanese locales.

src/components/ReadButton · high confidence

New scripts have been added to automate the browser extension release process. The \submit-stores.mjs\ script handles the automated submission of build artifacts to the Chrome Web Store, Firefox Add-ons (AMO), and Microsoft Edge Add-ons, requiring specific environment variables for authentication. Supporting this, \create-firefox-sources-zip.mjs\ generates the required source code zip for Firefox AMO review, while \update-firefox-metadata.mjs\ allows for updating Firefox extension metadata and release notes. Additionally, \check-edge-api-key-expiry.mjs\ provides a CLI tool to monitor the expiration of the Microsoft Edge API key, alerting users when renewal is needed.

scripts · high confidence

The extension now includes comprehensive translation files for German (de), Spanish (es), French (fr), and Indonesian (id), alongside updates to the English (en) locale. These changes add user-facing strings for new features such as independent conversation pages, custom API modes, selection tools, and web mode history settings, ensuring the interface is accessible in these languages.

_src/\locales · high confidence

The GitHub site adapter has been updated to detect and process GitHub issue and pull request pages in addition to commit patches. New path-matching logic identifies issue and PR URLs, allowing the adapter to parse thread titles, comments, and draft messages to generate structured summaries for discussions. The adapter also refactors its initialization to monitor URL changes without remounting unnecessarily and improves error handling for patch requests.

src/content-script/site-adapters/github · high confidence

A new Independent Panel page has been added, providing a dedicated interface for managing chat sessions independently of the main popup. This panel allows users to create, switch between, and delete multiple conversation sessions, with the ability to export all conversations as a JSON file or clear them entirely. The UI features a collapsible sidebar for session management and integrates with the existing settings page, offering a more persistent and organized workspace for ongoing chats.

src/pages · high confidence

Added a new \Claude\ client class in \src/services/clients/claude/index.mjs\ that enables interaction with the Claude Web API. This client requires a session key in the \sk-ant-sid02-\*\ format, allowing users to authenticate and send messages via the unofficial API. It supports proxy configuration, custom fetch implementations, and provides methods to manage conversations and retrieve model information.

src/services/clients/claude · high confidence

The popup settings now include dedicated sections for managing API Modes (allowing custom provider endpoints and API key overrides), Selection Tools (enabling users to create, edit, and activate custom prompt tools for text selection), and Advanced settings (providing controls for max response/conversation lengths, optional provider temperature overrides, custom API URLs for ChatGPT/OpenAI/Anthropic, context menu visibility, and web mode history).

src/popup/sections · high confidence

A new ReadButton component has been added to the library, enabling users to convert GPT text output into voice and read it aloud. This component is now exported from the main components index, making it available for use in applications.

src/components · high confidence

The content script now exposes a set of configurable menu tools that allow users to interact with the extension directly from the browser context menu. Users can now summarize the current page, open a new conversation in an independent popup window (with logic to reuse existing windows based on user configuration), open the Chromium side panel, or close all active chats on the current page. These actions are defined in the new \src/content-script/menu-tools/index.mjs\ module, which handles the specific browser API calls (such as \chrome.sidePanel.open\ with appropriate error handling for browsers like Firefox) and message passing to background scripts.

src/content-script/menu-tools · high confidence

This change introduces a suite of new React hooks in src/hooks to handle application state and environment detection. The useConfig hook now listens to browser storage changes to keep the local configuration in sync, while useTheme combines this config with the system's color scheme preference to determine the active theme. Additionally, useWindowSize and useClampWindowSize provide reactive window dimension tracking with optional boundary constraints, and useWindowTheme exposes the current system color scheme directly.

src/hooks · high confidence

The selection tools menu now includes new options: 'Explain' for simple content breakdowns, 'Code Explain' for code analysis, and specific translation tools for English, Chinese, and bidirectional translation. Existing tools like Translate, Summary, Polish, Sentiment Analysis, and Ask have updated prompts for better results, and the import path for language configuration has been updated.

