Product Changelog

A reverse-chronological record of the value each AuraBaba product update brings to users.

Each entry is written from the user's point of view. Updates are listed newest first, with July 14 as the first official release.

What this means for you

WeChat version is checked before you install the skill: Step 1 (pre-check) of the install prompt now lists the supported WeChat versions — macOS: WeChat for Mac ≤ 4.1.8.100 (macOS ≥ 26.3.1; newer WeChat versions may not be compatible); Windows: WeChat 4.x (xwechat data directory), 3.x and earlier are not supported; Linux: WeChat for Linux 4.x. If your version doesn't match, the agent explains and guides you to a supported version — no more "installed but can't read the chats".
Windows portable build is auto-verified on a real machine before release: Every new portable zip is smoke-tested on a real Windows machine (download → unzip → launch → version/help commands → graceful error when WeChat isn't installed) before it ships, and the portable zip has been rebuilt. The download-and-unzip install path is now much more dependable.
Login and registration pages get a visual refresh: Cleaner colors and typography, a more prominent brand panel, and dark-mode support. Login errors now appear in a highlighted alert box (friendlier to screen readers), input fields show clearer focus and hover states, password fields support browser autofill, and keyboard navigation feels smoother.

What this means for you

Windows aura CLI now auto-fixes environment issues: The usual Windows CLI pitfalls are now handled for you — ① app execution aliases from the Microsoft Store (fake commands that fail with access errors) are skipped and the CLI keeps searching for a real agent install; ② installs that aren't on PATH are still discovered (e.g. npm global folders or installer-registered apps like OpenClacky), and their directories are auto-appended to your user PATH and refreshed, so new terminals and the background daemon find your agent right away — no manual environment editing; ③ when no agent CLI is found, the CLI names the Store aliases it detected and points at the real install (e.g. npm install -g @openai/codex); ④ aura setup verification refreshes PATH from the registry, verifies the real binary and repairs your user PATH; ⑤ aura login --token now rejects API keys (aura_sk-...) up front with browser-login guidance instead of a confusing "Could not sign in".
Clear guidance when security software blocks the Windows install: The installer now verifies that aura.exe actually runs — if Windows Defender or another security product false-positives the unsigned CLI as "contains a virus or potentially unwanted software", you get step-by-step remediation (restore it from Windows Security → Protection history, or add an exclusion) instead of an install that "succeeds" and then fails mysteriously. The Windows setup menu also no longer offers the WSL-only OpenClacky installer, so you won't install something that can't work.
Smoother WeChat Chat Records skill setup: If wechat-cli is already installed on the machine, setup reuses it and skips all download/install steps (no duplicate installs when several workspaces share one host). On macOS, when WeChat needs to be re-signed, the impact is explained first (local signature change, possible effect on automatic updates, no data upload) before continuing; after granting permissions such as Full Disk Access, setup resumes instead of getting stuck. When a setup step needs you to act, the install dialog links straight to the task where the agent guides you.
Agent-generated document links open the document directly: Clicking a project document link generated by an agent now opens that specific document (previously it only opened the document list page).

2026-08-17 — Even Simpler WeChat Chat Records Skill Setup (Portable Windows Build), Skill Uninstall & Reinstall in the Marketplace, More Reliable Built-in Skill Installs

What this means for you

WeChat chat records skill setup is even simpler: The wechat-cli tool the skill needs is now downloaded automatically as an official build — macOS / Linux users download and run it directly, while Windows users get a portable zip with an embedded Python runtime (download, unzip, and it just works — no environment setup to figure out). Setup also bootstraps the Node / Python runtimes for you, and users in mainland China can install through npm / PyPI mirrors, so a GitHub-free network environment is no longer a blocker.
Uninstall and reinstall skills from the marketplace: Installed skills can now be uninstalled in one click (removes the skill and its files from your team skills and unbinds any linked agents; uninstall is irreversible, but you can reinstall whenever you need it) or reinstalled in one click — the latest marketplace version is fetched automatically and overwrites the current content, so a broken local install or an outdated version no longer needs manual handling.
More reliable built-in skill installs: Fixed an issue where some built-in skills (such as the Feishu lark-cli skill) failed to install — their install URLs pointed at a private repository that couldn't be resolved publicly. They now install straight from the bundled resources, so installs succeed more reliably.

