Show HN: Keypub.sh – OAuth for the terminal using SSH keys | Hacker News Reader

Show HN: Keypub.sh – OAuth for the terminal using SSH keys

keypub.sh

222 pointsmessh69 comments

Hi HN! I built KeyPub.sh to solve the problem of user verification for CLI applications. It's essentially OAuth for the terminal, but using SSH keys that developers and users already have.

- No installation needed - works with existing SSH setup - Privacy-focused: users control what email info is shared - Simple email verification process - Free public service - Perfect for CLI app developers who don't want to build user verification

Try it with: `$ ssh keypub.sh about`

Source code: https://github.com/skariel/keypub

Open articleView on HN

Sounds like a similar family of problems to [Wish](https://github.com/charmbracelet/wish) by charm.sh. They've been pushing this curious paradigm of "ssh apps", where ssh keys are used to automatically create identities for small self-hosted CLI/TUI apps.

As a useful comment for messh, it looks like you've committed the ssh_server binary file to git; you may want to add that to gitignore, as binary file handling isn't a traditional git strength. I _think_ it's better than it was a decade ago when I last investigated this, but I can see that Pro Git still recommends explicitly setting gitattributes to mark a file as binary https://git-scm.com/book/en/v2/Customizing-Git-Git-Attribute...

Regarding ssh-apps, this seems super duper niche to me even for developers. Like, I get that many devs are terminal savvy, and it's cool that it's possible to serve apps like this but is it practical (to pick for your new product /internal app)?

We are pretty successful over at https://pico.sh using the same techniques — and using wish

As a bare metal cloud service provider, we built a little TUI app over ssh so that people could upload their public keys to us without having to use a web interface. It has worked out really well because it makes it possible to "upload" the public key without having to do anything but login.

We automatically configure ssh on the servers to auth against our own internal api for serving up the keys. This way, customers can add to authorized_keys and also use our TUI for management as well.

As a developer, I agree. Give me a great CLI, please, not a server rendered TUI.

This gives me another thought though, a "server-rendered" CLI. A tiny shim binary that just sends argv to the server, and the server sends back stdout/stderr. Haven't seen anyone try that.

That sounds very similar to the command form of ssh i.e. "ssh foo.example.com ls" :)

a "server-rendered" CLI is exactly what `keypub.sh` is, and what many ssh apps are (some are more TUI)

Yeah, in retrospect, I'm an idiot. I wasn't considering `ssh foobar.io dothething` to be equivalent to what I was thinking in my head even though it is exactly equivalent.

I'm working on a different approach that doesn't use ssh at https://terminalwire.com/.

It's meant for inherently server-based CLI's like Github, Stripe, Fly, Heroku, or any other SaaS and can be plugged directly into web frameworks. My thinking is that more companies would ship command-line interfaces if they didn't have to maintain APIs for them, deal with distributing binaries (and all the compatibility issues that come with that), or deal with all the configuration involved in hooking up SSH to their app servers.

It's in beta at the moment, and I'm currently focused on Rails apps to get the ergonomics dialed-in. My goal is to make it possible for people in any web framework to build and ship a CLI to customers in under an hour and never have to touch an API or worry about binary distribution for as long as their CLI exists.

At my org the biggest issue with out CLI is not the initial distribution, but supporting and maintaining multiple platforms, code signing the binaries, and making sure customers actually update the binaries.

Yeah, I put all of that under the umbrella of “distribution”—the last mile is a huge pain in the SaaS that SaaS CLIs shouldn’t have to deal with or even think about.

Terminalwire will handle the builds for all the different platforms, signing, etc. so companies can instead focus on shipping their unique features to customers.

I’m also going to ship clients that automatically update to match the version of the client that’s compatible with the server in a way that the client won’t even notice.

At the large enterprises I've worked at, signing with our certs had to happen on our own build machines. And I don't think there was any precedent if we could sign third party binaries that had been built for us.

Having some my-company.terminalwire.com as the signing identity could work; I bet most SMB wouldn't bat an eye to ship like that.

hi, good catch of the binary! (will fix)

`Wish` and also `terminal.shop` were both great sources of inspiration. I hope to see many more ssh apps in the future. I'm already working on the next one

Another interesting set of ssh apps is https://pico.sh/ . These focus on networking and lightweight hosting (pastes, simple blog), all via ssh or rsync.

Looking at their demos immediately brought back the memories of parsing TUI's using ad-hoc "expect" rules.

I honestly don't get the point of TUIs...

A real command-line interface is extremely useful - it's trivially scriptable, works direct or with ssh, scroll-back buffer logs what happened, commands stay in my shell history, I can copy-paste commands to a friend over chat system, or into shell script, and it's easy for app authors too. Its my first choice for my apps.