src/content-script/selection-tools · high confidence

The background API module for \src/background/apis\ has removed the implementation files for the legacy ChatGPT Web interface (\chatgpt-web.mjs\) and the OpenAI API (\openai-api.mjs\). This change eliminates the code paths that previously handled streaming responses and session management for these specific backend integrations, effectively stopping support for generating answers via these legacy endpoints.

src/background/apis · high confidence

The API service layer has been restructured from a monolithic implementation into distinct, modular files for each provider (Azure OpenAI, Bard, Bing, ChatGPT Web, Claude API/Web, Custom, Moonshot, OpenAI, and OpenAI-compatible core). This change introduces a shared core for OpenAI-compatible requests, a provider registry for built-in and custom providers, and dedicated token and temperature parameter handling. Users benefit from improved stability, better support for diverse LLM providers (including Ollama, Mistral, xAI, and NVIDIA NIM), and more robust error handling and streaming logic across all API integrations.

src/services/apis · high confidence

The background script has been restructured to improve stability and security. A new \redact.mjs\ module now automatically strips sensitive fields (such as API keys, tokens, and passwords) from messages logged in the background, preventing accidental data exposure. Error handling has been hardened across keyboard commands (\commands.mjs\) and context menus (\menus.mjs\) to catch synchronous exceptions and observe asynchronous promises, ensuring that failures in tab queries or message sending do not cause unhandled rejections. Additionally, proxy generation state management (\proxy-generation-state.mjs\) has been introduced to better track and interrupt streaming sessions, while the main entry point (\index.mjs\) has been expanded to support a wider range of AI providers and configuration options.

src/background · high confidence

The Bing client has been upgraded to improve error reporting and add support for image generation. The update includes a new BingImageCreator class for generating images via Bing's API, integrated into the main BingAIClient through a genImage feature flag. Error handling has been enhanced with better WebSocket error messages and improved fetch request error reporting. The client now uses updated browser headers and connection parameters to maintain compatibility with Bing's services.

src/services/clients/bing · high confidence

The build system has been restructured to significantly improve build speed and flexibility. It now supports parallel builds of multiple extension variants (Chromium, Firefox, and minimal builds) controlled by the BUILD\_PARALLEL environment variable. Compilation uses esbuild for faster minification and enables filesystem caching with configurable compression (gzip/brotli) via BUILD\_CACHE\_COMPRESSION. Developers can tune thread-loader parallelism using BUILD\_THREAD and BUILD\_THREAD\_WORKERS, and the build now supports symlink resolution for workspace development via BUILD\_RESOLVE\_SYMLINKS.

(repo-wide) · high confidence

The content script has been significantly refactored to improve stability and user experience. The main entry point now supports a new 'always floating sidebar' mode, allowing the chat interface to remain visible and dockable regardless of the page context, while also implementing retry logic for element detection to handle dynamic pages better. Existing containers are now properly unmounted before re-rendering to prevent duplicate instances. Additionally, the UI styling has been updated with new CSS variables for toolbars and code blocks, improved scrollbar customization, and a font-family override for the chat container to ensure consistent rendering across different websites.

src/content-script · high confidence

The ConversationItem component has been refactored to support new user interactions and localization. Users can now retry failed or incomplete responses via a new Sync icon button, and answer text can be read aloud using the newly integrated ReadButton. The component also supports custom model names (descName) and wraps its output in a consistent 'chatgptbox-' CSS class prefix. Additionally, all visible UI text, including tooltips and labels, is now internationalized via react-i18next, and the component is memoized for improved render performance.

src/components/ConversationItem · high confidence

The DecisionCard component now supports internationalization by replacing hardcoded strings like 'Ask ChatGPT' and 'No Input Found' with translation keys via the useTranslation hook. Additionally, the component's CSS class names have been updated from 'sidebar-free' to 'chatgptbox-sidebar-free' to prevent conflicts with other websites, and the internal structure of the button and status text elements has been adjusted to wrap icons and text in a span for better styling.

