Short answer: don’t put secrets in

work_pool.job_variables.env

. In Prefect 3.x those values are resolved and stored on the deployment, so the UI will show them. Instead, load the secret at runtime inside your flow and, if you truly need an environment variable, set it in-process. Why you’re seeing plaintext -

{{ prefect.blocks.secret.password }}

in

job_variables.env

is resolved during

prefect deploy

and the resulting value is stored on the deployment. The Configuration tab then displays that stored value. Recommended pattern (keeps secrets out of the UI) - Load the secret inside your flow/task from a Secret block and either: - Pass it directly to the client/library you’re using; or - Set it as an environment variable only within the running process. Example

python
from prefect import flow, task
from prefect.blocks.system import Secret
import os
import subprocess

@task
def do_something_needing_pwd():
    password = Secret.load("password").get()
    # Option A: pass directly to a library/client
    # client.connect(password=password)

    # Option B: set env just-in-time for a subprocess
    env = {**os.environ, "PASSWORD": password}
    subprocess.run(["your_cmd"], env=env, check=True)

@flow
def my_flow():
    do_something_needing_pwd()

if __name__ == "__main__":
    my_flow()

This keeps the secret: - Encrypted at rest in the Secret block - Out of deployment configuration and the UI - Only present in-memory on the worker at run time If you absolutely need an env var before any library code runs - Put the logic that loads the Secret block at the start of your flow (as above) and set

os.environ["PASSWORD"]

before initializing your client. - Alternatively, mount a file (via

job_variables.volumes

) that contains the secret and read it in your code; the file path will be visible in the UI, but not the secret contents. - Avoid trying to inherit arbitrary host env into the container—Docker workers don’t automatically pass through host env (only Prefect settings like

PREFECT_API_URL

). Cleanup/rotation - Since the secret was stored in the deployment config, rotate that password/credential, remove the env entry from your deployment spec, and re-deploy. If you share any constraints (e.g., a third-party tool must read

PASSWORD

from env before your Python code runs), I can suggest a concrete wrapper pattern for your image or flow to meet that requirement without exposing the value.