If low-color fixed-width character grid is not cutting it, then native app or web app is a great second step, or even intermediate solutions, like generating HTML files from CLI tool and opening them in default browser. You have to invest some effort, but you have infinite ways to design your interface, and it's still in user's familiar environment.

But those charm.sh TUI applications seem utterly useless and highly annoying. They give up all of the terminal advantages: my shell's history is useless, my favorite ways to edit commands does not apply (as they have their own editor), I cannot scroll back and see what the program just printed, I cannot script it, I cannot log it, I cannot search it, I cannot redo the previous action, the color scheme is not the one I've picked... At least with web apps I can parse html and/or hit underlying API directly - no such luck with TUI apps.

At least the good news is that TUI apps are not getting any traction, and I can completely understand why.

> At least the good news is that TUI apps are not getting any traction, and I can completely understand why.

lazygit (54k stars on GitHub) and k9s got a lot of traction and are very popular:

https://github.com/jesseduffield/lazygit

https://github.com/derailed/k9s

I totally agree with you, I prefer good autocomplete to a TUI for all these reasons (and more). Well, I actually prefer any CLI to a TUI. The most TUI thing I use in my toolkit is `top` (yes, the "classic" one)

> I honestly don't get the point of TUIs...

There are one or two advantages over regular GUIs, but that's it.

The biggest is probably that they are lightweight since there are no GUI library dependencies (and if there are TUI ones, they are usually much lighter than their GUI sisters). This also means there are fewer (if any) dependencies to distribute compared to a GUI.

The only other advantage I can come up with is that a TUI will have to be usable by keyboard only (in almost all cases). This is not a given for regular GUI libraries.

I'm not a fan of TUIs either. I think the only one I am using regularly is `tig` (https://jonas.github.io/tig/). I guess the reason is that I don't have to remember the git revision list syntax that way and that `tig` allows for easy commit searching with `/` ...

TUI's have one thing going for them: Copy paste anything! IntelliJ doesn't let me copy any text it shows me.

Also, TUI's have their place. I haven't looked much at GUI alternatives, but k9s is really great.

Would also assume that interactive apps are simpler to implement if it's a TUI.

The TUI apps I use regularly are:

* lf (file manager) * vim * htop * moc and ncmpcpp (music) * emacs (which is its own world)

TUI apps are great for one-time actions that needs feedback, especially having a list of something, filtering it, and applying an action on one or more items.

But, I still like to have a CLI version so I can script any reccuring workflow.

When you say "OAuth", what exactly do you mean? It looks like this is really just a tool for associating email addresses with SSH keys.

This is an interesting concept, but it smells a bit like a solution in search of a problem. Perhaps it will feel more useful to me once there are two or three SSH apps that I want to access. Even then, I would suggest that prospective SSH app developers just lean on github's public SSH keys instead, as basically all developers will have a github account and this reduces your (already high, relative to webapp) startup friction.

github public ssh keys are not connected to mails. and using the api from the terminal requires installing some tool.

I hope there will be lots of apps for the the terminal, for e.g. cde (cloud dev env) managing, task management, project management, compute as a service, etc.

> using the api from the terminal requires installing some tool.

It really doesn’t. It’s just a HTTP API. So basically anything that can form a TLS connection will work.