src/components/DecisionCard · high confidence

The Markdown renderer now supports reasoning models by displaying collapsible 'Thinking Content' sections with expand/collapse functionality. Code blocks have been upgraded to include a font-size selector (10px–18px) and a copy button, replacing the previous simple copy-only behavior. Links to chatgpt.com, claude.ai, kimi.moonshot.cn, and kimi.com now trigger a browser notification and tab-jump action instead of opening in a new tab. The renderer also supports RTL text direction and uses a local KaTeX stylesheet (version 0.16.9) for math rendering.

src/components/MarkdownRender · high confidence

The content-script site adapters now include new integrations for Stack Overflow, Juejin, WeChat (mp.weixin.qq), Followin, and ArXiv, alongside updated configurations for existing sites like Reddit, GitLab, YouTube, GitHub, and Zhihu to improve compatibility. Search engine adapters for Google, DuckDuckGo, Kagi, Naver, Brave, and Presearch have been updated with new CSS selectors for input fields, sidebars, and result containers to handle recent UI changes, with DuckDuckGo and Brave also receiving specific initialization actions to handle rendering delays.

src/content-script/site-adapters · high confidence

The FloatingToolbar component has been significantly refactored to enhance stability and user control. It now supports a 'pin' feature that docks the window to a fixed position, preventing it from moving during scrolling, and includes logic to automatically correct position when the toolbar appears outside the viewport. The component also introduces support for custom selection tools, allowing users to define and activate their own tools alongside the default set. Under the hood, the implementation switches from browser storage listeners to a centralized config hook, improves mobile selection handling, and ensures proper cleanup of DOM nodes and timers to prevent memory leaks.

src/components/FloatingToolbar · high confidence

The GitLab site adapter now detects whether the user is viewing a specific file (blob) or a commit diff and applies a tailored prompt for each context. When viewing a file, the adapter fetches the raw content and asks the AI to analyze the code's purpose, functionality, and potential issues. When viewing a commit, it retains the previous behavior of generating a commit message and summarizing the patch changes, but with an enhanced prompt structure.

src/content-script/site-adapters/gitlab · high confidence

The Bilibili adapter now waits for the avatar element to exist before mounting to prevent rendering issues, excludes bangumi pages, and tracks video path changes instead of full URL changes. Subtitle retrieval has been updated to use the WBI-protected API endpoint, with robust handling for protocol-relative URLs and filtering of null content to prevent 'null' strings in summaries. The prompt for video summarization has been enhanced to request comprehensive markdown summaries with key takeaways. Additionally, the Quora adapter's prompt has been refined to better structure question and answer summaries.

src/content-script/site-adapters/bilibili · high confidence

The extension now extracts page content more reliably by using site-specific CSS selectors (for Google, Bing, Wikipedia, CSDN, FAZ, Golem, EETimes, and new.qq.com) before falling back to Mozilla Readability or a largest-element heuristic. SSE streaming is more robust, handling arbitrary chunk boundaries, non-JSON responses, and aborts, while fetch errors are classified with codes (FETCH\_REQUEST\_FAILED, FETCH\_RESPONSE\_STREAM\_FAILED, INVALID\_API\_ENDPOINT) and the request origin. Crop-text respects a new cropText setting and adapts its length limit to the model's context size (e.g., 8k, 32k) using maxResponseTokenLength. API mode handling is standardized with normalization, canonical model names, and lineage preservation across migrations. Additional utilities include background fetch via the service worker, element position helpers, mobile detection using userAgentData, and JWT generation via @noble/hashes.

src/utils · high confidence

The ConversationCard component now handles interrupted or disconnected sessions more robustly by finalizing partial answers and restoring retry records when a connection drops, ensuring no user input is lost. It also introduces generation ID tracking to prevent race conditions where old responses overwrite newer ones, and disposes of owned transports on unmount to prevent memory leaks. Additionally, the component supports dynamic window sizing via a clamp hook, allows user input while a response is being processed, and integrates with the new API modes system to display the correct model name and label. The CopyButton component has been updated to support internationalization for its tooltip text.

