macOS support, searchable conversations and custom skills

New features

Added macOS support for Apple Silicon.

Conversation storage is now persistent, and conversation search is supported.

Enhanced Skills with custom skill installation, plus enable/disable and delete controls.

Updated built-in Skills and added light/dark theme switching.

Bug fixes

Fixed long pasted text not auto-resizing in the input box.

Added a retry button when the model list fails to load, preventing the app from becoming unusable.

Fixed login popups being blocked by security software.

Input, workspace history and model routing improvements

Improvements and bug fixes

The input box now auto-resizes from a minimum of two lines up to a maximum of fourteen lines.

Invalid workspaces can still show historical conversations, together with a clear invalid-workspace notice.

Optimized login validation so conversations are not created before sign-in.

Historical file lists now support deletion, and main-flow tracking events were added.

Membership display now supports Ultra, and large/small screen adaptation was improved.

Fixed style corruption caused by XSS issues and optimized error log information.

Improved Gemini routing via Google with fallback to SkyRouter, upgraded gemini-3-pro-preview from 3 to 3.1, and added Skycode hot updates.

Standardized cost logs to distinguish Skybot and desktop versions, and updated the website page for the WSL2 version.

Feedback reporting, region handling and model updates

New features and fixes

Users can now submit bug feedback, including both virtual machine and frontend logs.

On startup, the app checks the user's region, shows a prompt for Mainland China users and disables unsupported functionality before backend validation is skipped.

Fixed “application not found” errors when opening certain file formats.

Added the Opus 4.6 model.

Improved tolerance around error codes during virtual machine acceleration installation, making the installer more robust.

Installer hardening, clearer errors and regional prompts

New features and fixes

The installer now includes virtual machine acceleration installation and packaged builds are signed.

Virtual machine startup failures now show specific error messages to make diagnosis easier.

Fixed line breaks not working in the middle of input text.

Fixed user-entered Markdown code being rendered as Markdown and AI-generated CSS interfering with the real project page styles.

The window title now displays the version number, and unsupported-region prompts were added for domestic users.

Health checks no longer go through the proxy.

Credit details, in-app subscription and purchase, and Markdown viewing

New features

New credit details dialog: view your credit usage details, with clearer spending records and balance changes.

In-app membership subscription and credit purchase: subscribe to membership and buy credits directly in the app for a smoother flow.

Improvements

Service access optimization: improved the API signing mechanism for more standardized and stable service access.

Better Markdown viewing: MD files can switch between preview and source, making reading and editing easier.

Bug fixes

Fixed DeepSeek-v4-flash issue: resolved request errors with DeepSeek-v4-flash in some scenarios for more stable model usage.

Personalized themes, instant skill installation and reusable workflows

New features

Personalized themes: Choose from multiple built-in themes, or upload an image and customize your own theme through conversation to match your preferences.

Skills take effect immediately after installation: Download and install skills from the Skill Marketplace during a conversation, then use them right away.

Prompts to save frequent workflows: When Skywork detects a repeated workflow, it suggests saving it as a skill for one-click reuse.

Improvements

Automatic updates for system skills: Built-in skills now stay up to date automatically, so you continuously benefit from improved task handling.

Improved skill discovery: The Agent can match the right skills more accurately, improving its understanding and output for complex tasks.

Improved interruption experience: Interrupting a conversation no longer collapses the current content, so you can keep reviewing the context while adding new instructions.

Improved new-task experience: The summary panel is expanded by default when creating a task, making task details and execution status easier to see.

Task quick start, next-step suggestions and a better input experience

New features

New task quick-start entry: New users can quickly explore supported scenarios and task examples, making it easier to get started.
Next-step suggestions: After an Agent finishes, recommended questions appear above the input box. Click one to send it instantly and keep the task moving.

Improvements

Improved payment guidance: When your credits may not be enough to complete the next task, Skywork warns you in advance to reduce interruptions during execution.
Summary information moved into the right sidebar: Task results, key information and related content are now available in one place.
Improved message input: Rich-text list editing is smoother, attachments can be mixed with text, and image files show a hover preview for easier complex input.

Fixes

Fixed conversation auto-naming: Resolved an issue that could cause incorrect automatic conversation names in rich-text scenarios.

Conversation controls, context management, and richer input

New features

• Choose how to open links from the context menu: links in Agent messages now support right-click open options for more flexible access to external content.
• Enhanced conversation pinning: conversations inside pinned projects can now also be pinned, making important projects and key conversations faster to access.
• Copy and specify conversation IDs: copy a conversation ID and specify it when starting a new conversation to continue with existing context.
• Automatic conversation naming: conversations can now be named automatically for clearer, more readable titles.
• Quick navigation to previous questions: quickly locate earlier questions within a conversation, making long discussions easier to navigate.
• Rich-text composer: the message composer has been rebuilt with Tiptap for a richer text input experience.
• Context status and compaction: the Composer now shows context status and supports manual input or compact to compress the current conversation context.
• Slash Commands: enter /review, /compact, /feedback, /goal, or /plan in the composer to quickly open the corresponding feature.
• Conversation feedback shortcut: enter /feedback in the conversation composer to open the feedback dialog for the current conversation.
• Skywork image-generation Skill: use the Skywork image-generation Skill directly for more powerful image creation.

