OpenClaw 3.12 Update: What's New (Release Notes Breakdown)

0:00I just updated OpenClaw to 3.12 and in this video I'm going to be breaking down all the security updates. This might be the most security focused release that they've ever shipped. We're talking 15 separate security fixes in just one update. So, if you are running OpenClaw in production like I am, you need to update right now, not tomorrow, not yesterday, right now, because it's not just security in this release. There's a completely rebuilt dashboard. There's a fast mode for both Claw and GPT. So, this is going to be an amazing update if you're just getting going and there's some more that goes into it. So, they added some smart architectural changes as well that are going to matter a lot

0:45going forward. So, I'm going to break it all down. First up is the new control UI. They call it dashboard V2 and it's basically a complete rebuild of the gateway dashboard. So, now you get modular views for overview, chat, config, agents, and sessions a bit more organized. There's a command palette which if you've ever used VS Code, you know how powerful that can be. There's a mobile bottom tab so you can actually use this thing from your mobile phone, which is nice. And the chat interface got a huge upgrade with slash commands, search, export, and pinned messages. So, this matter because the old dashboard was functional but pretty basic. Now it's actually something you want to use every day. I spent a lot of time in my mission control dashboard which is my

1:30custom dashboard built for things like automating content, running my social media, you know, producing videos for my SaaS products, reaching out to people for podcast interviews. It's tailored towards me and I built it how I want. And for people that are just getting started, this new built-in dashboard is going to be really nice because it's going to be a little bit easier to navigate. So, you can manage your own agents, you can check config files, you can monitor sessions all from one clean interface. Then here's something that I'm genuinely excited about is the fast mode. For both Claude and GPT 4 5.4, this means that you can now toggle a fast mode at the session level. So, for OpenAI, it shapes the request to use their faster interface path. For

2:16Anthropic, it maps their service to their tier API, and you can set it with a {slash} fast in the chat and in the TUI as well and in the control UI or through the ACP. Why does this even matter? Because sometimes you don't need the full thinking power of Opus. Sometimes you just need a quick answer. Fast mode gets you that toggle without switching models. So, same model, faster interface, lower latency. Real quick, before we keep going, if you're watching this and you want to actually build with some of these tools, not just watch videos about them, then you're going to want to check out our community down below, Shipping School. We have a full Claude code course, a full open Claude course, and four live bootcamps every single week where we actually help you get set up from scratch. Like actually

3:03set this thing up, not just watch a tutorial and figure it out by yourself. And we also provide one-on-one coaching, so you could book a call with me, we could share screens, and I can help you get Claude code or open Claude running on your machine. That's it, no fluff. I built this community because watching YouTube's only gets you so far. We launched it just 3 days ago and we have over 55 members. You need people around you who are actually building, people who hold you accountable, and coaches who can help you when you get stuck. I'll put the link in the description down below. Get in now before the price goes up. So, for the cron jobs that I have that run at 5 and 6 a.m., this could cut execution time significantly on those subtasks. Okay, so now to the big one, security is the top of mind

3:48topic when it comes to open Claude. With every release, the security gets much more robust. So, there's 15 fixes in this latest update that are critical. I'm going to walk through the ones that actually matter for people running Open Claw. So, first is the workspace plugins. Before this update, if you had cloned a repository that had a workspace plugin code in it, that code can auto execute without you ever approving it. So, that could be a little bit scary if you think about it. You clone Open Claw, loads it, and suddenly untrusted code is running on your machine. Yay! This is why I never really encourage people to go to Claw Hub to download skills. I only suggest people go to Claw Hub to get skills that then you can give the URL to your Open Claw and say, "Hey, I

4:33want this skill. Now, recreate that skill for me." And I would highly avoid downloading anyone else's skill because all they are is just a markdown file. So, if you're looking to download skills as well, I would take that route as it is a much safer approach. You don't know what prompt inject in is in these skills, and you don't want to open yourself up to that kind of vulnerability. So, now that that's all fixed, plugins require an explicit trust decision before they can execute. So, that's a pretty big one. Exec approval Unicode tricks. Someone found that you could use invisible Unicode characters to make a command look safe in the approval prompt while actually running something completely different. So, you would approve what looks like a harmless command, but hidden

