Where .env Went Wrong
secretspec.dev
secretspec.dev
> Environment variables only deliver values
Yes, the problem .env files try to solve is having "environment variables" be injectable from a file so that different applications can have different environment variables by default.
> A string is not a schema
Yes, input validation is an application concern. The application should know what these values represent / how to parse them and error if they're invalid.
I could go on, but the article is all about trying to use a hammer as a screwdriver and complaining that the hammer doesnt work.
This seems like a misunderstanding: The `.env` file should be `source`d into the current shell (environment). The application reads values using whatever mechanism it uses to read these values from the environment. Nothing should be “injected” into the application, i.e. the application should not read `.env` directly.
(Not sure if you were just being loose with your terminology, trying to clarify.)
Because well behaved applications play nice with their host system.
I have a strong suspicion that you’re thinking about application configuration, not environment configuration.
https://x.com/0xPrajwal_/status/2084163731534311627
https://x.com/nilbuild/status/2084339074639200488
The issue is that agents need structure and context.
Your local .env should NOT be shared, but should also be assumed to be leaked at any given time. Security should be done using a secrets manager through the cloud platform that's being used, e.g. AWS'secrets manager (or ssm param store too i guess)
Pasting it into chatgpt should not be a problem.
It also doesn't belong in git repos, but a much bigger issue is what process led to it. It's pretty standard to ignore it in a .gitignore, I'd be very surprised if modern agents made that mistake. And even if they did, agents should call tools that scaffold deterministically so that this isn't a problem.
I genuinely don't even know what we're talking about anymore, .env files are just values (it even says so in the article). People are making it way more complicated than it needs to be for no discernable reason other than an insane amount of laziness.
The difference is that we're providing an interface for applications to build with, with 8 SDKs available so you can have first-class support for secrets whatever you're bulding. I do hope fnox copies that too!
Those inherent similarities aside, I don't see it. They don't seem to have a lot in common as far as design or implementation or configuration.
I'm using 1Password's .env integration[1] and although the UX is a bit clunky, I really like it. My API keys are secure, tools that ordinarily support .env just work and there's a team-sharing feature too (although I'm yet to use it). It's pretty neat.
A big difference between our tool and many other similar tools is that we combine the schema and value setting into one surface, with a way of merging many definitions together, much like cuelang - but in a way that feels more intuitive. It's extremely flexible, and can even do credential brokering for untrusted workloads.
I've been enjoying the secretspec content lately, and watching it evolve :)
Varlock is clever. Sub string types for .envs pretty cool, why not
Personally, I tend to have a `source ./source-me-auth` which does stuff like `DB_PASSWORD="$( pass show blah.com | head -1 )"`, but that still means I end up with secrets dangling around in my environment. The "source-me" file is safe to share, but my runtime environment is tainted (for good and bad...).
I saw something w.r.t. the way open-claw handles things where you basically say `export DB_PASSWORD="REDACTED@SECRET_001"` and then it basically outbound filters network requests to hydrate that with `s/REDACTED@SECRET_001/$REAL_SECRET/g`, and I _really_ like that mechanism b/c it defers secret usage to runtime and keeps it physically "out" of the application itself.
Basically, having `with $SECRETS -- some_app.sh --some params ...` wouldn't be terrible (conceptually).
It allows you to generate a public/private key for each environment: test, dev, and prod (by default). The 'prod' private key can't be committed to the repository. Once the cryptographic keys are generated, you can encrypt any secret you want, and the framework will automatically decrypt it and allow you to access it as an environment variable at runtime.
The encrypted values for all environments are committed to the repository. For example, the file for the `$STRIPE_SECRET_KEY` environment variable for the 'dev' environment is a file that has these contents (shortened for readability):
<?php // dev.STRIPE_SECRET_KEY.c33678
return "\x1Dq\x11B6\x5B\xFE7\x9B\xA4\xFE...";
1. Keys for the 'dev' and 'test' environments can easily be shared with the team because their public and private keys values are committed to the repository. No more sharing secrets file over Slack or through some other mechanism.
