Vertex AI Agent Engine

Format id: vertex-agent-engine. Reads saved responses of the Agent Engine API (reasoningEngines.get or reasoningEngines.list, JSON) at inventory tier.

Marshal never calls the API. A job the organization runs where it already has access saves the response, for example:

$ gcloud ai reasoning-engines list --region=us-central1 --format=json > engines.json

The job then commits the file, or pushes it with marshal push engines.json --event observed. See Network posture.

Ledger field	Source	Provenance
agent id	vertex-agent-engine/<display name slug>
display_name, description	displayName, description	declared
permissions.identity	identityType: AGENT_IDENTITY → the agent's own identity; otherwise the service account it runs as	declared / derived
permissions.principal	serviceAccount; without one, the project's Reasoning Engine Service Agent	declared / derived
model, instructions, tools	Unknown: set in the agent's code	unknown
extensions.vertex-agent-engine	Framework, class methods, environment variable names, secret references (secret name only), deploy source (Developer Connect repository, entrypoint, container image), resource limits, labels	declared
bindings	Resource name, package locations, create and update times

Environment variable values are never copied. To fill in the model, instructions and tools, read the agent's source with the Python reader; the deploy source in the extension says where it lives.