Devin

Sessions authenticate with a subject you compose from the claims you choose.

What proves the identity

The token has to come from https://app.devin.ai, and its sub claim has to match the workload you named. penv compares that claim byte for byte, with no wildcards.

Every token also has to name your workspace as its audience. Ask for this audience and no other. A token requested for two audiences fails every exchange.

What the console asks for

A field marked advanced carries a default and stays behind the disclosure until you open it.

The console also asks which project and environment this identity reaches and which role it gets, with the credential lifetime behind Advanced.

What the trust holds

Built from the example answers above:

penv names the identity for you, so the form never asks for one.

The snippet the console prints

# In a Devin session, as one command:
# --subject-keys must name the same keys as the Subject you pinned in penv, or the token
# arrives with a sub that matches nothing.
T=$(devin-oidc token --audience YOUR_WORKSPACE_ID --subject-keys "org_id")
C=$(curl -sS "https://penv.cloud/api/v1/auth/oidc" \
  -H 'content-type: application/json' \
  -d "{\"token\":\"$T\"}" | jq -r .credential)
PENV_TOKEN="$C" penv pull

The snippet finishes with penv pull, which writes a file. penv run is the path the docs teach for a process, and penv pull is for a host that needs a file on disk. Connect your CI has the wrapped step.

Related

• Machine identity
• Connect a Platform in the console
• Devin documentation, where the value people get wrong is always the subject