esbuild does not read tsconfig.json, so it defaults to classic JSX
transform (React.createElement) while tsconfig specifies react-jsx
(automatic). This produces 432 bare React.createElement calls in the
bundle that reference a nonexistent React global, crashing all TUI
commands (status, deploy, etc.) with "React is not defined".

Constraint: esbuild ignores tsconfig jsx settings by default
Rejected: Adding React as external | would require React in node_modules at runtime
Confidence: high
Scope-risk: narrow

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The Starter Toolkit injects memory ID via BEDROCK_AGENTCORE_MEMORY_ID,
but CDK constructs use MEMORY_{NAME}_ID pattern. After import, agent
code still references the old env var, causing memory to silently fail.

Add a yellow warning during import telling users the correct env var
name to update in their agent code.

Constraint: CDK env var naming is controlled by AgentCoreMemory.getEnvVarName()
Rejected: Auto-rewrite agent source code | too fragile across frameworks
Confidence: high
Scope-risk: narrow

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The Starter Toolkit injects memory ID via BEDROCK_AGENTCORE_MEMORY_ID,
but CDK constructs use MEMORY_{NAME}_ID pattern. After import, agent
code still references the old env var, causing memory to silently fail.

Show a git-diff-style warning during import with red/green highlighting
so users see exactly what to change in their agent code.

Constraint: CDK env var naming is controlled by AgentCoreMemory.getEnvVarName()
Rejected: Auto-rewrite agent source code | too fragile across frameworks
Confidence: high
Scope-risk: narrow

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Import is a CLI-only command that requires --source flag and doesn't
have a TUI screen. Remove it from the interactive command picker.

Confidence: high
Scope-risk: narrow

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Import fails with "The specified bucket does not exist" when the AWS
account hasn't been CDK-bootstrapped. The deploy command handles this
via useCdkPreflight, but import bypasses that since it's CLI-only.

Check bootstrap status after CDK synth and auto-bootstrap if needed,
before disposing the toolkit wrapper.

Tested on two unbootstrapped accounts (509471412906, 126432121770).

Constraint: Must bootstrap before disposing toolkitWrapper since bootstrap requires it
Rejected: Prompt user to manually run cdk bootstrap | poor UX for import flow
Confidence: high
Scope-risk: narrow

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Remove unused waitUntilChangeSetCreateComplete import
- Prefix unused assetHash with underscore
- Add eslint-disable for necessary any cast in STS credentials
- Remove unnecessary type assertion in test
- Fix prettier formatting in multi-agent test

Confidence: high
Scope-risk: narrow

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

OAuthCredentialProvider.discoveryUrl was made optional to support imported
providers that already exist in Identity service. Update tests to match:
- Schema test: expect success when discoveryUrl is omitted
- Pre-deploy identity tests: add discoveryUrl to fixtures that test
  update and error paths (without it, the code now skips the provider)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…rAllMocks

checkBootstrapNeeded and bootstrapEnvironment were created as inline
vi.fn() inside the vi.mock factory. vi.clearAllMocks() in beforeEach
wipes their mockResolvedValue, causing handleImport to fail in CI
where test execution order differs from local runs.

Hoist the mocks and configure them in setupCommonMocks like all other
mock functions in the file.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

When import fails during CDK build/synth or CloudFormation phases, the
merged config (agentcore.json) was already written to disk. On retry,
the import would detect existing agents and report false success without
completing the CloudFormation import. This snapshots the original config
before merging and restores it if any later phase fails.

Constraint: Rollback only triggers after config is written (configWritten flag)
Constraint: Snapshot must be a deep clone since projectSpec is mutated in-place
Rejected: Deferred config write until after all phases | too invasive, changes control flow for all paths
Confidence: high
Scope-risk: narrow

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

STM_ONLY memories incorrectly received a SEMANTIC strategy in the import
flow, while the CLI create flow and starter toolkit both deploy STM_ONLY
with zero strategies. This mismatch caused drift/errors during reconciliation.

Constraint: Must match CLI create flow which maps shortTerm to strategies: []
Rejected: Keep SEMANTIC for STM_ONLY | causes reconciliation drift with deployed resources
Confidence: high
Scope-risk: narrow

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>