Improvements

• Improved left project list: refined expand and collapse behavior and added drag-and-drop sorting.
• Improved Subagent display logic: refined how subagent content is presented for a clearer multi-agent collaboration process.
• Experience improvements and bug fixes: refined additional interaction details and fixed several known issues.

Multi-tab right sidebar, message-based follow-ups, and system notifications

New features

Multi-tab right sidebar: the right sidebar has been rebuilt to support opening review, file preview, and side chat at the same time, making it easier to switch context during complex tasks.

Continue from Agent messages: select an Agent message and add it to the main conversation or side chat, then keep asking, editing, or expanding around that specific content.

Subagent chat visibility: view subagent chat processes and context to better understand task ownership and execution during multi-agent collaboration.

System notifications added: key states such as conversation completion, authorization, and Q&A prompts now notify you through system notifications, making task progress easier to track in time.

Improvements

Message-stream card styling optimized: artifact cards and review cards have been refined so task results, code reviews, and related information are clearer to read.

User messages now collapse automatically: longer user messages are collapsed by default and can be expanded manually, reducing the impact of long text on conversation reading.

Left menu actions are easier to access: archive and pin shortcuts are now exposed in the conversation list, making common actions easier to find and use.

Icon visuals unified: icon design rules across the project have been standardized and refined for a more consistent interface.

More accurate summaries: summary sources now filter out local search tool content, reducing noise caused by repeated web searches.

More stable artifact display: generated documents, PDFs, PPTs, and other outputs can be displayed and opened more reliably, reducing cases where a file was generated but not visible.

Fixes

Fixed Claude subagent creation: resolved an issue where subagents could not be created normally in some scenarios when using Claude models.

Fixed conversation ownership issues: resolved cases where conversations could be assigned to the wrong project, preventing conversations from appearing in the wrong place.

Fixed artifact cards disappearing after restart: artifact cards now remain available after restarting the app, improving stability when reviewing historical results.

Sub-agent visibility, artifact preview cards, and a better summary experience

New features

Sub-agent information display: the message stream and summary panel now show sub-agent-related content, helping you understand multi-agent collaboration and task results.

Artifact preview cards in the message stream: generated files, images, documents, and other artifacts can now be previewed quickly in the message stream, with support for opening them in the app or with your system default application.

Improvements

Left menu interaction optimized: improved hover layer expansion speed and secondary-menu operation experience, reducing accidental closes and making interactions more stable.

Summary panel UI rebuilt: the summary panel now groups content by project type, making the information hierarchy clearer and task results easier to review.

Main chat UI adjusted: improved layout and reading experience in the main chat, making long conversations and task processes easier to browse.

Fixes

Fixed image message card misalignment: fixed an issue where image message cards could appear misaligned in the message stream while the Agent was processing, improving message display stability.

Skywork Desktop 2.0 — rebuilt for local, sandboxed agent work

New features

A new local sandbox execution model replaces the previous virtual machine workflow, so agents can work closer to your real development environment while keeping operations controlled by sandbox permissions.

A fully redesigned interface and interaction model makes tasks clearer, smoother and easier to follow from planning through completion.

Coding capabilities are significantly upgraded: import local projects and let the agent read code, edit files, inspect diffs, run commands and assist with Git submissions.

Worktree and branch management lets different features, bug fixes or experiments move forward in isolated workspaces, then review and merge changes safely.

Automation tasks are now available for scheduled work such as routine checks, report generation, information syncs and project follow-ups.

Improvements

Conversation management now supports project-based organization, making related tasks and context easier to find, with pinning for important projects and frequent conversations.

Local machine and MCP configuration has been expanded, enabling agents to connect with more external tools, data sources and private systems.

Task execution, project collaboration and multi-tool calling were optimized for better reliability and control on complex work.

Breaking changes

Because 2.0 moves to a new underlying architecture, historical conversations from earlier versions will not be retained after upgrading.

Local folders, memory and smoother browser automation

New features

Take over local folders for batch reading and writing while keeping data within your own environment for better privacy.

Added Gemini-3.5-flash for faster responses on lightweight tasks.

All members can now freely choose from all available models.

Memory is now available, so Skywork can remember your habits and preferences without repeated setup.

Browser Use can automatically click, fill and operate websites with a smoother end-to-end flow.

You can now interrupt and add instructions while generation is in progress.

Improvements and bug fixes

Improved low-credit prompts with one-click top-up.

App hot updates make new features available without reinstalling.

Fixed feedback submission after runtime environment exceptions, message failures after reconnecting, and context handling issues in long conversations.

Built-in skills and better file-based prompting

New features

Curated high-quality skills are now built in, including skywork-video, skywork-audio, frontend-design and web-design-guidelines, greatly improving out-of-the-box video, audio and web generation quality.

HTML preview is supported, so generated webpages can be viewed directly.

Added skill shortcuts: type “/” in the input box to bring up skills.

Improvements and bug fixes

Improved message composition by showing attached local files.

Tasks can now be created with instructions only, files only, or a combination of instructions and files.

Fixed an issue that prevented selecting the same file across multiple conversations.