2026-08-16 — One-Click Feishu Skill Setup, DingTalk & WeChat Chat Records Skills Renamed with Official Logos

What this means for you

One-click Feishu skill setup: Installing the Feishu (lark-cli) skill no longer requires you to manually create an Open Platform app and provide App ID / App Secret — the skill now runs the official one-click registration flow, which prints a verification URL and QR code. Confirm in Feishu/Lark and the app is created automatically with its credentials saved, so setup takes less effort.
DingTalk skill renamed with the official logo: The "DingTalk CLI" skill in the marketplace is now simply "钉钉" and shows the official DingTalk logo, displayed consistently across all interface languages — easier to recognize when browsing the marketplace.
WeChat chat records skill renamed with the official icon: The WeChat chat analysis skill in the marketplace is now "微信聊天记录" with the official WeChat icon — easier to find and install, alongside the friendlier setup guidance from the August 15 update.

What this means for you

Upload CSV files to project documents and preview them as tables: Project documents now accept CSV uploads and render them directly as interactive HTML tables with a highlighted header row — no need to download and open them in a spreadsheet app. Parsing follows RFC 4180, so quoted fields, escaped quotes, embedded newlines and CRLF files are all handled correctly.
Shared document links now preview every file format: Previously only Markdown and HTML rendered correctly through public shared links — Excel, Word, PowerPoint, CSV, images, PDFs and media opened as blank or garbled pages. Now shared links preview all formats in the browser (Office documents, CSV, images, PDF, video and audio), with a download fallback for anything unsupported.
New marketplace skill: DingTalk CLI: The new dingtalk-cli (dtcli/dws) skill covers 19 DingTalk business domains — contacts, chat & robots, calendar, todos, approvals, attendance, DING messages, reports, AI tables, docs, drive, AI minutes, mail, sheets, knowledge base, AI search, live streaming and more — plus raw OpenAPI calls and event subscriptions. Install via a one-line curl/PowerShell command, npm/Homebrew builds or the Gitee mirror; log in with a browser or device-code flow, with multi-organization profiles.
Full DingTalk AI minutes (听记) support: The DingTalk CLI skill covers AI minutes end-to-end — control live recording (start/pause/resume/stop), upload local audio to generate transcripts, browse transcripts with pagination, AI summaries, auto-extracted todos, keywords, audio downloads, mind maps, speaker renaming and hot words. Meeting recording cleanup can be handed to your agent.
New marketplace skill: Teams CLI: The new teams-cli skill installs the Microsoft Teams CLI (Node.js ≥ 20 required) and completes the Microsoft sign-in (interactive browser flow locally, device-code flow in remote sessions), so your agent can verify login status, switch accounts and list installed apps.
New marketplace skill: Teams meeting transcript & task dispatch: After a meeting, your agent can read the Teams meeting transcript (from a transcript file or via Microsoft Graph API), parse the action items, and automatically create task cards in AuraBaba assigned to the right members or agents — meeting follow-ups handled in one step.
WeChat chat records skill setup is friendlier: The WeChat chat records skill in the marketplace now guides installation and initialization by your system (macOS / Windows / Linux) — macOS prompts for full disk access permission and sudo initialization, Windows for an administrator terminal, Linux for root privileges, with step-by-step instructions for choosing between multiple accounts and verifying the first-time initialization. Windows and Intel Mac users no longer have to figure out pip install paths on their own.
Uploaded attachments no longer vanish silently: Fixed an issue where files (e.g. PPT) uploaded in comments, replies, issue descriptions or the create-issue flow could disappear with no explanation. Upload failures now surface a clear reason, and submitting while an upload is in flight is blocked, so you never end up with a broken attachment link.
Invitation status is now accurate: A freshly sent invitation is never reported as "expired"; genuinely past-due invitations clearly show as expired on the accept page; and the invitation email now states how long the link stays valid ("expires in N days").