src/components/ConversationCard · high confidence

The extension has been rebranded as ChatGPTBox and upgraded to version 2.7.1, significantly expanding its capabilities beyond its original scope. It now supports multiple AI providers including Bing AI, ChatGPT, Claude, Google Bard, and Kimi, facilitated by new declarative net request rules in rules.json that correctly set origin and referer headers for these services. The architecture has shifted to Manifest V3, introducing a dedicated side panel, keyboard shortcuts for actions like creating new chats or summarizing pages, and broader host permissions for sites like bing.com, google.com, and claude.ai. Additionally, the legacy config.mjs module has been removed, and content scripts now execute on HTTP and file URLs to support intranet usage.

src · high confidence

The application now uses a new configuration structure in src/config that defines model keys for various providers (OpenAI, Anthropic, Moonshot, Mistral, DeepSeek, NVIDIA NIM, OpenRouter, AIML, ChatGLM, Ollama, Google, xAI) and includes legacy model key migrations to ensure backward compatibility. The system also provides provider mappings for API keys and language data support with BCP 47 normalization.

src/config · high confidence

The popup interface has been redesigned to support two distinct display modes: a compact popup mode (460x560px) and a larger page mode (min 460x560px, full height/width), with the footer now fixed to the bottom and tabs slightly enlarged. Additionally, users can now configure the extension icon's click action to either open the popup or trigger a custom action (such as a context menu tool) when the popup area is too small, with the logic handled in the new entry point script.

src/popup · high confidence

The Safari extension build process has been updated to use the 'ChatGPTBox' name and bundle identifier (dev.josStorer.chatGPTBox) instead of the previous 'chatGPT-for-Search-Engine' naming. The build script now explicitly sets the app category to 'utilities' and configures architecture support for both arm64 and x86\_64. Additionally, the marketing version is reset to 0.0.0, and the patch file previously used to apply these settings has been removed as the changes are now integrated directly into the build configuration.

safari · high confidence

The session initialization logic has been extracted into a dedicated module, introducing a unified session structure that supports multiple AI providers including ChatGPT web, Bing, Bard, Claude, and Kimi. Local session storage operations (create, delete, update, get) are now centralized in a new service file, which also handles legacy model key migrations to ensure backward compatibility. Additionally, the API access wrapper for ChatGPT has been updated to fetch cookies from chatgpt.com instead of chat.openai.com, and includes a guard to prevent errors when browser cookies are unavailable.

src/services · high confidence

The InputBox component now supports manual resizing via a draggable handle and keyboard shortcuts (Arrow keys, Home, End), allowing users to adjust the input area height within defined limits. It also introduces logic to prevent accidental submission during IME composition and adds a configuration option to control whether the input box automatically focuses after an answer is received.

src/components/InputBox · high confidence

The Reddit site adapter now uses updated CSS selectors to extract post titles, descriptions, and comments, reflecting changes in Reddit's DOM structure. The internal prompt has also been refined to provide a more structured input for the AI model, explicitly labeling the title, description, and comments sections.

src/content-script/site-adapters/reddit · high confidence

The YouTube adapter now correctly fetches video subtitles by handling YouTube's new 'pot' parameter requirement, ensuring the video summarization feature works again. Additionally, the prompt used to generate summaries has been updated to request comprehensive, markdown-formatted outputs with key takeaways, and HTML entities in subtitle text are now properly decoded for better readability.

src/content-script/site-adapters/youtube · high confidence

The Zhihu content-script adapter has been updated to generate more detailed and structured summaries. It now distinguishes between single-answer pages and question pages with multiple answers, collecting up to four answers for the latter to provide a comprehensive overview. Additionally, the internal prompts have been rewritten in English to improve clarity and consistency across different content types, such as articles and Q&A discussions.

src/content-script/site-adapters/zhihu · high confidence

The DuckDuckGo site adapter now waits for the search results container to appear before initializing, preventing mounting failures caused by the site's initial rendering delay. This ensures the extension correctly detects when the page is ready to be modified.

src/content-script/site-adapters/duckduckgo · high confidence

Added a comprehensive suite of Node.js unit tests for the extension's background scripts, content-script components, and core utilities. To enable running browser-extension code in a Node environment, the diff introduces a minimal browser API shim (mocking chrome.runtime, storage, tabs, and windows) and a set of ESM loader hooks that intercept module resolution to provide stubs for React, Preact, i18n, and other dependencies. These test infrastructure files support new unit tests covering command registration, proxy generation state management, sensitive data redaction, conversation card lifecycle and session handling, and floating toolbar stale task cleanup.

tests · high confidence

The project has upgraded its dependency tree, including major bumps for jsdom (21 to 28), webpack (5.75 to 5.92), sass (1.58 to 1.91), and uuid (9 to 11), alongside numerous minor updates to libraries like preact, react-bootstrap-icons, and eslint. The Node.js engine requirement has been raised to version 22 or higher, and new build tools such as esbuild and c8 for test coverage have been added to the development environment.

