The Weekly Claw
Skill Workshop is your homework. Peter built the thing that builds the thing.
One hour. The week, in claws.
Skill Workshop is your homework. Peter built the thing that builds the thing.
One hour. The week, in claws.
Drop questions in text chat as we go. Henry and Tim drive their own slides.
Every Friday, same slot, same Stage. New here? Drop your questions in text chat as we go and I'll catch them between segments. Prefer not to be on the recording? Hang out in text chat — you're still in the room.
OpenClaw has no token, no coin, no blockchain component. If you see a wallet, an airdrop, a presale, or any account claiming an association with this project — it's a scam. /report and move on.
Episode 18. Friday, June 19th, 4 PM Eastern. Settle in.
And — ever — no crypto, betting, or trading talk. That bar is what keeps serious builders and high-profile guests coming back. No self-promo without staff sign-off; don't DM members about your project.
Help goes to #help & #users-helping-users — not #general. Model talk to #models. Something looks off? /report and a mod will see it.
If you've been talking a lot, make space for someone who hasn't. Keep the language kid-safe-ish — kids show up. /report-vc pings Audrey if something goes sideways.
Stuck on an install? The fastest unblock is almost always the git install: curl -fsSL https://openclaw.ai/install.sh | bash -s -- --install-method git
Lighter week, and there's a reason. Peter's moved out to SF and is getting his head around his new role at OpenAI, so the project's running a little quieter while he gets settled.
One item worth real time → Skill Workshop
Instead of you hand-building custom skills, it watches what you do, notices the recurring patterns, and proposes new SKILL.md files for you to review and activate. A loop that improves on its own.
This is OpenClaw's answer to "but I don't know how to configure it." It lowers the bar for non-technical users — and the more you use it, the smarter it gets at knowing what you need. That self-improving loop is what separates this from the toys.
If you haven't switched it on yet — do it. Then come back next Friday and tell us what it proposed for you, and whether you kept it.
docs.openclaw.ai/tools/skill-workshop
I won't run the momentum lap. The star count keeps climbing, and OpenClaw's showing up in home robots in China now. We've covered all of it.
The interesting question stopped being "is this the future" a long time ago. It's "what are you building with it."
GitHub star history · openclaw/openclaw — surpassed React's decade-long record in ~60 days
Everything known has been patched. And NemoClaw was built specifically to add enterprise security guardrails for that exact concern.
The line: known issues, all patched, and the enterprise hardening layer exists.
Not something to lead with — good to have straight if anyone asks.
For what actually shipped this week and the developer-experience read on it — the changelog and DevRel desk is yours.
Host's slides come back for the video
Peter Steinberger · Microsoft Build 2026 · session BRK245 · Microsoft Developer on YouTube · ~42 min
How the OpenClaw maintainer team uses agents to maintain OpenClaw. They used the thing to build the thing — and Peter walks you through it at a technical level.
Thousands of issues, Discord in a dozen languages at once, PRs faster than humans can review. His answer wasn't to hire — it was to build a system of agents.
Every few minutes an orchestrator wakes up and hands a queue of issues to worker agents. Triage, label, respond, route. Mechanical work — dupes, missing repro steps, stale issues — gets handled; hard cases get flagged for a human.
By the time something's a GitHub issue, the context was already hashed out in chat. A crawl system ingests Discord and distributes that signal into GitHub as comments, labels, and links. You don't have to live in Discord to benefit from what Discord knows.
A dashboard that tracks how ready a release really is — not just "are the tests passing," but open critical issues, blockers, recent CVE activity, and the PR backlog in one view. Closing tickets is one thing; knowing the true state of the codebase is another.
Agent-driven testing across Mac, Windows, and Linux. Then VNC: an agent operates a full desktop, clicks the UI like a user, reproduces a vague bug, records a 30-second clip, and attaches it to the issue. A completely different way to work.
The models are capable of more than we're asking of them. The Discord crawl, the Release Bar, the VNC workflow — none needed a smarter model. They needed someone to imagine you could wire things together that way.
A working maintainer showing you his actual system, down to the architecture — not a vision talk. If you maintain a plugin, skill, fork, or contrib workflow, the ideas translate. Closes on composability: keep each skill narrow and replaceable.
For where to meet this community in person over the next couple of weeks — the events desk is yours.
Pulled from the ClawExplorer calendar · Host returns to close
Name, where you're calling in from, and what brought you to OpenClaw.
What did it propose for you — and did you keep it? I want to hear it.
If Peter's talk sparked something you'd wire together — drop it in chat.
Taking questions from text chat here.
📍 Austria · the reason I was away
While I was off the grid in the Alps with the family, these two kept The Weekly Claw running without missing a beat. Henry on the changelog and DevRel desk, Tim on live events — they carried the show.
That’s the whole point of this community: people show up for each other. I came back to a show that never skipped a week. I owe you both one.
Two things to take with you, then we head straight into happy hour.
Turn it on this week and report back next Friday — what it proposed, and whether you kept it.
The agent system that maintains OpenClaw itself. "Models aren't limited by intelligence anymore — they're limited by imagination."
Grab a drink and stick around. 🦞 · "The Claw is the Law."