2026-08-14 — Agent Env Var Import/Export, WeChat Official Account Publishing Skill, More Reliable Skill Installs

What this means for you

Copy agent environment variables as JSON, or import them from JSON or a dotenv file: The environment variables tab of an agent now has "Copy as JSON" — one click exports the whole variable list to your clipboard — and "Import JSON", where you paste either a JSON object or KEY=VALUE lines and they merge into the editor (same keys are overwritten). Moving an agent's configuration to another workspace, or bringing in a .env file, no longer means retyping every variable.
New marketplace skill: WeChat Official Account publishing: A new built-in skill in the marketplace covers the whole WeChat Official Account publishing flow — configure your AppID/AppSecret, manage drafts (图文消息 articles and 小绿书 posts), upload media, check article statistics, and convert Markdown to styled HTML with 29 themes. AI writing and de-AI prompts are included for drafting content, and you can also call the official account API directly.
Skill installs are more reliable: Installing a skill from the marketplace no longer fails on a single temporary hiccup — transient network errors and 5xx responses from the skill catalog are retried automatically (3 attempts with a short backoff) before giving up. Permanent errors such as a missing skill still fail immediately.
GetNote skill points to the right credential page: The install flow for the GetNote (Get笔记) skill now links directly to the page where you obtain your client ID and API key, so setup no longer ends up on a page that does not have them.

2026-08-12 — Office/PDF Preview in Project Docs, Version History with Diff & Restore, Desktop Local Mirror Sync

What this means for you

Preview Word, Excel, PowerPoint and PDF files directly in the browser: Word (.docx) files render as readable pages, Excel (.xlsx) files display sheet by sheet, PowerPoint (.pptx) files browse slide by slide, and images and PDFs open right in the preview pane — no need to download anything.
Document version history: timeline, line-by-line diff and one-click restore: Project documents now have a version history panel where you can browse all past versions on a timeline, see the line-level differences between any two versions, and restore any older version with one click. Each document keeps up to 50 versions — the oldest are pruned automatically, so history never piles up without bound.
Desktop local mirror with bidirectional sync: The desktop app can mirror your workspace's project documents into a local folder — files added or changed in the cloud are pulled down automatically, local additions and edits are pushed back up, and files whose content is already on the server are skipped. Text conflicts resolve to the most recently modified version, with the overwritten content still recoverable from version history; binary conflicts never overwrite the cloud copy.
Full-text search across project documents: You can now search your whole workspace's project documents by keyword to quickly find the document you need, instead of browsing folder by folder.
Anonymous website bubble chat for external agents: After enabling public access in an Agent's external-access settings, you get a token-based web chat entry where visitors on your website can talk to the agent without logging in. You can restrict the allowed embedding domains (origins) and IP ranges to keep the entry safe from abuse.
Return to your original page after logging in: Fixed an issue where unauthenticated visitors opening an external agent chat link were dropped on a blank page after signing in — now you land back on the page you originally wanted, so visitors coming in from a website bubble can continue the conversation after login.
New aura uninstall CLI command: The command-line tool can now be cleanly removed with a single aura uninstall command (add -y to skip the confirmation prompt). Uninstalling keeps the configuration and data under ~/.aura, so a fresh install still picks up where you left off.

2026-08-11 — Video/Audio in Project Documents, Clone Website Skill, External Agent Chat History Restored

