He leads engineering on Gemini at Google. instead of keeping his Claude setup private, he open-sourced it. Addy Osmani. That Google: Chrome DevTools lead, "Learning JavaScript Design Patterns" author. 'agent-skills' - his personal loadout. Drop-in for any project. 68,925 stars. MIT. → bookmark it. This is how your Claude... setup goes pro.show more

Not just Addy, all of us should make our Claude setups public! Here's mine for reference - Includes skills that: - Deliver tickets end-to-end - Enforces TDD on every commit - Discusses design-decisions via AskUserQuestion tool

He left Google

Leads gemeni -> claude setup. Ngmi

model ain't the main thing, it's the wrapper fr

loosing him is googles great loss .

He wasn’t leading Gemini engineering. He is successful as is why make up random attribution?

Open source is still the fastest way to level up with AI.

Nice article

No he doesn’t

Dude, this is gold! Thank you for sharing.

you can also bookmark and try this and thanks me later, instead of remeber which phase you should trigger, this single skill guide you through the whole dev lifecycle

Addy is a rare gift to this world.

👀

Cool guy!

If he’s using clause he’s a pleb

Nice move sharing it—open sourcing shows real confidence. Congrats to Addy and the team on the Gemini win. Big respect for the transparency.

@grok how would you evaluate and rank the skills in that repo?

cron-based loops are very efficient. Agents should know who they should talk to next

It's always interesting to see how experienced engineers structure their workflows. The setup often teaches more than the prompts

Google engineer shares Claude setup

I just open-sourced mine too, do you have one for inspo?

Exactly. The model is table stakes. The skills layer — structured workflows, verification gates, personas, anti-rationalization — is where production capability compounds. Addy open-sourcing this (pulling real Google-scale engineering discipline into reusable agent skills) accelerates the entire ecosystem. Teams should study and adapt it. At enterprise scale, though, these autonomous trajectories create the accountability gap: cost attribution, loop detection, trajectory drift, and policy enforcement before spend compounds. PromptKing is the economic accountability and governance layer on top. Seat classification. Simulate-before-enforce. Native governance APIs across Claude, Gemini, Copilot, Bedrock, Watsonx. Velocity without governance is just faster waste.

The fact that a Gemini engineering lead uses Claude as his personal setup and then open sources it is the most honest signal about model preference you'll ever get. Actions over employer loyalty. Which specific skill from his loadout had the biggest impact on your workflow?

There is no point for them keeping this setups private. If they won't someone else will. And they pretty much are close to each other. As the approach is not deterministic, many will come up with any new line of setup. Of course some are better some are not.

He left Google.

The harness is indeed the floor! Mind blown by the simplicity and power of these loops. Kudos to Addy Osmani for sharing his expertise

The man behind google cloud AI

这种配置公开出来，细节比名头更有看头

Open-sourcing your workflow does something interesting. It turns individual expertise into a community asset, and the community often improves it faster than any single person could.

Output like this is downstream of setup, not luck. Once memory and skills are wired right the bottleneck disappears. Whole stack here: