gcp-pipeline-resource-provisioning

How to use this skill

deployment.yaml

Mandatory labels

[!IMPORTANT] Whenever you generate resource definitions in

deployment.yaml

, you MUST directly populate the

datacloud

label under

definition.labels

for every resource to track the source of creation. Determine the value based on your current IDE environment:

• For Antigravity, set

  datacloud: "antigravity"

• For VS Code, set

  datacloud: "vscode"

• For any other environment, set

  datacloud: "other"

Do not use a variable substitution for this label; hardcode the appropriate string value directly into each resource definition (e.g., replacing

__REQUIRED_LABEL__

placeholders).

Special rule for BigQuery DTS Ingestion: Whenever you generate a

bigquerydatatransfer.transferConfig

in

deployment.yaml

, you MUST also explicitly define its target destination

bigquery.dataset

in the same file and apply the

datacloud

label to it. You must do this even if the dataset already exists, to ensure the destination dataset's labels are patched and updated.

Step 1: Supported Resource Types

gcloud beta orchestration-pipelines resource-types list

deployment.yaml

Step 2: Discover Environment Parameters

[!TIP] If

deployment.yaml

already exists in the repository root, prioritize extracting

project

and

region

from the target environment configuration (e.g.,

dev

).

1. Project ID:
   bash

   gcloud config get project

2. Project Number:
   bash

   gcloud projects describe $(gcloud config get project) --format="value(projectNumber)"

3. Region:
   bash

   gcloud config get-value compute/region

4. Commit SHA:
   bash

   git rev-parse HEAD

5. Environment Name: If initialization is needed, you MUST ask the user for the environment name. If the user does not provide it, use dev as the default.

[!TIP]

Use these commands to replace placeholders like

YOUR_PROJECT_ID

with actual values. Always remove associated comments that start with TODO once replaced.

Step 3: Generate or update deployment.yaml

deployment.yaml

[!TIP]

Use the Reference Spec: The agent can use the

references/gcp_pipeline_resource_provisioning_spec.md

file as a template. It includes sample definitions for select supported resource types. Copy and adapt the required resource blocks into the

deployment.yaml

. Use

gcloud beta orchestration-pipelines resource-types list

when needed.

[!IMPORTANT]

Handling Secrets & Privacy (CRITICAL): NEVER hardcode plain-text secrets in

deployment.yaml

.

• Sensitive Data (Secrets): Sensitive information such as passwords, API keys, and other sensitive information MUST be stored in Secret Manager and declared in the

  secrets:

  block of

  deployment.yaml

  .
• Non-Sensitive Data (Variables): General configuration (e.g., dataset names, table IDs, regions) could be declared in the

  variables:

  block.
• Substitution via

  {{ VAR }}

  : Both

  variables:

  and

  secrets:

  MUST be used as

  {{ VARIABLE_NAME }}

  substitutions in resource definitions.
• No Creation: The agent MUST NOT use the framework to create new secrets. If

  gcloud

  indicates the secret does not exist, the agent MUST ask the user to create it manually and then re-verify.
• Reference Only Policy: The agent's role is strictly limited to referencing existing secrets. The agent MUST NEVER read, print, or inspect the values of secrets.
• Safe Deployment: The actual value injection happens during deployment execution. The agent only provides the reference.
• Manual Secret Management: Advise the user to manage secret payloads and versions manually.

Step 4: Validation

deployment.yaml

gcloud beta orchestration-pipelines validate --environment=<ENV_NAME>

Step 5: Deployment

gcloud beta orchestration-pipelines deploy --environment=<ENV_NAME> --local

[!NOTE] If a new transfer is being created, make sure to NOT remove the DTS transfer resource from

deployment.yaml

after it completes the run.

Definition of Done

• deployment.yaml

  exists in the repository root with actual discovered values (no placeholders) and correct resource definitions.
• The agent runs the deployment command to perform the deployment, and it executes successfully (exit code 0).