What this means for you

Project documents now support video and audio: You can upload video/audio files to project documents and play them inline in the preview — drag the progress bar to seek, no download needed. Product demos, training recordings and meeting audio can all live inside a project document and be shared with it.
Storage usage indicator for project documents: Each workspace gets 10 GB of document storage by default. The project docs page shows a usage bar that turns yellow near the limit and red when exceeded, and uploads over the limit are rejected with an explicit "X used of Y" message instead of failing silently. Workspace admins can adjust the quota in settings — set to 0 for unlimited.
New "Clone Website" skill: The marketplace now carries the clone-website skill. Your agent can reverse-engineer one or more target URLs and rebuild them section-by-section as pixel-perfect clones — colors, spacing, typography, animations and interactions — with multiple pages built in parallel. Handy for quickly replicating reference sites.
External agent chats resume your history: Opening an external agent's chat page no longer starts from a blank window — the page resumes that agent's most recent non-archived session, so your previous conversation is still there. A fresh session is started only when there is no history.
CLI updates and installs are more reliable: When GitHub is unreachable, CLI update downloads now fall back to a mirror automatically, so updates are less likely to fail on slow networks. Update checks honor your configured timeout, and when a check times out the CLI tells you the exact retry command instead of hanging silently.
More accurate PDF export for HTML documents: Fixed an issue where scripts and styles embedded in an HTML document could interfere with the layout when exporting to PDF — exports now match the preview more closely.
Skill bundle install fix: Fixed incorrect internal paths in marketplace skill bundles (e.g. the Google Stitch bundle) that could leave installed skills incomplete.
Self-hosted: fuller object storage support: Tencent COS now supports virtual-hosted-style storage endpoints (no more path-style request limitation), and a new sync_uploads_to_s3 tool migrates locally stored attachments to S3/COS (including MinIO/R2) — dry-run by default, with per-row size and SHA256 checksum verification, so migrations are safer.
Self-hosted: track email delivery status: Delivery events from your email provider (e.g. Tencent Cloud SES) are now recorded, and bounce/dropped events are summarized to your ops mailbox — failed verification codes and invites are no longer silent.

2026-08-10 — Project Documents Downloadable as PDF, "Collapse All" Fixed in Device Management

What this means for you

Project documents now download as PDF: The download menu in the project docs preview and file list now lets you choose between the original file (Markdown/HTML) and a PDF — documents are laid out on A4 pages and exported as a PDF, perfect for sharing and archiving. The PDF is generated locally in your browser using your system fonts — no extra setup needed.
"Collapse all" in device management actually collapses devices now: The "collapse all" button in device management previously left every device expanded, so clicking it had no visible effect. It's now fixed — one click collapses the whole device list.

2026-08-08 — Knowledge Planet (zsxq) Skill Added, Marketplace Skill Bundles, Clearer Install Messages

What this means for you

New "Knowledge Planet" (zsxq) skill: The marketplace now offers the Knowledge Planet skill, wrapping the full zsxq-cli. Your agent can now handle the whole scope of Knowledge Planet operations — publishing/editing topics, replying to comments, curating highlights and hashtags, checking members and expiry dates, daily rounds, ops reports and weekly reviews — and can even generate planet-poster images and vertical animated videos.
Knowledge Planet login happens entirely inside the install dialog: On install, the bound agent sets up zsxq-cli on its device and starts an OAuth 2.0 device flow. The dialog shows a login code (with a one-click copy button) next to the authorization link — open the link and authorize, no terminal needed.
Marketplace now supports skill bundles: A new "bundle" item type packages multiple skills/plugins into one curated collection — a single install imports the whole set into your workspace, and the detail page lists every skill included. The first bundle is Google Stitch's Stitch Skills (design, build, and utility plugins).
Install script messages no longer look like failures: Right after a release, the domestic mirror takes a few minutes to sync. Install scripts now clearly state "mirror not yet synced (normal right after a release), downloading from GitHub Releases" — no more misreading the fallback as a failed deployment.

2026-08-07 — Guided Agent Install Fixed on Windows, OpenClacky Added, OS-Aware Homepage Install Command

What this means for you

aura setup guided agent install now works on Windows: When no agent CLI is detected, the guided setup flow fully supports Windows now — installers that relied on curl | bash (CodeBuddy, Kimi) fall back to PowerShell irm | iex, and npm installs are made PowerShell-safe, so setup no longer fails on systems without sh/bash. If PATH needs fixing after install, Windows users get a "restart your terminal" hint instead of a source ~/.bashrc instruction that doesn't apply.
OpenClacky joins guided install: aura setup can now walk you through installing the OpenClacky runtime, running its official install script automatically on both Windows and macOS/Linux.
Homepage install command matches your OS: The CLI install command on the landing page now adapts to your system — Windows visitors see the PowerShell command, macOS/Linux visitors keep the curl one — so copying it always works.

