Mmmm vibecrypto, my favourite. I don't see anything obviously broken (at a glance) but as a perf improvement, there's little reason to use Argon2id for the "verification hash" step, might as well use sha256 there. There is also no need to use ConstantTimeCompare because the value being compared against is not secret, although it doesn't hurt.
The "Crash-safe rotation WAL" feature sounds sketchy and it's what I'd audit closely, if I was auditing closely.
Thanks for the look. On the verification hash, you're right, SHA256 would work there. Argon2id was overkill, I agree 100%.
The crash-safe WAL is the part I'm most nervous about too. That's exactly why I posted this. I want eyes on the rotation logic specifically.
And yeah, single bbolt db is a limitation. I could have used pebble or any other, but trade-off for simplicity (a single *.db). A true WAL will need external file. The storage is pluggable though also open to improvement.
Per-bucket DEKs with HKDF, hashed policy keys to kill enumeration, HMAC audit chain. This is the kind of boring-correct crypto design I rarely see in Go libraries. memguard for the master key is a nice touch too.
Definitely … agents cannot access your password unless you save it to the environment too. However it's better to use resolvers ... depending on your use case
Honestly… the initial use case is to hide certs from the file system and secrets from the environment. However, this can be extended.
The primary issue has been not being able to manage an encrypted storage system… the main goal is to have something that can be audited, not just secured.
Hey I ran this request through my AI harness (beigeboxoss.com), first with a smaller local model and then validated with Trinity Large via OR. https://github.com/agberohq/keeper/issues/2 -- YMMV but wanted something to do with my coffee, thanks!
The first bug has been confirmed however The second `vulnerability` would only be exploitable if an attacker could also break SHA-256 preimage resistance to forge valid checksums ??? correct me if am wrong
Not sure, I've seen common things like this pop up a lot too, the same errors being tripped over. I'm not sure if it is a context thing or just a limitation of how the models work presently? For stuff that I'm using myself, I will run these through like the top 10 reasoning models on OR and just see where everything pans out.
Thanks for sharing this. secret looks really well thought out, the three-layer key hierarchy is impressive. And using `age` is a solid choice. once considered it.
Different trade-offs though, Keeper is library first embedded. secret does per version keys with symlink switching - nice, Keeper does per-bucket DEK isolation + audit chains. Both solve "encrypted local storage" but for different workflows.
I'll definitely be looking through your code for ideas
I haven't used it, don't advocate for it, and have no opinion on either its viability or your product's viability for any specific use case. Mostly I just think it's a bit confusing to have two separate products in a very similar space with the same name.
Mmmm vibecrypto, my favourite. I don't see anything obviously broken (at a glance) but as a perf improvement, there's little reason to use Argon2id for the "verification hash" step, might as well use sha256 there. There is also no need to use ConstantTimeCompare because the value being compared against is not secret, although it doesn't hurt.
The "Crash-safe rotation WAL" feature sounds sketchy and it's what I'd audit closely, if I was auditing closely.
Thanks for the look. On the verification hash, you're right, SHA256 would work there. Argon2id was overkill, I agree 100%.
The crash-safe WAL is the part I'm most nervous about too. That's exactly why I posted this. I want eyes on the rotation logic specifically.
And yeah, single bbolt db is a limitation. I could have used pebble or any other, but trade-off for simplicity (a single *.db). A true WAL will need external file. The storage is pluggable though also open to improvement.
Still very young.
Name could conflict with Keeper Security
So have been told. Will definitely look for a better name
Per-bucket DEKs with HKDF, hashed policy keys to kill enumeration, HMAC audit chain. This is the kind of boring-correct crypto design I rarely see in Go libraries. memguard for the master key is a nice touch too.
I was thinking its better to be boring-correct :)
Could I use this to store secrets to hide env vars from agents?
Definitely … agents cannot access your password unless you save it to the environment too. However it's better to use resolvers ... depending on your use case
I have been looking for something like this. I know openbao, hashicorp vault.
But they require to be placed on a separate server, and come with their own infra management.
Is the idea of this project to embed this into you app, instead of relying on .env or an external vault?
Honestly… the initial use case is to hide certs from the file system and secrets from the environment. However, this can be extended.
The primary issue has been not being able to manage an encrypted storage system… the main goal is to have something that can be audited, not just secured.
yes 100% ... embeded
Genuine question: what's your thread model?
Vault gives time limited Tokens with Network Boundary. Instead of Keeper, i would just use age:
# write
echo "my secret" | age -r <recipient-pubkey> > secret.age
# read
age -d -i key.txt secret.age
https://git.eeqj.de/sneak/secret
This is an age+filesystem secrets manager that I made that is basically what you wrote, but with more organization.
not when you need an audit system
True, but AFAIK an audit system is worthless if it resides on the same potentially compromised machine, no?
Hey I ran this request through my AI harness (beigeboxoss.com), first with a smaller local model and then validated with Trinity Large via OR. https://github.com/agberohq/keeper/issues/2 -- YMMV but wanted something to do with my coffee, thanks!
The first bug has been confirmed however The second `vulnerability` would only be exploitable if an attacker could also break SHA-256 preimage resistance to forge valid checksums ??? correct me if am wrong
> The VerifyHMAC() function unconditionally returns true when the HMAC field is empty
This kind of thing is super common in vibecoded crypto, I wonder why it keeps happening.
Not sure, I've seen common things like this pop up a lot too, the same errors being tripped over. I'm not sure if it is a context thing or just a limitation of how the models work presently? For stuff that I'm using myself, I will run these through like the top 10 reasoning models on OR and just see where everything pans out.
Edit: here is an example of the process and output with something I put together the other day: https://github.com/RALaBarge/garlicpress/blob/master/portfol...
Even when you have a proper function and use AI for auto documentation, it silently changes it (insane) … I will defiantly fix this.
I have a similar one called “secret”, also in Go, that is more CLI-focused and uses the filesystem as database.
https://git.eeqj.de/sneak/secret
Thanks for sharing this. secret looks really well thought out, the three-layer key hierarchy is impressive. And using `age` is a solid choice. once considered it.
Different trade-offs though, Keeper is library first embedded. secret does per version keys with symlink switching - nice, Keeper does per-bucket DEK isolation + audit chains. Both solve "encrypted local storage" but for different workflows.
I'll definitely be looking through your code for ideas
Keeper is already the name of a popular enterprise secrets store: https://docs.keeper.io/en/user-guides/web-vault
I haven't used it, don't advocate for it, and have no opinion on either its viability or your product's viability for any specific use case. Mostly I just think it's a bit confusing to have two separate products in a very similar space with the same name.
thanks for the update ... will definitely look for a better name