Fix TanStack Start XSS Flaw: VibeBuilder Weekly, Oct 5
A TanStack Start flaw affects some AI-built apps. Here is the fix, plus Sonnet 5.5 in Replit, Lovable and Microsoft, and upcoming events.
TL;DR
1A security flaw (CVE-2026-102989) was disclosed in TanStack Start, a framework many AI-built apps use. Lovable-hosted apps are protected; apps hosted elsewhere need an update and a redeploy.
2Anthropic released Claude Sonnet 5.5 on Sep 28, and Replit added it to Power Mode on Oct 2.
3Lovable added Microsoft 365 connectors and Microsoft sign-in, and a voice feature built on GPT Live.
Top Story: TanStack Start Patch for Apps Built with AI
On Sep 30, the TanStack team published a security advisory for a reflected cross-site scripting (XSS) flaw in TanStack Start. In plain English: an attacker could craft a link that makes a vulnerable app show attacker-controlled content, which can run malicious code in a visitor's browser. TanStack says it affects server-function handling in several react-start, solid-start, vue-start and start-server-core versions, and that fixed versions are out.
Lovable published its own post the same day. Lovable says a researcher found the issue on Sep 14, that it reported it to the maintainers, and that it added firewall rules for hosted apps. Lovable also says it saw no active exploitation, and that the fix applies on your next edit or immediately from the Security page.
Why it matters to beginners: you did not write this code, but you ship it. Check whether your project uses TanStack Start, update it, and redeploy. Updating on your computer alone does not fix the live site.
Why it matters to business owners: a flaw like this can put customer trust at risk even when your own code is fine. If your app is hosted outside Lovable, ask whoever built it to confirm the update and redeploy are done.
Plans and Pricing
1Claude Sonnet 5.5 (Sep 28): Anthropic lists API prices of $2 per million input tokens and $10 per million output tokens. Anthropic says it is 30%+ faster and costs up to 30% less than Sonnet 5 for most work. These are vendor claims.
2Replit (Oct 2): Replit's changelog lists Claude Sonnet 5.5 in Power Mode and GPT-6.1 Sol in Max Mode. Check what each mode costs on your plan before switching.
3Lovable (Sep 28 and Oct 1): Microsoft 365 connectors, Fabric integration and Microsoft sign-in work on all plans, per Lovable. Workspace sign-in with Microsoft Entra needs Business or Enterprise. Business and Enterprise also got IP allowlists for API keys, and Enterprise admins can turn on EU-only AI processing.
4Windsurf, now under Devin Desktop (Sep 29): you can sign in with ChatGPT and bill GPT model use to a ChatGPT Plus or Pro plan, per the vendor changelog.
5Vercel AI Gateway (Oct 1): Vercel lists the Laya model as free through Oct 31, 2026, when served by Boundless.
Security Watch
1Check your TanStack Start version. The advisory lists affected ranges, for example react-start 1.143.12 through 1.168.59, with fixes starting at 1.168.60. Update, then redeploy.
2Lovable API key controls (Oct 1): Business and Enterprise plans can restrict API keys to approved IP addresses. If you share keys with clients or tools, this limits damage if a key leaks.
3Vercel Sandbox (Sep 30): Vercel says Sandbox now supports Secure Compute connectivity, which matters if your agent runs code against private services.
Not Seen This Week
No verified credit, usage-limit or price changes from Lovable, Replit, Cursor or Bolt.new inside the Sep 28 to Oct 5 window. Cursor's newest changelog entry was Sep 23, so it is left out. Claude Code releases ship often, but the changelog I could read has no dates, so they are left out. Also noted: Bolt.new announced its first acquisition, Dokai, on Sep 29, with agent features promised this fall.
Try This Week
1Open your project's package.json, search for tanstack, and confirm you are on a fixed version. Redeploy afterward.
2If you build on Replit, try the same small task in Power Mode with Claude Sonnet 5.5 and compare it with your usual setting.
3In Lovable, try the new GPT Live voice feature (Oct 1) in a small test app, such as a spoken FAQ for your business.
Events: Next 30 Days
Oct 15: LA Tech Week Vibe Coding Camp, Los Angeles, 10 AM to 6 PM PT, free, per the AllAI Events listing.
Oct 16-18: JunctionX Turku Hackathon, Turku, Finland, per the Lovable community events listing.
Oct 18-22: Lisbon AI Week, Lisbon, Portugal, per the Lovable community events listing.
Join the Conversation
Questions about the TanStack fix or anything above? Join the free VibeBuilder Facebook group at facebook.com/groups/1113060757712466 and share what you are building. To get this roundup every Monday, subscribe to the newsletter on vibebuilder.space.
Sources
TanStack Start Security Update CVE-2026-102989: tanstack.com/blog/tanstack-start-security-update-cve-2026-102989
A vulnerability in TanStack Start: lovable.dev/blog/how-lovable-protects-your-app-from-a-tanstack-start-vulnerability
Lovable + Microsoft: lovable.dev/blog/microsoft-partnership
Lovable Changelog (Oct 1): docs.lovable.dev/changelog
Introducing Claude Sonnet 5.5: anthropic.com/claude-sonnet-5-5
Replit Updates (Oct 2): docs.replit.com/updates
Devin Desktop Changelog: docs.devin.ai/desktop/changelog
Vercel Changelog: vercel.com/changelog
Bolt.new Blog: bolt.new/blog
Cursor Changelog: cursor.com/changelog
LA Tech Week Vibe Coding Camp: allai.events/event/-la-tech-week-vibe-coding-camp--build-fast-ship-faster-learn-the-entire-ai-dev-stack
Lovable Community Events: whos-local.lovable.app/events
Originally reported by VibeBuilder Space
Read the originalMore news
Show HN: Pi pod โ Run your pi coding agent in sandboxes on your own server
Pi Pod lets you run custom AI coding agents inside secure, self-hosted sandboxes so you can build software safely on your own terms.
Source: Hacker News: Claude Code
Getting the most out of Opus 5.5 in Claude and Claude Code
Practical tips for getting better results out of Claude Code by giving clear finish lines and ditching old prompt habits.
Source: Hacker News: Claude Code
Supabase Select 2026 Recap
Supabase unveiled new features that let AI coding assistants configure backend databases, run locally without Docker, and monitor app health directly.
Source: Supabase Blog
Weekly vibe-coding tips โ coming soon
We're getting this ready. Leave your email and we'll let you know the moment it launches. No fluff, unsubscribe whenever.