Loading secrets at runtime: op read, op inject, op run

Three commands solve three different problems. Pick by destination, not by habit.

op read — single secret to stdout

op read [flags] <secret-reference>

Reference query parameters

Append ?attribute=... or ?ssh-format=... to extract derived values:

op read "op://prod/2fa/otp-field?attribute=otp"            # current 6-digit code
op read "op://prod/login/password?attribute=type"          # field type metadata
op read "op://prod/ssh-key/private key?ssh-format=openssh" # OpenSSH-formatted key

Inline in commands

This is fine when the secret is needed once and won't end up in ps output for long:

docker login -u "$(op read op://prod/docker/user)" -p "$(op read op://prod/docker/pass)"
psql "$(op read op://prod/db/connection_string)"

For long-running commands, prefer op run — exposing secrets via $(...) puts them in the parent shell's argv briefly.

op inject — render a templated file

op inject [flags]

Template syntax

Two reference forms are accepted in templates:

unenclosed:   op://vault/item/field
enclosed:     {{ op://vault/item/field }}

Use enclosed in nearly every case — they're robust around adjacent text:

database_url: postgres://{{ op://prod/db/user }}:{{ op://prod/db/pass }}@{{ op://prod/db/host }}/{{ op://prod/db/name }}

Unenclosed references end at the first character outside [A-Za-z0-9_.?\-], which makes adjacent characters dangerous.

Environment-variable interpolation in templates

Templates can also expand env vars, with default-value support:

${VAR}            ${VAR:-default}
$VAR              op://${ENV:-dev}/db/password

Combined: a single template handles dev/staging/prod by exporting ENV differently before each op inject.

Streaming when you don't need the file on disk

op inject -i nginx.conf.tpl | docker run -i --rm nginx nginx -c /dev/stdin -t
op inject -i app.yaml.tpl | kubectl apply -f -

This avoids the "delete the resolved file when done" rule entirely.

When you must write the file

Always treat output as plaintext-secret material:

op inject -i config.yml.tpl -o config.yml
trap 'shred -u config.yml 2>/dev/null || rm -f config.yml' EXIT
# ... start app that reads config.yml ...

Add the resolved filename to .gitignore.

op run — exec a command with secrets in env

op run [flags] -- <command> [args...]

The dotenv pattern (recommended for app dev)

.env contains references, not values. Commit it. The plaintext secrets never touch disk:

# .env — safe to commit
DATABASE_URL=op://prod/db/connection_string
STRIPE_SECRET_KEY=op://prod/stripe/api-key
JWT_SIGNING_KEY=op://prod/auth/jwt-key
SENTRY_DSN=op://prod/sentry/dsn

op run --env-file=.env -- npm run dev
op run --env-file=.env -- python manage.py runserver
op run --env-file=.env -- ./bin/server

Multi-environment with one template

.env references a variable substituted at run time:

# .env
DATABASE_URL=op://${APP_ENV}/db/connection_string
API_KEY=op://${APP_ENV}/api/key

APP_ENV=staging op run --env-file=.env -- ./bin/server
APP_ENV=prod    op run --env-file=.env -- ./bin/server

Chained references (env var → secret reference)

Set the env var to a reference, then use a subshell so $VAR expands inside the subprocess (after op run has substituted):

DB_PASS=op://prod/db/password op run -- sh -c 'echo "$DB_PASS" | psql ...'

The wrong way (broken):

DB_PASS=op://prod/db/password op run -- echo "$DB_PASS"
# parent shell expands $DB_PASS to the literal "op://..." string before op run sees it

Masking

By default, secrets passed via env are redacted from the child process's stdout/stderr. Helpful for screenshare/CI logs. Disable with --no-masking (or OP_RUN_NO_MASKING=true) only for debugging.

Multiple env files

op run --env-file=.env.shared --env-file=.env.local -- ./bin/server

Later files override earlier files for the same key.

CI / production patterns

GitHub Actions

- uses: 1password/load-secrets-action@v2
  with:
    export-env: true
  env:
    OP_SERVICE_ACCOUNT_TOKEN: ${{ secrets.OP_SERVICE_ACCOUNT_TOKEN }}
    DATABASE_URL: op://prod/db/connection_string
    STRIPE_KEY: op://prod/stripe/api-key
- run: ./bin/migrate && ./bin/start

If you don't want the Action, raw op works:

- run: |
    curl -sSf https://downloads.1password.com/linux/keys/1password.asc | sudo gpg --dearmor -o /usr/share/keyrings/1password-archive-keyring.gpg
    echo "deb [signed-by=/usr/share/keyrings/1password-archive-keyring.gpg] https://downloads.1password.com/linux/debian/$(dpkg --print-architecture) stable main" | sudo tee /etc/apt/sources.list.d/1password.sources
    sudo apt update && sudo apt install -y 1password-cli
    op run --env-file=.env -- ./bin/migrate
  env:
    OP_SERVICE_ACCOUNT_TOKEN: ${{ secrets.OP_SERVICE_ACCOUNT_TOKEN }}

Docker (build vs run)

Don't bake secrets into images. Inject at run time:

docker run --rm \
  -e OP_SERVICE_ACCOUNT_TOKEN \
  -v "$PWD/.env:/app/.env:ro" \
  -v "$(which op):/usr/local/bin/op:ro" \
  myapp:latest \
  op run --env-file=/app/.env -- /app/bin/start

Better: use the official 1password/op image multi-stage to bake op into your runtime image, then run as above.

Kubernetes

Use the 1Password Operator or render Secret manifests with op inject at deploy time:

op inject -i secret.yaml.tpl | kubectl apply -f -

Anti-patterns to flag

• A .env file with plaintext values committed to a repo. Replace values with op://... references and run via op run --env-file.
• op read $REF >> ~/.bashrc. Bakes the value into the dotfile. Use op run or read on demand.
• A script that does export PASS=$(op read ...) then runs many subcommands. Acceptable for ad-hoc use; for anything checked in, prefer op run --env-file=.env -- script.sh.
• Calling op read in a hot loop. Each call hits the local agent / 1Password.com. Read once into a variable, or run the inner work under op run.
• Using --no-masking in production. It's a debug flag.