What this means for you

Team invitation links work again: Clicking "accept" in an invitation email now actually accepts the invitation — you see a clear accept/decline page and land straight in the joined workspace. Previously the link opened the referral page and silently ignored the invitation, so it stayed pending and asked you to accept again later.
Devices can be deleted from team management: Each device in the team device list now has a delete action that tears down every runtime on that machine in one step. If any runtime still has active agents, the flow asks you to confirm archiving them first, so nothing is deleted by accident.
One-click "my devices" filter: The device list now has a filter that narrows it to machines running at least one of your runtimes — quick to find your own devices as the inventory grows.
Collapse-all in the device list: A new "collapse all" button folds every expanded device entry at once instead of collapsing them one by one.
Shared HTML documents load better on mobile: Opening a shared HTML document on a phone now shows a proper loading state and renders at the right height, instead of a broken or stretched viewport.
aura daemon start no longer looks stuck: The command now reports success as soon as the daemon is up and running — agent detection and workspace sync continue in the background without making a restart look like it failed.

2026-08-05 — New WeChat & Xiaohongshu Skills, Agent Failures No Longer Silent, Faster Project Doc Editing

What this means for you

New "WeChat chat analysis" skill: The marketplace now offers a WeChat chat analysis (wechat-cli) skill. Your agent can search WeChat messages, read chat history, check unread messages, inspect contacts and group members, export conversations, and produce chat statistics — all from your local WeChat data. It only reads the local WeChat database and never uploads chat content, so your privacy stays safe.
New "Xiaohongshu account starter" skill: The marketplace now offers a Xiaohongshu (小红书) account-building skill. Your agent can break down viral video scripts, rewrite/remix spoken-word copy, produce titles and post text, and turn a finished script into a shootable storyboard — a repeatable workflow for growing your account.
Agent failures are no longer silent: When an agent task fails with no retry left, the issue is now marked "blocked" on your board, and the failure comment explains why — for example the upstream account ran out of credits or the selected model is unavailable — along with a suggested next step (top up credits or switch models, then retry). Problems are visible at a glance.
Faster, safer project doc editing: Project docs now autosave drafts locally so your work survives network hiccups, show live save states (saved / saving / unsaved), and detect conflicts when the server-side doc changed while you were editing — you can then choose "load server version" or "overwrite server".
Agent template instructions fixed: Creating an agent from a marketplace template (such as "Project Manager") no longer loses the template's built-in instructions — they now load in full.
Feishu skill install flow hardened: Fixed a case where the Feishu (lark-cli) skill setup could never finish because a background process was interrupted mid-authorization; authorization state is now kept in per-task files, so concurrent setup tasks on the same device no longer interfere with each other.
Stability: The daemon no longer exits when it hits transient API failures during startup — it retries automatically.

2026-08-04 — Easier, Safer CLI Installation, New Karpathy Personal Wiki Skill

What this means for you

CLI installation is easier and safer: The macOS CLI is now shipped as a DMG installer signed and notarized by Apple, so downloading and installing it is more secure and needs no manual terminal commands. Windows gains an interactive setup wizard — double-click the .exe and the CLI installs into %USERPROFILE%\.aura\bin and is added to your user PATH automatically (the scripted install still works too).
New "Karpathy personal wiki" skill: The marketplace now offers a Karpathy-style personal LLM wiki management skill. Your agent can ingest web pages, papers, PDFs, Markdown, or pasted text into a raw/ source library, compile them into durable wiki/ knowledge articles, answer questions with citations, and lint the wiki for index, link, and evidence quality — your personal knowledge base stays solid as it grows.
Feishu skill install flow fixed: Installing the Feishu (lark-cli) skill no longer fails with "lark install not configured". The agent you pick now installs the CLI on its own device and guides you through OAuth authorization, with progress shown live in the install dialog — the same experience as other skills.
Cleaner marketplace browsing: Agent templates (such as the "Project Manager" template) no longer appear on the skills marketplace page — they only show up in the partner market when you add an agent, so browsing skills isn't cluttered with templates.
Polish: Fixed the Tencent ima skill's API key install dialog, and the mobile landing page header overlap.

2026-08-03 — Images in HTML Project Documents, Skills Built by Your Agents

What this means for you

HTML project documents now support images: The HTML editor has a new insert-image button — upload an image to embed it in the document, or click any image in the preview to swap it out in one click. When you export an HTML document, its images are packaged into the ZIP, so image-rich documents stay portable and easy to share. The HTML editor preview scrolling issue was also fixed.
No more hand-writing skills to create one: The "manual creation" option is now "let an agent create it" — pick an agent and describe what the skill should do (purpose, inputs, workflow, acceptance criteria), and it builds the skill as a shared team capability using skill-creator, with a task created to track the work.
Localized marketplace skill names: Platform skill names now display in your interface language (Chinese, English, Japanese, Korean), making it easier to understand what each skill does while browsing the marketplace.
New Tencent IMA skill: The marketplace now includes the Tencent ima (ima.qq.com) skill, letting your agents manage notes and knowledge bases — create, search, browse, and edit notes, upload files, or save web pages to the knowledge base.
Computer Use marketplace review: The earlier QwenPaw-only Computer Use listing has been retired while we redesign Computer Use as an agent-agnostic capability with explicit runtime, desktop helper, and OS permission boundaries.
"Project Manager" agent template (Tencent Meeting workflow): The partner market now seeds a PM template that books Tencent Meetings and syncs them to the project calendar, generates minutes and task dispatch 2 hours after the meeting ends, and archives the full transcript and minutes into the project documents folder — end-to-end meeting management.
Custom device names: Devices in the team device inventory can now be renamed inline with a click, and the label survives daemon restarts — easier to tell devices apart as your inventory grows.
Live Markdown preview: Fixed the issue where previews went blank for up to 5 minutes after saving; the Markdown editor is now a split-pane live preview, so what you see is what you get.
Stability and polish: ZIP skill uploads no longer report false "SKILL.md not found" errors, SKILL.md detection is now case-insensitive so it works reliably on Windows and macOS, project documents with Chinese filenames no longer hit false duplicate-name errors, and agents bound to retired runtimes (libtv/superun) are automatically detached instead of getting stuck.

2026-08-02 — New Marketplace Skills and a Smoother Skill Install Experience

What this means for you

New "Code Video" platform skill: The marketplace now includes a code video (HyperFrames) skill that renders HTML/CSS directly into MP4 video. Your agents can now generate code-related videos — such as captioned explainers — from a website URL, a GitHub PR, a Figma design, or a text brief.
New LibTV skill, installed entirely in-app: The marketplace now offers the LibTV (liblib.tv) skill, letting your agents drive the LibTV canvas from the command line. You no longer need to open a terminal: the bound agent installs the CLI on its own device and completes login authorization, with install, configuration, and verification progress shown live in the install dialog — you just click to authorize.
Interactive OAuth login for the Feishu skill: The lark-cli (Feishu) skill now comes with an interactive OAuth authorization flow during installation, supporting an auth link and QR code login, guided by your agent inside the install dialog.
Bigger skill imports: URL import caps for skills have been raised substantially (128 → 512 files, 8 → 32 MiB), so skill repositories with many supporting files now import cleanly instead of failing the "import cap exceeded" check.
Agent backend adjustments and stability: The libtv and superun local runtime backends were retired (LibTV capability now comes via the marketplace skill), and a LinkAI CLI command-wait issue was fixed for more stable runs.
Email verification in the admin: Admins can now verify an account's email directly from the admin panel.
Polish: Platform skill logos in the marketplace were refreshed for easier recognition, and skill install/detail dialogs now scroll when content overflows, so nothing becomes unreachable on smaller screens.

2026-08-01 — Feishu CLI Skill, Richer Project Document Info, and Smoother Installation

What this means for you

New Feishu CLI platform skill: The marketplace now includes a lark-cli skill wrapping the full Feishu CLI with interactive login. Your agents can now take on Feishu-related automation tasks directly.
Project documents carry more context: When you share a document via a public link, the preview now shows the author and created/modified time, so readers can instantly see the source and version. In the file tree, hovering over a truncated filename reveals the full name.
A more recognizable marketplace: Skills like Tencent Meeting, Baidu Netdisk, and GetNote now display brand logos, making it easier to spot what you need.
Dev environment setup, phase two: A new AI tool installer module sets up common AI tools with a single command, so getting a local development environment ready takes less effort.
More reliable Windows installation: The PowerShell install scripts now enforce TLS 1.2 and fix version parsing, so installation works on more Windows environments, including older PowerShell versions. Install timeout behavior is also documented clearly.
More stable agents and CLI: Fixed qwenpaw backend registration and the LinkAI CLI protocol, model discovery, and migration issues, with hardened CLI backend support overall. aura daemon status now warns when the daemon version differs from the CLI, and aura setup --token clarifies how to use user access tokens.
No-source private deployment for enterprises: Documentation now covers private deployment without delivering the source tree, so enterprises can self-host without receiving the codebase.
Landing page polish: Chinese copy has been normalized and the homepage layout refined.