2. The private key for the 'prod' environment can be stored in a 3rd party vault so it can be made available to the production servers. You can also decrypt 'prod' secrets during deployment to reduce the decryption overhead for each request.
3. At worst, agents would have access to the secrets in the 'test' and 'dev' environments only because they couldn't decrypt 'prod' values locally.
4. It forces good practices to ensure you're not using the same secret value for production and non-production environments.
It's made secret management so easy I don't even give it a second thought. Do other web frameworks support a system like this?
[0] https://symfony.com/doc/current/configuration/secrets.html
It may be time for Hackernews to integrate a Pangram detector into the UI, similar to what substack is doing :)
It's not exactly a wide breadth of varied text for such an all-encompassing task.
You can also just try it yourself I guess really what convinced me was how it perfectly agrees with my own judgement.
Now, I could just assume I'm an outlier within that small false-positive rate. But I find it far more likely that the efficacy of these tools is much lower than advertised and they rely on confirmation bias.
REDIS_URL=redis://localhost:6379
This is what went wrong with .env
Yes, I know
docker --env-file
Also see jt2290's comment:
Mise env vars management is much better. If you really need a script to generate env vars (like what you can do with direnv, there is a clean mechanism for that).
If you need to do something when you enter the directory of a project, there is also a mechanism for that.
Mise is really nice.
Are these valid observations so far?
- .env is the simplest and oldest / most boring solution. Secrets are passed to teammates by DM?
- .env.example seems like a nice home for documentation about secrets
- but I'd rather avoid writing sensitive credentials at all -- instead, teams can use a cloud-based secrets manager and the project fetches secrets at runtime. Is this ever a hassle / any downsides? How standard is this practice these days? I hear this also helps updating secrets so you don't have to e.g. tell every teammate when you rotate a secret. Generally not a fan of introducing a network call and a saas dependency, tho
- non-sensitive environment variables also need a home. I don't like the idea of cluttering the project with a .env.local or worse like .env.development.local
- I hear lots of secrets and config management happens in container orchestration, but what if I barely have a container in the first place? I guess docker compose is one of the simplest tools at this layer?
I guess I'm just looking for a safe, simple solution for a small team, and maybe the problem is that every team does things differently, and that many tools are marketed towards huge enterprise teams.
I don't see a reason to ever pass them. Sensitive production values should never be accessible to everyone, and for development the team should be able to create them themselves.
I usually keep very stupid values in .env.example, with the instructions to `cp .env.example .env`. In development it doesn't matter if the password is "password". In production the privileged admin sets up actual values.
> - non-sensitive environment variables also need a home. I don't like the idea of cluttering the project with a .env.local or worse like .env.development.local
`cp .env.example .env` and the application uses .env just like in production, only with dummy values.
- Secrets don't belong in config https://secretspec.dev/blog/secrets-dont-belong-in-config/
- You want to have flexibility of choosing between any secrets provider: https://secretspec.dev/blog/but-i-use-sops/
start with making it a dot file. Why would you want to hide the fact your app was using loaded environment variables?
I did discover one reason for it not to be just "env" though, which surprised me, which is that "source env" will yield "bash: source: /usr/bin/env: cannot execute binary file". I did not expect the source command to use the $PATH to resolve the filename. Probably some minor security issues that can result from that out there in the world. Arguably anyone loading it should use "source ./.env" or equivalent, with full path.
It's documented in the bash manual, of course. But it is rather complicated as to when it will and will not use $PATH.
. [-p path] filename [arguments]
From "The Open Group Base Specifications Issue 8 - IEEE Std 1003.1-2024" [2]:
. file
So it seems that it's not something Bash specific.
[1]: https://www.gnu.org/software/bash/manual/html_node/Bourne-Sh... )
[2]: https://pubs.opengroup.org/onlinepubs/9799919799/utilities/V...
Granted, but on the other hand: the entire history of this profession.
~/code
» cd foo
~/code/foo
» git init
± foo:main:/
» touch .env
± foo:main:/
» git add .
