keenv

keenv puts secrets from a KeePass database into the environment of one command and nowhere else. The values never reach a file, an export, or the shell history: they exist between unlocking the database and execing the command, and the process that held them is replaced.

It is the tool the Oberon Systems keyring policy names for local secrets, and the way indech hands the Cloudflare R2 keys to OpenTofu.

Contents

• Why
• Installation
• Configuration
• Usage
• How it works
• Development

Why

The usual ways of getting a secret into a process all leave it somewhere: export puts it in every child of the shell for the rest of the session, a .env full of plaintext puts it on disk, and eval $(something) puts it in the history file as well.

keenv reads the value out of the database at the moment it is needed, builds the environment for exactly one command, and replaces itself with that command. There is no keenv export and no keenv eval, on purpose.

Installation

pip install keenv

The database itself is read in process through pykeepass, so KeePassXC does not have to be installed.

Configuration

keenv reads two files, both optional. keenv.yaml names the database and maps environment variables onto entries:

vault: ~/Dropbox/oberon.kdbx
keyfile: ~/.keys/oberon.keyx

env:
  AWS_ACCESS_KEY_ID:
    entry: Oberon/R2/indech-state
    field: username
  AWS_SECRET_ACCESS_KEY:
    entry: Oberon/R2/indech-state
    field: password

.env is the same mapping in the shape people already write, where a value may be a keenv:// reference instead of a literal:

AWS_ACCESS_KEY_ID=keenv://Oberon/R2/indech-state/username
AWS_SECRET_ACCESS_KEY=keenv://Oberon/R2/indech-state/password
TF_LOG=INFO

A reference is keenv://<entry path>/<field>. The last segment is the field and everything before it is the path to the entry, so a field whose name contains a slash cannot be addressed. username, password, url, notes and title are matched case-insensitively; any other name is looked up as a custom attribute, with its spelling preserved.

The entry path starts at the top-level groups, one level below the root group KeePass shows at the top of its tree, so Oberon/R2/indech-state and not Root/Oberon/R2/indech-state. A path that does name the root group first is accepted too, under either the group's real name or a plain root, so a path copied straight out of KeePass works as it stands.

Values that are not references pass through literally. A # only starts a comment at the beginning of a line, never in the middle of one, because a secret may contain it.

The layers apply in this order, each one overriding the last:

A file that is not there is an empty layer, not an error. Having nothing to resolve after both layers is an error.

keenv.yaml is validated against a pydantic model that rejects keys it does not know, so vualt: is reported as a mistake rather than quietly ignored.

Usage

Run a command with the resolved environment:

keenv run -- tofu -chdir=circuits/live/ramnode/compute plan

Check that every reference still points at something, without printing any value:

keenv check

Expected output:

AWS_ACCESS_KEY_ID            keenv://Oberon/R2/indech-state/UserName               20 chars  keenv.yaml
AWS_SECRET_ACCESS_KEY        keenv://Oberon/R2/indech-state/Password               40 chars  keenv.yaml
TF_LOG                       literal                                                4 chars  .env

The last column is the file the variable came from, which is the quickest way to see that a .env is overriding keenv.yaml.

Point at another database and another mapping:

keenv run --vault ~/other.kdbx -e deploy.env -- ./deploy.sh

Exit codes are 0 on success, 1 for anything keenv can explain, 2 for a usage mistake, and 127 when the command does not exist. Otherwise the exit code is the command's own, because the command replaces keenv.

How it works

1. Both layers are read and merged into one list of variables. A name defined in both takes its value from .env, and keenv check names the file each variable came from.
2. If any of them is a keenv:// reference, the database is opened once. The master password is asked for on /dev/tty, never on stdin, so a password prompt can never swallow the first line of a pipe. A key file replaces the prompt.
3. Every reference is resolved from that one open database.
4. The resolved variables are laid over a copy of the current environment and handed to os.execvpe.

Step 4 is what keeps the secrets contained: execvpe replaces the process image, so nothing that held the values is still running once the command starts, and the shell that invoked keenv never saw them.

Development

make init
make test
make lint

make init creates .venv, installs the package in editable mode and wires up the pre-commit hooks. The test suite builds its own throwaway database in a temporary directory; no test opens a real vault.

Commits go through commitizen:

.venv/bin/cz commit