ecesena

I have a tmux split in 4. Personally I just use claude, but in any 4th you can run whatever you pref. Each agent has its own workspace (run in a diff dir), but they all share memory, skills, and can of course read each other work. Four is nothing special, just easy to still read in a 13” screen, easy to switch each pane to full screen and so on.

Edit: forgot the ssh part. Same setup on my mac, on my server, at work. So my UX is pretty much always the same. Most of the time it’s via ssh, yes, only rarely I use my mac.

Ops, I may have misinterpreted. I read it as take a passkey and copy/store it inside a hw sec key. Thanks for clarifying!

> By design, you cannot create a backup of passkeys on a hardware key

In fact, I think it's technically possible. But it's true, as of today I don't think anybody supports it.

I was experimenting on solokeys with ios, in principle we could backup Passwords (the ios/mac app) into a solo key, space permitting, including passkeys, regular passwords and totp (not wifi passwords). I believe the same is for Android, but haven't tested yet. This is all experiments I've been doing on my own, there's nothing ready to be released.

A better solution (than Shamir secret sharing) are threshold signatures.

The difference is that with Shamir you have to reconstruct the private key in one place before you can sign. With threshold signatures multiple servers can collaboratively sign without ever reconstructing the private key in a single place.

For chains like Solana, Aptos, SUI that use ed25519 (schnorr signature), there's a pretty clean solution called FROST.

For Bitcoin and Ethereum/EVMs that use ECDSA it's a bit trickier but there's been a lot of research recently, so there are solutions.

google drive where you don’t see the files, but you see the people who you shared to, and if you click you can see which files.

I do not have experience with Copperhead, but I made 2 PCBs with Claude. He made a py script that generates Kicad sources, I ended up generating all files (in particular schematics and PCB) from a shared netlist, and use Kicad cli to validate and generate gerbers.

I manufactured both PCBs via PCBWay (Claude also red all the specs and configured the constraints accordingly) and both work. I'm pretty impressed, especially because I've never done a PCB myself and didn't know I could.

What I was doing manually, co-working with Claude, was placing and routing (I was doing those via Kicad, then saving, then Claude was re-importing my work in his py files).

I did 2 pcbs recently with claude (opus 4.8 and 5). One was simple, the other pretty complex. I’ll open source them at some point.

I did manual routing myself, including most positioning, but claude was super helpful in identifying components, reviewing, etc. It was truly a collaboration.

These are my first pcbs ever, I don’t have any edu at all at making pcbs and was really excited to receive working prototypes. The other pretty amazing thing is that I have py scripts that generate the all the kicad files, as a human I wou look d have never done that (for the cost of building the py files) but I find it pretty neat.

One per device seems a good rule of thumb. If you have 2 laptops, each gets a diff ssh key, and then you config both in github/server(s).

It's slightly more painful to setup, but you get more flexibility later on in case you loose one key you have the other, and in case you need to cancel just one you can.

If you use hardware security keys, you get one per device naturally.

I believe the main one right now is digital identities. https://www.w3.org/reports/identity-web-impact/

Classic use cases would be like 1) show that you have a national id (like a passport) without revealing which one, 2) show that you are >= 18 without revealing your date of birth. More generally, assuming you have digital credentials with metadata, pretty much any statement can be proved in ZK (relatively efficiently, especially if the digital id is designed to be ZK friendly).

Nowadays you can run arbitrary verified computation at 100-300MHz using so-called zkvm. State of the art are probably risczero and succinct.

Practically speaking, write any program you want, compile it to riscv, imagine to run it on a pretty fast microcontroller, and in addition to the result you get a proof of correct execution.

I’d say it’s pretty practical, all major unlocks happened like in the past 3-4y and of course there’s a ton of research still happening.

This is the slow/generic version. For specific problems (aka dedicated circuits) it can be much faster.

There’s FIDO CXF/CXP: https://fidoalliance.org/specs/cx/cxf-v1.0-ps-errata-2026030...

To my understanding both Apple Passwords and the Android equivalent allow you to export passkeys to a different app (password manager), but I haven’t tried it yet.

If anyone has direct experience I appreciate to know how it was.

Nice. This is relatively old work (2020) but Pinterest had/has skin tone features, for example if you're searching for cosmetic products. Worth a check if anything might be useful for your work. https://medium.com/pinterest-engineering/powering-inclusive-...

The hardware: https://baochip.com

no, for now just multiple copies of the same repo. typically the agents work on independent things.

but I will try. the limit tbh is me and my ability to keep up with what’s happening. I work mainly on security/cryptography so I’m expected to audit the results of my work very carefully.

I have a tmux split in 4 panes, 4 agents running in 4 workspaces with shared memory. Most of the times I use 2 concurrently, but here and there I have extra analysis or reviews to do. I like this setup with tmux, works over ssh, easy to bring any one full focus, etc.

I recently added ml-dsa-44 to solokeys, both piv and fido2. To my understanding ssh+fido2 doesn’t support pq yet, but if anybody’s reading and knows how to make it happen, I’d be really interested.

I think I found a mini bug. Say I do a score of 10. Then I change board. Then I die with score < 10. I still have 10 for that board. It makes it easy to cheat the leaderboard :) Fun game!

For firmware dev, Claude is amazing. Just plug the dev board and tell him what you need. Great learning tool too!

I’m surprised that the Seeker is not even mentioned. I understand it’s Solana, it’s blockchain, whatever people think about it... to me it’s the only serious attempt at creating an app store that can compete with apple or google. What’s the point of independent hardware if the whole sw stack is them?

Same. To add some details, I used Authy because at the time it was the only app that would just work after upgrading my iphone. I never enabled their cloud mode, so only local 2FA codes.

I’m very familiar, we made Somu :) I meant USB-C.

I dream of an open board like the yubikey nano. This is very nice!

Try a prompt that helps claude iterate until it can verify the result.

For example, if you tell it to compile and run tests, you should never be in a situation with syntax errors.

But if you don’t give a prompt that allows to validate the result, then it’s going to get you whatever.

Has anyone tried to "turn off some cores" (eg using multi-instance gpu feature) and see if/how that increases reliability?

If you have the opportunity, try out a zen5. Significant improvements.

See also https://www.numberworld.org/blogs/2024_8_7_zen5_avx512_teard...

It’s explained in the link. I actually misremembered, it’s 196 bits.

Cool demo, but this is only log2(43 quintillions) = 65 bit security.

Kind of related is DiceKeys, with 192 bit security: https://www.crowdsupply.com/dicekeys/dicekeys