5:19hidden in zero-width character change what actually can be executed. Now, Open Claw strips and escapes all the invisible Unicode before showing you the approval prompt. So, what you see is what runs. So, device pairing tokens is the third thing I want to talk about. The old pairing flow embedded shared gateway credentials in chat messages and QR codes. Not great if someone intercepts that. Pairing uses short-lived bootstrap tokens that expire quickly, so it's a much more secure. The fourth thing is session status sandbox escape. The sandbox sub agents could previously inspect parent session metadata or even write model overrides to parent sessions through session status. So that's a sandbox violation that has been fixed. The fifth thing is

6:05gateway off scope escalation. On the shared token websocket connections, a deviceless client could self-declare elevated scopes, basically giving them selves admin access they shouldn't have. Yay! Now unbound scopes get cleared on connect. And there's more, wait. Browser profile creation is blocked from the browser control surface. Agent callers can't override workspace boundaries anymore. Ruby exact approvals now catch require flag tricks. The exact allowlist respects case sensitivity properly. So I know that's a lot, but here's the takeaway. If you're just getting started with Open Claw, you are via your If you're running Open Claw, you're exposed to the inner internet in any way through a reverse proxy, through Telegram,

6:51through anything. These fixes close the real attack vectors. This isn't theoretical security theater. These are actually researchers finding actual exploitable bugs, and the team is patching them fast. I just made a video last night about 3.11, and this one just came out 24 hours later. I love it. So let me talk about the architectural stuff because this is going to matter long term. A llama, LLM, and SG Lang have been moved to a provider plugin architecture. So what this means is instead of these being hardcoded into Open Claw, they now are modular plugins with their own onboarding, discovery, model picker setup, and hooks. This makes Open Claw more extensible. So if you want to add a local model provider,

7:37the plugin architecture makes that much cleaner. So for my setup, I run everything through Anthropic and Open Router, but for people in Shipping School community who are running local models on their hardware, that's a bigger deal, especially combined with a Llama onboarding improvements from the latest update. So, they also added Kubernetes docs and smarter manifest kind setup and deployment docs. If you're running OpenClaw in a team or enterprise setting, you can now deploy it on K8s properly, not something most of us need right now, but it does show where OpenClaw is headed. This is the beginning of production infrastructure, not just a hobby project. One more feature that is subtle but more powerful, sessions yield. So, this lets orchestrator agents end its current turn

8:22immediately. They can skip any queued work tool and carry a hidden payload into the next turn. Why does this even matter? Because sometimes an agent realizes mid-execution that it needs to change course. Instead of finishing all the queued work and then pivoting, it can yield immediately. So, for complex multi-agent setups like mine, the Rizza, Ghostface, and Inspected Deck, this kind of control flow does matter. Also worth mentioning is Slack got a Block Kit support. So, if you're running OpenClaw through Slack, your agent can now send rich formatted message with button selections and all the Block Kit components. So, the cron system fixed a bug as well where transients and retries

9:07could replay duplicate proactive messages after a restart. So, I actually have bit might have been hitting that one. So, how do you update? Like, how do you do this? Just run OpenClaw update in your terminal or if you're on -g OpenClaw @latest. Then go and run OpenClaw doctor - -fix to handle any breaking changes. OpenClaw doctor is going to be your friend anytime your OpenClaw breaks. I literally just did this 5 minutes ago and it took 30 seconds to update. Here's what I would do after updating is check your exec allow list if you have one because the case sensitivity fixed might affect your rules. If you're using workspace plugins, you will need to explicitly trust them on first load after updating. And if you have any custom pairing

9:53setups, the new short-lived tokens might mean you need to repair your devices. Yay! Bottom line, this is a serious release. The security fixes alone make it mandatory if you're running OpenClaw anywhere that touches the internet. The fast mode and dashboard are a nice quality of life improvement, but nothing groundbreaking. And the architectural changes show that OpenClaw is maturing into a real production infrastructure. I'm loving these updates. If you want to learn more on actually how to set up your OpenClaw properly and you want a community of like-minded individuals, maybe you want to run agents in production and build your own content pipeline like I have, that's exactly what we teach in Shipping School. I'll put the link in the description. You can check it out. We have over 50 members in just 3 days. We have four live coaching

10:39calls and we also have a one-on-one that you could do with me and I'll walk you through the setup. So, I hope you guys have a blessed rest of the night. We'll see you in the next one. Bye-bye.