(dependencies) · high confidence

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

• CAI 42 → 42 (+0.6)
• Rubric changed (rubric-2026.09.12 → rubric-2026.09.18) — scores are not directly comparable.

• Code Health 38 → 40 (+1.2)
• Architecture 69 → 67 (-2.5)
• Maturity 62 → 61 (-1.6)
• Readiness 52 → 53 (+0.3)
• Security 82 → 81 (-1.5)
• Accessibility 33 → 33 (+0.0)
• Performance 85 (new)

• Dependency hygiene PARTLY measured — npm pinning read, dependency currency not (no pnpm-resolved versions to grade)
• Documentation: no architecture or design documentation (README.md)
• High CVE: [GHSA redacted] (package-lock.json)
• Hotspot: build.mjs (build.mjs)
• Hotspot: src/background/index.mjs (src/background/index.mjs)
• Hotspot: src/content-script/index.jsx (src/content-script/index.jsx)
• Hotspot: src/popup/sections/GeneralPart.jsx (src/popup/sections/GeneralPart.jsx)
• Hotspot: src/popup/sections/api-modes-provider-utils.mjs (src/popup/sections/api-modes-provider-utils.mjs)
• Hotspot: src/utils/model-name-convert.mjs (src/utils/model-name-convert.mjs)
• Off-boarding risk: anonymized user #1

• FunctionTooLong: index.Browser.runtime.onMessage.addListener() (src/background/index.mjs)
• High CVE: [CVE redacted] (package-lock.json)
• High CVE: [GHSA redacted] (package-lock.json)
• High CVE: [GHSA redacted] (package-lock.json)
• High CVE: [GHSA redacted] (package-lock.json)
• High CVE: [GHSA redacted] (package-lock.json)
• Medium CVE: [GHSA redacted] (package-lock.json)
• Off-boarding risk: anonymized user #1
• Orphaned files with no living knowledge
• Outdated (npm): @babel/runtime
• Outdated (npm): @noble/hashes
• Outdated (npm): @picocss/pico
• Outdated (npm): @primer/octicons-react
• Outdated (npm): countries-list
• Outdated (npm): i18next
• Outdated (npm): js-sha3
• Outdated (npm): katex
• Outdated (npm): parse5
• Outdated (npm): preact
• Outdated (npm): react
• …and 16 more

• 14 commits — 14 feature/other, 0 fixes

• badges/coverage.json — 6 commits
• tests/unit — 3 commits
• .github/workflows — 2 commits
• tests/setup — 2 commits
• src/services — 1 commit

• change: Add Anthropic Claude Opus 5.5 API support (#1089)
• change: Add Anthropic Claude Sonnet 5.5 API support (#1097)
• change: Add OpenAI GPT-6 Sol and GPT-6 Luna model (#1090)
• change: Add OpenAI GPT-6.1 Sol support (#1098)
• change: Dispose conversation-owned transports on unmount (#1083)
• change: Update coverage badge [skip ci]
• change: Update coverage badge [skip ci]
• change: Update coverage badge [skip ci]
• change: Update coverage badge [skip ci]
• change: Update coverage badge [skip ci]
• change: Update coverage badge [skip ci]
• change: Update pullfrog CI Action from v0.1.81 to v0.1.84 (#1092)
• change: Verify floating toolbar subscription cleanup with real hooks (#1080)
• change: chore(deps): bump pullfrog/pullfrog from 0.1.78 to 0.1.81 (#1088)

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

ChatGPTBox-dev/chatGPTBox was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.