OpenClaw v2026.5.16-beta.7 landed on May 18, 2026 at 3:47 AM America/Chicago time, which was May 18, 2026 at 08:47 UTC. The headline here is not one giant feature. It is a release full of sharp little fixes and a few new tools that remove the kind of friction you only notice after you have been living in OpenClaw for a while.

The three things I would pay attention to first are better browser dialog recovery, a real typed plugin workflow, and reply delivery fixes that stop finished answers from quietly disappearing after preview updates. If you automate web flows, build your own tools, or run OpenClaw in rooms where one dropped reply can confuse everybody, this beta is worth a close look.

What users can now do

Browser automation gets a lot less blind. OpenClaw now surfaces pending and recently handled modal dialogs in snapshots, reports blockedByDialog when a click opens a modal, and lets you answer the dialog directly. That means fewer dead-end browser runs where the page looks stuck and you have to guess what blocked it.

Plugin authors finally get a cleaner starting point. This release adds defineToolPlugin plus openclaw plugins build, validate, and init. In practice, that gives you a more normal path for building simple typed tool plugins with generated manifest metadata instead of hand-rolling everything from scratch.

The Mac app is easier to live in. Settings got a real layout pass, with steadier navigation, cleaner panes, and fewer awkward little spacing problems. That is not glamorous, but it matters if you spend real time in the app instead of treating it like a setup wizard you open once a month.

There is also a lot more surface area for power users. New bundled skills cover meme generation, node inspector debugging, Python debugging, fused diagrams, and throwaway spikes. None of those will matter to everybody. One or two of them will matter a lot to the people who needed them yesterday.

What got safer or less annoying

The reply-path fixes are the ones I would not ignore. One fix keeps final payload delivery alive after live preview updates, so channels can still send the completed answer instead of getting stuck with a preview draft. Another keeps approved async commands from stalling on follow-up approval calls. Those are the kinds of bugs that make operators doubt the whole stack, even when the underlying work actually succeeded.

xAI login cleanup also keeps moving in the right direction. The release notes call out OAuth and sidecar auth fixes, including loopback callback handling, video generation polling defaults, and native-host user-agent attribution. If you tried the new xAI path recently and hit a weird edge, this is the beta I would retest on.

There is also a steady stream of recovery and safety work across sessions, plugins, channels, and QA. OpenClaw now fails closed in a few places where it used to guess, hides internal transcript rows that should never have leaked into chat history, and gets more honest about missing coverage in parity checks. Good. Systems like this should be strict in the weird corners.

One practical footnote: the release raises the minimum supported Node.js 22 line to 22.19. That is easy to miss. It is not optional if you are still pinned lower on the 22.x line.

What I would test right after updating

1. Run one browser flow that used to get stuck on alerts, confirms, or auth popups. Make sure you can see the dialog state and recover without restarting the whole run.
2. If you maintain plugins, scaffold one throwaway tool with the new plugin commands. I would rather find manifest or typing issues in a five-minute spike than in the plugin I actually care about.
3. Do one real reply test on the channel that matters most. Watch the full path from preview to final delivery. This release specifically targets the class of bugs where the final answer gets lost.
4. If you use xAI, do one fresh login and one actual media or model call. Not a config check. A real task.
5. Confirm your Node version before blaming anything else. On hosts pinned to Node 22, make sure they are at 22.19 or newer before you start debugging side effects that are really just version drift.

Should you install this beta?

Probably yes, especially if you build on top of OpenClaw instead of only chatting with it. This is a builder's release. The flashy part is modest. The useful part is everywhere.

I would still treat it like a beta and spend ten minutes on the retest list above. But I am glad they shipped this one. It fixes the sort of stuff that wastes afternoons.

---

Full changelog: View on GitHub