Issue #4 · Sent Nov 6, 2026 · Week of Oct 30–Nov 5, 2026
A tampered Claude Code extension in Cursor, one shared ChatGPT and Codex allowance, and a GitHub Actions deadline
This week: what our weekly fact check changed, then eight releases and updates that matter for builders, each with a source. If you use Cursor, read the first item before anything else.
What our weekly check changed
On Nov 4 we rechecked 70 of our 71 listed tools against their own pricing pages, docs, changelogs and repos. Railway's pricing page now lists a Free plan: a 30-day trial with $5 in credits, then $1 a month. Railway also reopened signups on Nov 2, but for now new customers have to pick a paid plan (Hobby or Pro) before deploying, and free trial deployments are still unavailable. Smithery's Oct 31 shutdown date has passed. Its paid plans are gone, and smithery.ai stays up as a free public MCP registry. Locofy's pricing page couldn't be read this week, so its price stays as last confirmed.
New on the site: a Background jobs & workflow automation category, for running tasks on a schedule, on a trigger or in the background. n8n is our top pick, with Trigger.dev, Inngest and Activepieces as alternatives. Trigger.dev's listing links to its Nov 1 security disclosure (item 2 below).
Every change is on our public What changed log, with the before and after wording and the sources.
What changed · Background jobs & workflow automation · Deploy & hosting
This week's verdicts
Cursor's marketplace served a tampered Claude Code extension
Versions 2.1.312 and 2.1.313 of the Claude Code extension in the Cursor VS Code Extension Marketplace were modified copies that Anthropic didn't build. Cursor says they were up from about 2:27 AM to 1:35 PM PT for macOS and Windows. The code tried to add an attacker's SSH key, open a reverse SSH tunnel that keeps retrying, and send host details out. Cursor removed the versions, rebuilt the marketplace and got the attacker's server taken offline. If you installed or updated the extension in Cursor that day, Cursor's notice has a one-line check for a key labeled svc-telemetry and cleanup scripts for macOS and Windows. Uninstalling the extension may not be enough, and Cursor recommends rotating credentials that machine could reach.
SourceTrigger.dev reports exposed environment variables
Trigger.dev says an attacker got unauthorized access to environment variables for a subset of organizations. It says it blocked the exploit and emailed every affected account owner with steps to mitigate, and it will post updates on its status page. If you got that email, follow it. If you run jobs on Trigger.dev, it's a good week to look at which secrets live in your environment variables.
SourceChatGPT and Codex will share one usage allowance starting in December
On Plus, Pro and Business plans, ChatGPT and Codex will draw from the same allowance, measured in tokens. Bigger models and harder tasks use more of it, and Instant chat doesn't count. You'll see your remaining usage and reset times in Settings, under Usage. If you run out, you can wait for the reset, upgrade or buy credits. OpenAI says it expects most people won't wait for resets more often than they do now. The help article doesn't say how many tokens each plan includes, so if you lean on Codex, watch that Usage page in December.
SourceGitHub Actions will block pull_request_target in public repos on Dec 2
Starting Dec 2, GitHub will block workflows triggered by pull_request_target in public repositories unless an Actions event policy explicitly allows them. Private and internal repos aren't affected. The trigger runs with your repo's secrets, so if a workflow runs untrusted code from a fork, that code could steal them. Before then, GitHub will enforce the block during two brownouts, on Nov 9 and Nov 16 from 16:00 to 18:00 UTC (6 to 8 AM in Hawaii), when affected workflows will fail. If any of your public repos use pull_request_target, check GitHub's policy Insights now, then either switch to pull_request or allow only the specific workflows that need it.
Source · Securely using pull_request_targetVercel Security Review is in public beta, free until Nov 10
For Vercel Pro and Enterprise teams, Security Review checks every push to a connected GitHub repository for vulnerabilities, including direct pushes to your production branch. Each review covers only what changed, and findings come with a summary, a suggested fix and a severity rating. You can hand a finding to Vercel Agent, which opens a pull request for you to review, or pull findings into your own coding agent with the Vercel CLI. Vercel says it analyzes source code only and isn't a security audit. After Nov 10 it's billed at the Vercel Agent rate: model tokens at the provider's price plus $0.25 per million tokens.
Source · AnnouncementLovable now remembers what you tell it, and tidies that memory up
Since Oct 30, Lovable keeps the preferences, decisions and facts you state while building and uses them in later chats. Project memory is shared by everyone who can edit the project, and personal memory follows you across projects. You can see a plain-language summary in your Memory settings, fix it there or turn it off. On Nov 4 Lovable added Dreaming on paid plans: it checks its memory and your AGENTS.md rules against your code in the background and rewrites or removes what no longer applies. Lovable says your app's code never changes and Dreaming doesn't use credits. Lovable also now follows AGENTS.md and CLAUDE.md files in subfolders.
SourceGitHub Copilot now works inside Xcode's own coding assistant
If you build iPhone or Mac apps, you can now pick GitHub Copilot in Xcode's built-in coding assistant instead of installing a separate extension. It's a public preview that needs Xcode 27.3 beta on macOS Tahoe 26.6 or later, and every Copilot plan is supported. Set it up in Xcode, Settings, Intelligence. The existing Copilot for Xcode extension is still available.
SourceCognition releases SWE-2 Flash, a cheaper, faster coding model in Devin Desktop
Cognition, which makes Devin Desktop (formerly Windsurf), says SWE-2 Flash costs one-fifth as much and runs 3x faster than its larger SWE-2 model. It's available now in Devin Desktop and the Devin CLI. The weights aren't public: Cognition says it's building a program to share them with trusted partners and researchers. In Cognition's own benchmark table it trails Claude Haiku 5.5 on seven of the eight tests listed, so treat it as a cheap option for routine work inside Devin, not a step up.
SourceIf your app deploys on Vercel Pro from GitHub, enable Security Review in your team's Agent settings before Nov 10, when reviews are still free. Push a normal change and see what it flags. Then you'll know whether the findings are worth paying for at the Vercel Agent rate before billing starts. Nothing it suggests merges until you approve the pull request.
SourceThe weekly verdict
New AI tools for builders, and which picks changed, once a week. We will email you a confirmation link first. Unsubscribe anytime.