Support multiple deployment targets in the same account+region (scoped by target name)

[allandsouza]

Problem Statement

Currently, the AgentCore CLI assumes a 1:1 mapping between a deployment target and an account+region pair. This forces teams to either provision separate AWS accounts or use different regions to maintain isolated environments (dev, staging, preprod) - both of which add operational overhead and cost.

Many teams, especially during early development or in organizations with limited AWS account provisioning, need to run multiple isolated environments within a single account and region. The CDK layer already supports this (it scopes resource names by target name, e.g., QAgents_dev, QAgents_stg), but the CLI's state management and status commands do not.

Proposed Solution

Allow multiple targets in aws-targets.json to share the same account and region, using name as the disambiguator across the entire CLI lifecycle:

[
  { "name": "dev", "account": "123456789012", "region": "us-west-2" },
  { "name": "stg", "account": "123456789012", "region": "us-west-2" },
  { "name": "preprod", "account": "123456789012", "region": "us-west-2" }
]

Specifically:

1. Deploy: When persisting deployed-state.json after deploy, read CloudFormation outputs from the target-scoped stack (AgentCore-<project>-<target>) rather than stackNames[0] from the cloud assembly.

2. Status: agentcore status --target stg should resolve the correct runtime ARNs from the stg entry in deployed-state.json.

3. Invoke: agentcore invoke --target stg should invoke the correct runtime.

4. Destroy: agentcore destroy --target stg should only tear down the stg stack and remove its entry from state.

Use Case

Our team deploys AI agents to AgentCore and needs isolated environments for:

• dev - active development, frequent redeployments
• stg - integration testing with downstream services
• preprod - final validation before production (which lives in a separate account)

All three share the same sandbox AWS account and region. We don't want to:

• Create 3 AWS accounts just for pre-production environments
• Deploy to 3 different regions (introduces latency differences, model availability differences, and makes cross-service integration testing harder)
• Maintain 3 separate project directories with duplicated configuration

What Would Need to Change

The CLI currently assumes one target per account+region. To support this feature:

1. CDK synthesis: cdk/bin/cdk.ts would need to create separate stacks per target (e.g., AgentCore-Project-dev, AgentCore-Project-stg) and scope resource names by target to avoid collisions.

2. State persistence: After deploy, the CLI should read CloudFormation outputs from the target-scoped stack name rather than stackNames[0] from the cloud assembly.

3. Status/Invoke/Destroy: These commands should resolve the correct runtime ARNs using the target name as the key into deployed-state.json.

4. Schema validation: The aws-targets.json schema already enforces uniqueness on target.name — no schema changes needed.

Alternatives Considered

Alternative	Downside
Separate AWS accounts per env	Account provisioning overhead, cross-account networking complexity
Different regions per env	Model availability varies by region, latency differences, harder integration testing
Single target, manual environment switching	No isolation, risk of overwriting production-bound code
Post-deploy script to patch state	Fragile workaround, breaks on CLI updates