± foo:main:/
» g s
On branch main
No commits yet
Changes to be committed:
(use "git rm --cached <file>..." to unstage)
new file: .env
± foo:main:/
»
And even with it in .gitignore, I've dealt with multiple security incidents where someone has managed to commit it anyways. (And yes, I'm aware there are commands to do this, but what I haven't figured out is why someone would work around the safety and not think "what's the point of this?" prior to the bullet ending up in the foot.)
Type "dotprot" to copy your .env file into 1Password using the 1P CLI, it verifies it's there, then deletes the local copy of .env. When you're ready to work again, type "dotprot" again to restore the .env file from 1P. There is a .prot file in the directory that you can declare other files to store in 1P as well.
Varlock (https://varlock.dev ) is also good for this and more mature. I use it on some work projects, but for local dev, I wanted something to allow me to be even lazier than that.
I did a ^F for "update" and "dynamic" - didn't find any hits in that doc. Managing your dynamic environment values is table-stakes in most large deployments - and I didn't see any reference in `secretspec` as to how they plan to do that.
I love Django's approach to configs: your config is just a plain old Python file with some constants. As simple as possible, and works great.
This was really popular, and lots of people bought the factor III of only configuring via environment variables.
The whole point of it was: keep config simple. Env vars are austere and will keep you from complexity.
But people found a way to complicate them, so they could have both the 12 factor literal pass and the complexity they wanted.
That resulted in the mess we are today.
Not affiliated but those troubles are now finally solved with Infisical for me. It's open source so I happily give them a plug.
(Am I getting thrown out the window now? :)
Developers came up with .env because they would run apps on their own machine and wanted to pass variables to their application without having to set them in a command-line or environment-variable every single time they ran their app. And it works great for that. If you're one developer, running something locally, sure, just read some lines from a file. And if you're a team of developers, and most of you have the same lines you want to use, but maybe just a few of them you want to change per developer machine, fine, either keep .env out of Git, or keep a .env.local for the non-Git stuff. Again, simple, works fine.
Then you want to run your apps in production, with different lines. And then you want to run it on a test/stage/qa machine too - again, different lines. And maybe you hard-code those into separate files (.env.prod, .env.staging, .env.test) and load them depending on which of the 2 or 3 machines you have running your app. Again, simple, works fine.
Until the problems start.
Secrets in the .env? Now those are in the code, which gets cloned everywhere, and can be stolen. "Encrypted" secrets in .env? Now you have to manage a secret key out of the file, in addition to the encrypted secret in the file. Everyone has access to the secret? Now people can use those secrets, or access different machines, potentially creating problems or exceeding their authority, and there is no way to know who did what because it's one shared secret. Somebody leaves the company? Now you need to rotate the secret (which nobody does).
Ephemeral containers/deployments? Now the .env entries don't match the new hostname. Want to support multiple hosts? One hostname in the .env line doesn't work anymore. Want to scale horizontally? Now your "prod/test/dev" files are more of an environment type than a specific host. Your RDS database's hostname, or an old IP address, has changed? App is broken, time to update all the .env lines referring to it and re-deploy the app.
These are all problems that you might or might not run into. But they are problems that do exist in the world; we know they happen, and we know how to avoid them. You decide whether you're going to wait for them to bite you, or avoid them altogether from the very start. If you do the former, you're acting like a Systems Engineer, designing a system to be resistant to known failures. If you do the latter, you're not.
You may not want to do extra work you feel is unnecessary just to avoid a possible problem. But other professions do this anyway, by regulation, because (for example) as a society we we don't want to allow houses to burn down from a preventable problem. Example: If you run wire in a conduit, the conduit must be a minimum size depending on the number and type of wires you run in that conduit. You may think it's annoying that you have plenty of space left in your conduit; why should I have to change my wire size or conduit size? But if the wires take too much current, and don't have enough airflow/space between them, they can heat up and start a fire, or a short, which could cause a larger problem somewhere else (like taking out a hospital's ventilators, as one example). Doing the extra "unnecessary" work prevents fires. That's why Systems Engineers don't use .env files. Their job is to build reliable systems, not just roll out a feature and cross their fingers.