But the bigger problem here is that email addresses can be private in GitHub. Also you never know when GitHub might change their API.

``` ).WHERE( AND( table.SSHKeys.Email.EQ(String(to_email)), table.SSHKeys.Fingerprint.EQ(String(fingerprint)), ), ) ```

If not using a `citext` column then you're going to want to normalize (ie downcase/tolower) everywhere you're doing arbitrary string comparisons, or you're going to get incorrect counts.

Also I don't see any null or "" checking taking place before querying...

I'm not going to trust that your service can give me any reasonable confidence about the identity of the ssh key or the email it's attached to.

That aside, I'm not understanding what the goal here is. I've never once needed my ssh key tied to my email address, but if I did, it's included in the public key already...

there is email validation, and the public key is handled between the ssh client and server. If there is no valid public key this code would not even run. The fingerprint is prepared from this valid public key.

Regarding having the email in the ssh pub key: maybe it is there, but it is no validated. Anyone could write anything there

No SSHFP record, TOFU clients. This is not secure.

I wish people would stop trying to use SSH for things like this. PKI has features that are missing here and those features matter.

This is unsafe.

"but raggi, mitm before tofu is a really unrealistic scenario"

ok, well, consider that some large percentage of gas stations in the US have hardware installed to skim your credit cards. those same folks are perfectly well motivated to drop a wifi dns mitm in conference buildings (trivial). new tech conference, handful of credit cards as people gossip about exactly these kinds of things in the hallway track. the roi on these installs would be pretty high, because tech folks tend to have a high credit balance. so yeah, "totally unrealistic" (this is more about terminal.shop, but same principle here as soon as someone uses it for actual value).

Are you serious? A MITM here cannot sign things with your private key. People who care can add the keypub’s fingerprint before connecting. Adding an sshfp record is completely trivial too. There is no more tofu here than any other ssh connection and the solution is the same. And the added advantage of a PKI means, in a properly designed system, a threat actor can’t do anything useful sitting on the wire. I’m not saying ssh apps are going to take off, just that your reaction is quite hyperbolic.

There’s no separate signing in use here except for the ssh connection, which can be trivially mitm’d in common targeted scenarios because of the lack of webpki and the lack of other preparations. SSHFP would help, but only if configured both in dns, and if the client is both configured to look for it, using secure dns, and the user understands the failure UX and doesn’t just bypass it. On DNS: DoH would help but it is only in widespread use in browsers. DoT would help but it is only in widespread use on android.

In addition to this a further scan of the code reveals it’s also using a btree index lookup for code comparison and no limitation on attempts, so it is likely that this is relatively trivial to attack with timing as well.

Trivialize mitm all you want, you say concerns of mitm are hyperbolic, I gave a practical example of a target rich environment and there are plenty more folks could come up with. SSH may have long been skirting the lack of a better host key distribution system, but this is largely a matter of luck, access and bespoke usage. These new deployments demonstrate a change on two of these factors, increasing risk substantially if this grows.

I agree there is an issue here -- btw, you would notice when eventually the server key changes.

User friendly and secure-by-default clients will leverage the domain HTTPS CA to solve this (fetching the server key using https). The downside is that it will require d/l and install

- TOFU: Trust on first use (https://en.m.wikipedia.org/wiki/Trust_on_first_use)

- SSHFP: SSH FingerPrint record in the domain name system (https://en.m.wikipedia.org/wiki/SSHFP_record)

- MITM: Man in the middle (https://en.m.wikipedia.org/wiki/Man-in-the-middle_attack)

You missed some.

- SSH: Secure SHell (https://en.wikipedia.org/wiki/Secure_Shell)

- PKI: Public Key Infrastructure (https://en.wikipedia.org/wiki/Public_key_infrastructure)

- US: United States (https://en.wikipedia.org/wiki/United_States)

I agree with other commenters that the “oauth” keyword in title is misleading. I assumed this was compliant with oauth spec.

It’s an interesting service, but I doubt I would use this for anything serious beyond toy apps.

There was an article in October mentioning CF acquiring a company that dabbled in this space though. [1]

It’s more for managing authN/Z for internal infra (k8s, servers,) but via existing IdP (with MFA). Underlying tech is “openpubkey”. [2]. I can see this adapted towards authN for CLI apps.

Wanted to test this on my homelab infra but life has gotten in the way.

[1] https://news.ycombinator.com/item?id=41923429

[2] https://github.com/openpubkey/openpubkey

Oauth usually implies some variability of access - ie selective permissions. Is this more a pure authn layer as opposed to authz?

You have a repo on GitHub... Have you looked at using account public keys for anything? Ie https://github.com/hpsin.keys I hear a lot about how those keys should get used to bootstrap pki systems but I've not seen it happen yet.

Ubuntu's installer supports bootstrapping ssh access with this method. Put in your GitHub username and it takes care of the rest. Works for interactive and headless installs.

The `ssh-import-id` package is the basic mechanism for this.

it does have authorization with the `allow` and `deny` commands for granular control of who can can see what. Currently very simple, only for email but maybe other fields in the future

Care should always be taken when using an SSH client to connect to untrusted hosts. Make sure you’re not actually forwarding your SSH agent to the remote host, or they’ll be able to hijack your keys. Consider also that any output is processed by your terminal, and there have been a number of serious security bugs in terminal escape sequence handling in a variety of terminal emulators in the past.