2026-07-31 — ERP Enterprise Solution, New Agent Runtimes, and Developer Tooling

What this means for you

AuraBaba ERP enterprise solution is here: A full enterprise resource planning solution is now available on the AuraBaba platform, bringing finance, HR, supply chain, and operations management into your AI-native workspace.
Four new AI agent backends are now supported: LinkAI CLI, libtv CLI (for liblib.tv), superun-cli, and qwenpaw CLI (via ACP) are all now recognized and executable. Your agent ecosystem just got significantly wider — choose the backend that fits your workflow.
One-click dev environment setup: Phase 1 of the development environment setup scripts lets new contributors get a fully working local environment with a single command. No more hunting down dependencies or configuration steps.
Aura CLI mirror updates are now faster: Package and update downloads from Aura mirrors complete noticeably quicker.
Windows installation is more robust: The batch install command now works correctly inside PowerShell, and denied PATH updates are gracefully tolerated instead of failing the entire install.
Landing page and marketplace polish: Stale documentation links have been removed from the landing page, and Baidu Drive marketplace setup metadata has been corrected.

2026-07-30 — HTML Project Documents, New Marketplace Skills, and Brand Polish

What this means for you

HTML files are now first-class project documents: You can upload and preview HTML files inside project documents. When shared via a public link, HTML documents render with full interactivity — embedded scripts, styles, and links all work. The shared viewport maximizes to fill the screen, and document content streams in for faster loading.
Two new skill categories are available in the marketplace: DevOps platform skills and academic research skills have been added. Your agents can now handle DevOps automation and academic research tasks out of the box.
Skill downloads are more reliable: The server now skips binary assets before downloading skills, reducing download failures and making skill installation faster and more stable.
Installation guides use consistent branding: All user-facing installation guides and onboarding pages now consistently use Aura branding. External install sources are restored with appropriate warnings so you can see all available options.
Invite page stays in your workspace navigation: The referral and invite page now uses an account-scoped route and stays within the workspace navigation shell, so you stay oriented while inviting team members.

2026-07-29 — WeChat Mini Program Voice Input, Chat Input Redesign, and External Agent Access

What this means for you

The WeChat Mini Program now supports voice input and has a redesigned chat interface: You can now dictate tasks using the WechatSI voice-to-text plugin. Voice input appends to existing text instead of replacing it, making it easy to combine speech and typing. The send button has been redesigned — a clean black circle that turns blue when you have content. The agent picker and workspace picker now sit inline inside the input box for a more compact and focused experience. SVG icons replace emoji throughout, and the voice button now appears as a green mic in the tab bar.
External users can now access agent chat sessions: External agent chat access is now available. Session replies from external participants remain visible, making it possible to share agent conversations with people outside your workspace.
Skill uploads are more reliable: SKILL.md file detection in folder uploads has been fixed, and missing file content is now handled properly. Skills you upload manually or via agent automation are less likely to fail silently.
Workspace names are cleaner: The server now strips special characters (@ and #) from workspace names, preventing formatting issues in URLs and mentions.
Brand and architecture updates: The license and copyright attribution have been refreshed for mobile. All marketplace references now use "Aura CLI" branding. An architecture review for domestic expansion has been published. Account referral page now renders globally.

What this means for you

The WeChat Mini Program now supports the full path from signup to use: Users can bind a WeChat account, verify email, create a workspace, and complete onboarding inside WeChat. A new Workbench tab gives users a real starting point, and the new Mini Program marketplace skill lets agents handle Mini Program-related work directly.
CLI updates and downloads are more reliable: CLI updates now retry automatically, use longer download timeouts, and defer updates when the daemon is busy. Temporary file permissions and metadata preservation were fixed. New project-doc download command lets you download project documents from the terminal. OpenClacky runtime detection was also added for broader local AI tool support.
Project document sharing feels more polished: Shared documents now generate AuraBaba-branded PNG posters. Document preview loading is faster for large files. A new move-file dialog lets teams reorganize files inside project docs more easily.
Calendar supports cross-workspace views: You can now view events from multiple workspaces at once instead of switching back and forth. Created events now display reliably.
Credits and permissions are clearer: New users receive starter credits after email verification. Project document credit usage is more transparent. Workspace admins can see member emails for easier troubleshooting, and partners can now be removed for safer access control.
Docs and guidance are more accurate: New AI project management best-practice docs help teams use AI agents more effectively. SEO sitemap and LLM skill links were fixed, and not-yet-implemented feature docs were removed so the docs match the product.
Interaction details are smoother: Create-task and agent-mode dialogs are larger on big screens. Org-chart connector alignment was fixed. Outdated workspace-management guidance and broken onboarding links were cleaned up.

2026-07-19 — Simpler Agent Creation, Faster CLI Installation, and Copy Polish

What this means for you

New tasks now default to Agent mode: Clicking + New Task opens the agent-description flow directly. You describe the outcome you want, and the agent helps fill in the task details. The manual-mode toggle was removed to make the flow simpler.
Project document sharing works for every file type: File-based project documents can now be shared externally. Any project document, including uploaded files, can be shared inside the workspace or through a public link.
CLI installation and updates are faster and more resilient: CLI binary downloads now prefer China mirrors, with GitHub as a fallback. The install script output was refreshed so the install process is clearer.
Product copy and localization were tightened: Chinese smart-quote issues were fixed, and the agent creation panel, chat starter text, visibility labels, and other strings received updated English, Chinese, Korean, and Japanese translations.

What this means for you

The WeChat Mini Program supports a complete task workflow: Task management, inbox, and agent monitoring are now available with native WeChat styling. You can review tasks, process messages, and monitor agents without leaving WeChat.
Project documents can be shared externally: Project docs now support workspace sharing and public links. Agents can also upload and manage workspace documents, making it easier to prepare and publish project materials.
Offline device warnings prevent wasted waits: When a teammate's device is offline, the system now shows a clear warning so you do not assign work to an unavailable machine.
A video editing skill is now available: The marketplace now includes a video editing skill, so agents can handle video editing automation tasks.
The credit model is simpler: Credit consumption was simplified to project document creation. Invitation credit display and calculation were fixed so balances are easier to understand.
UI details were cleaned up: The CLI now uses the brand logo, update checks prefer the official site, the create-task dialog no longer has a continue-creating toggle, Inbox was renamed to Messages, and the autopilot navigation entry is clearer.

What this means for you

Navigation icons were refreshed: Top navigation and inbox controls now use Tabler icons, giving the product a more consistent and polished feel. A new business navigation icon set makes key entry points easier to recognize.
Brand references are now consistent: The final old-brand references were replaced in runtime update checks, analytics, and integration identifiers. The product now consistently uses Aura branding.
The skill marketplace is safer: Server-side filtering hides retired platform skills, so users only see currently available skills.
WeChat Mini Program production fixes shipped: Production API address and URL validation fixes help Mini Program users use the product normally.

2026-07-14 — First Official Release

AuraBaba's first official release gives 2-10 person AI-native teams a complete platform for managing agent work.

What this means for you

AI agents are first-class team members: Agents have profiles, can be assigned tasks, can comment, and can update status. Humans and agents collaborate in the same activity timeline.
Tasks have a full execution lifecycle: Tasks move through queueing, claiming, starting, completion, and failure states. Agents report blockers and stream progress so work does not fail silently.
Skills are reusable capabilities: A skill packages code, configuration, and context as a reusable agent capability. Write it once, then every agent on the team can use it.
Runtimes are managed in one place: Local daemons and cloud runtimes share one management view with online/offline status, usage charts, and activity heatmaps. The product automatically detects 12 installed AI coding tools.
CLI support covers every major platform: macOS, Linux, and Windows are supported. The one-line install script detects the system and downloads the latest release automatically.

Product Changelog

On this page