agent-platform-model-registry

Operational guidance for managing machine learning models in the Agent Platform Model Registry through the gcloud command line. It covers the full lifecycle of registry entries: listing models to discover their numeric IDs, describing a model or a specific version for full metadata, uploading new models or versions, updating metadata such as display name, description, and labels, and deleting a model and all of its versions. It is explicitly scoped away from model training, deployment to endpoints, and managing models outside the Agent Platform.

Safety is the organizing principle, expressed as three confirmation tiers. Tier R read-only actions like list and describe run immediately with no confirmation. Tier M mutating but reversible actions like upload and update require an interactive Yes or No confirmation whose prompt must contain the exact literal command string with all required flags, and must never execute in the same turn as the prompt. Tier D destructive actions, meaning delete, require explicit typed confirmation requested immediately, before any pre-flight checks, and also may not run in the same turn. These same-turn restrictions are stated as hard rules.

A required Phase 0 environment setup precedes any command: authenticating with gcloud auth login and configuring Application Default Credentials with gcloud auth application-default login, setting the active project, and always specifying the region on each command rather than using global. The commands themselves are standard gcloud ai models invocations for list, describe (including the model-at-version form), upload with flags such as container image URI and artifact URI, update, and delete. A note reminds that upload is a long-running operation and that all versions must be undeployed from endpoints before a model can be deleted.