"Hijack your keys" is somewhat overstating it, all they can do with it is log into other hosts that you can log into, assuming key auth is the only factor. And only while you are connected to the original host.

Terminal security bugs are not more or less serious than your average Chrome 0day.

My yubikey always requires a physical touch so even if the agent is forwarded my keys can't be used. And I'll notice when it tries because I see it flashing.

Best to use an agent that prompts you for signing. 1Password and Yubikey do this. Are there others?

OAuth means a specific technical thing, and this is not that thing.

"X for Y" should make you draw a metaphorical, not literal comparison to X - it is something like X, applied in the context/audience/etc. of Y. "Tinder for boats"¹ is not Tinder.

¹https://www.buzzfeed.com/michellerial/is-your-startup-idea-a...

Based on current downvotes it seems that people object to this. And I did take this literally myself before going to the comments. Still, I think the metaphorical thing has value as a succinct explainer. Maybe a small change in wording or some quotation marks is the way to go?

Part of the issue is that "bootstrapping real oauth with ssh" is something people actually want, so there's some (unearned) disappointment in finding out that this isn't it.

I have no horse in this race, was just commenting from a generic point of view. The project's README.md says "Lightweight alternative to OAuth for CLI applications" in section "Use Cases", and there is no further mention of OAuth in it, maybe that's more acceptable.

Indeed. It also doesn't have anything like PKCE.

PGP keys are associated with emails and there are various fully decentralized options to discover these keys (keyservers, WKD, ...). (How) does Keypub improve on this?

Best practice:

Use SSH keys for SSH connections and authentication

Use PGP keys for email encryption and file signing

Keep these systems separate as they're designed for different purposes

PGP is hairy and many developers don't have it installed and configured. It's full of non-obvious options and capabilities.

SSH is a common tool which most developers have installed, keep in a working order, use often, and reasonably understand.

Guess which tool is easier to piggyback on and achieve easier and faster adoption.

I like PGP keys for ssh auth because they can be hardware backed through openPGP. That makes it a great option (and also the actual crypto is then executed on the key itself, not by the gpg software)

You can do this with ssh keys too. And ssh keys can sign git commits. There’s no reason these days to even install a pgp suite. Stop using it.

Why not use the device grant?

It's designed for this use case (among others), is standards compliant, and leverages the OAuth ecosystem.

RFC: https://datatracker.ietf.org/doc/html/rfc8628

An example golang CLI tool that uses it: https://fusionauth.io/blog/building-cli-app-with-device-gran...

I love that you are approaching this space. SSH keys are an under-valued identity credential. Most big companies i've worked at have a clumsy sync & storage system that I think you could sell to.

I'd encourage the next version using a more distributed approach (instead of a centralized credential DB & email token). A gpg signature or CA (using openssl) are alternative trust models that would provide trust without the DB

I'm not sure I understand what this is useful for. I have dozens of ssh keys I use for various services. Do I register all of them?

Few years ago, I have worked to integrate SSH trough a major IDP via PAM modules. It was a bit tricky (especially for a bug in PAM in CentOS) but in the long run it worked fine.

"about" didn't work for me but "help" did.

"why" doesn't work either.

will add them both. Initially I looked at "about" and "why" as suitable only for the landing page

That's pretty reasonable, although the styling of the landing page makes it look like those are actual command outputs, so I expected them to work.

I really wish I could have the opposite. To authenticate with my SSH key (openpgp on yubikey) on the web.

I'm not really happy with the Fido options.

Why? What is wrong with fido?

Several things. I can't use it for file encryption as well, and it's not possible to make a software key as backup (as far as I know) in case I lose both my yubikeys. With pgp I have a break glass key on an encrypted pen drive.

Also, with SSH it requires a recent version which is not on all the boxes I need access to.

It requires pin code entry for every authentication unlike openpgp where I enter the pin once and then I just touch it for every subsequent authentication. I like that because I use it not only for login but also for other stuff like root elevation.

And I think agent forwarding doesn't work with it. It's not really something I've looked into though as it's just not something I can use yet for the other reasons.

For now pgp is just much better for me, though I wish I could use it on the web as well.

Nice job! Have you thought about implementing 2fa? I know platform.sh supports 2fa via ssh.

Just a reminder for everyone that by default SSH leaks all of the below public keys (if they exist) + all public keys in your ssh-agent to every server you connect to.

    ~/.ssh/id_ecdsa.pub
    ~/.ssh/id_ecdsa_sk.pub
    ~/.ssh/id_ed25519.pub
    ~/.ssh/id_ed25519_sk.pub
    ~/.ssh/id_xmss.pub
    ~/.ssh/id_dsa.pub

Your "leaked" public keys can be used to check if a certain server recognizes that key.

Loading an alternate client, like Dropbear or PuTTY for Linux is one way to avoid this.

Or just set `IdentityFile` with the path to the key and set `IdentitiesOnly` to yes for the specific host you connect to, and disable public key authentication by default for everything else.

Looks interesting but your tls cert isn’t valid on your site

What do you mean?

https://www.ssllabs.com/ssltest/analyze.html?d=keypub.sh says the cert is OK.

> Lightweight alternative to OAuth for CLI applications - just use SSH keys that users already have

I'd love to see more support for this application. I've contributed a number of oauth-based CLI apps and the user onboarding step (3-legged init) is atrocious.

Reply on news.ycombinator.com