Manages and orchestrates prompts in Agent Platform. Use when you need to create, list, retrieve, version, or delete managed prompts in Agent Platform. Don't use for model training, model deployment to endpoints, or managing non-Agent Platform prompts.
A management and orchestration skill for prompts in Google Agent Platform (Vertex AI), covering creating, listing, retrieving, versioning, and deleting managed prompts. It is scoped specifically to Agent Platform prompts and explicitly excludes model training, model deployment to endpoints, and managing non-Agent Platform prompts. Rather than searching the file system for scripts, the skill provides Python snippets for the user to run, using the high-level Prompt class in the Vertex AI SDK preview module (vertexai.preview.prompts).
Safety is the defining feature, organized into three confirmation tiers. Tier R covers read-only operations (list and get) and executes immediately with no confirmation. Tier M covers mutating but reversible creation and requires interactive Yes/No confirmation that spells out the proposed display name, target model, and template text, with a strict same-turn restriction that forbids running the creation code in the same turn as the confirmation prompt. Tier D covers destructive, irreversible deletion and requires explicit typed confirmation such as I confirm, asked before any pre-flight checks and never executed in the same turn. The doc repeatedly stresses that asking for confirmation and providing deletion code in a single parallel turn is a severe safety violation.
A Phase 0 environment setup section instructs the user to authenticate with Google Cloud and configure Application Default Credentials (gcloud auth login and gcloud auth application-default login), create and activate a dedicated virtual environment, and install the google-cloud-aiplatform and google-genai SDKs, activating the venv before every snippet. Uppercase placeholders such as PROJECT_ID, LOCATION_ID, and PROMPT_ID must be replaced with real values before scripts are provided. The reference snippets initialize Vertex AI, list prompts, retrieve a prompt (including a specific version_id) and assemble it with variables via assemble_contents, and delete by numeric prompt ID. Best practices cover idempotency, double-curly-brace placeholder syntax, recording version IDs for production prompts, specifying a target model such as gemini-2.5-pro, and using the correct metadata schema when working through the Dataset API. Detailed create instructions live in references/create.md.
It creates, lists, retrieves, versions, and deletes managed prompts in Agent Platform. It is not for model training, model deployment to endpoints, or managing non-Agent Platform prompts.
Deletion is a Tier D destructive operation requiring explicit typed confirmation such as I confirm, asked before any pre-flight checks. The deletion code must never be provided or executed in the same turn as the confirmation request.
Authenticate with gcloud auth login and gcloud auth application-default login, create and activate a dedicated Python virtual environment, and install google-cloud-aiplatform and google-genai. The venv must be active each time a snippet is run.
The high-level Prompt class in the Vertex AI SDK preview module, imported as vertexai.preview.prompts. After vertexai.init() you can call prompts.list(), prompts.get(), and prompts.delete().
Yes. prompts.get() accepts a version_id argument, for example version_id='2', and a retrieved prompt can be assembled with variables using assemble_contents() where keyword arguments match the template variable names.
Quick Setup:
.claude/skills/Repository
google/skills