aicore-lifecycle-management

Rules

1. Run all scripts from the skill root directory with uv run scripts/<name>.py.
2. Never print or log credential values — the SDK reads them from the environment.
3. Scripts that delete or mutate state require --confirm. Always pass it explicitly; never bypass it.
4. When the user does not specify a resource group, omit --resource-group so the SDK uses its configured default.
5. For LLM deployments: the scenario ID, executable ID, and model parameter names come from genai-hub-foundation-models — invoke it first if those values are unknown.
6. Deployment configuration naming convention: use {executable_id}--{model_name} for LLM deployments (e.g. azure-openai--gpt-4o-mini). Before creating a configuration, check whether one with that name already exists via get_configurations.py --scenario-id <id> and reuse it to avoid duplicates.

List / Inspect Scenarios

# List all scenarios
uv run scripts/get_scenarios.py

# Inspect a specific scenario and its versions
uv run scripts/get_scenarios.py --scenario-id <id>
uv run scripts/get_scenarios.py --scenario-id <id> --response-format detailed

# JSON output for piping
uv run scripts/get_scenarios.py --json

List Executables

# List executables for a scenario
uv run scripts/get_executables.py --scenario-id <id>

# Inspect a specific executable (includes parameters and input artifacts)
uv run scripts/get_executables.py --scenario-id <id> --executable-id <id>
uv run scripts/get_executables.py --scenario-id <id> --executable-id <id> --response-format detailed

# JSON output
uv run scripts/get_executables.py --scenario-id <id> --json

Parameters are only returned when inspecting a single executable (--executable-id). The list view returns id, name, and description only.

Configurations

# List configurations
uv run scripts/get_configurations.py
uv run scripts/get_configurations.py --scenario-id <id>
uv run scripts/get_configurations.py --response-format detailed
uv run scripts/get_configurations.py --json

# Inspect a specific configuration
uv run scripts/get_configurations.py --configuration-id <id>

# Create a configuration (binds parameters to an executable)
uv run scripts/create_configuration.py \
  --name my-config \
  --scenario-id <id> \
  --executable-id <id> \
  --param modelName=gpt-4o \
  --param temperature=0.7

If a deployment fails or stays PENDING after several minutes, verify parameter names and values against the executable's declared parameters:

uv run scripts/get_executables.py --scenario-id <id> --executable-id <id> --response-format detailed

Deployments

Deployments serve inference endpoints. The standard flow is: create (or reuse) a configuration, then create a deployment from it.

Configuration naming convention

Always check whether a matching configuration already exists before creating one, to avoid duplicates:

# Check existing configurations before creating
uv run scripts/get_configurations.py --scenario-id <scenario-id> --json | jq '.[] | select(.name == "<expected-name>")'

Naming convention:

• LLM: {executable_id}--{model_name} — e.g. azure-openai--gpt-4o-mini
• Orchestration: orchestration
• Custom: choose a descriptive name

LLM deployment (from genai-hub-foundation-models handoff)

# 1. Create configuration (skip if one with this name already exists):
uv run scripts/create_configuration.py \
  --name azure-openai--gpt-4o-mini \
  --scenario-id foundation-models \
  --executable-id azure-openai \
  --param modelName=gpt-4o-mini \
  --param modelVersion=latest

# 2. Deploy:
uv run scripts/create_deployment.py --configuration-id <config-id> --wait

Orchestration deployment

# 1. Create configuration (skip if "orchestration" config already exists):
uv run scripts/create_configuration.py \
  --name orchestration \
  --scenario-id orchestration \
  --executable-id orchestration

# 2. Deploy:
uv run scripts/create_deployment.py --configuration-id <config-id> --wait --timeout 600

List and inspect deployments

# List all deployments
uv run scripts/get_deployments.py

# Filter by status or scenario or executable
uv run scripts/get_deployments.py --status RUNNING
uv run scripts/get_deployments.py --scenario orchestration
uv run scripts/get_deployments.py --executable-id azure-openai --status RUNNING
uv run scripts/get_deployments.py --executable-id azure-openai --status RUNNING --json

# Inspect one deployment
uv run scripts/get_deployments.py --deployment-id <id>
uv run scripts/get_deployments.py --deployment-id <id> --response-format detailed
uv run scripts/get_deployments.py --deployment-id <id> --json

Stop, patch, delete, logs

# Stop a deployment
uv run scripts/stop_deployment.py --deployment-id <id> --confirm
uv run scripts/stop_deployment.py --deployment-id <id> --confirm --wait

# Patch (swap configuration on a running deployment)
uv run scripts/patch_deployment.py --deployment-id <id> --configuration-id <new-config-id>

# Delete (must be STOPPED or DEAD first)
uv run scripts/delete_deployment.py --deployment-id <id> --confirm
# Stop and delete in one step:
uv run scripts/delete_deployment.py --deployment-id <id> --confirm --auto-stop

# Deployment logs
uv run scripts/get_deployment_logs.py --deployment-id <id>
uv run scripts/get_deployment_logs.py --deployment-id <id> --tail 50
uv run scripts/get_deployment_logs.py --deployment-id <id> --order asc
uv run scripts/get_deployment_logs.py --deployment-id <id> --json

# Remove duplicate LLM deployments (only affects LLM deployments with a model name)
uv run scripts/dedup_deployments.py --dry-run
uv run scripts/dedup_deployments.py

Handoffs

• Need a Bearer token or credential setup? → invoke aicore-admin-resources
• Need executable ID, model name, or model capabilities? → invoke genai-hub-foundation-models
• Want to list available models or providers? → invoke genai-hub-foundation-models