fix(deploy): scope TUI deploy to the picker-selected targets (#1267)

[aidandaly24]

Description

#1267

In the TUI deploy flow, when aws-targets.json has more than one target the multi-select picker lets the user choose a subset, but the choice is dropped: the deploy screen header lists every target (cosmetic), and far worse, the deploy itself runs against ALL synthesized stacks (all targets), not the selected one(s). A user who picks a single target silently deploys real infrastructure to every configured account/region. Only affects multi-target projects; single-target (the default) is correct.

Fix

Thread the picker selection through preflight AND scope the deploy. (1) Display + context (what open PR #1302 does): add selectedTargets?: AwsDeploymentTarget[] to PreflightOptions/DeployFlowOptions; DeployScreen derives it from awsConfig.selectedTargetIndices over availableTargets and passes it down; useCdkPreflight filters preflightContext.awsTargets to the selected names after validateProject(). This fixes the header (DeployScreen.tsx:279) and makes awsTargets[0] reads (persist/teardown/transaction-search; checkBootstrapNeeded preflight.ts:307) reflect the user's choice. (2) ADDITIONALLY (NOT in PR #1302, and the load-bearing part): scope the actual deploy by passing a stacks selector into the TUI cdkToolkitWrapper.deploy() at useDeployFlow.ts:784, built from the selected target names mapped through toStackName(projectName, targetName) with StackSelectionStrategy.PATTERN_MUST_MATCH, e

Related Issue

Closes#1267

Documentation PR

N/A — bug fix; no devguide change required.

Type of Change

• Bug fix
• New feature
• Breaking change
• Documentation update
• Other (please describe):

Testing

How have you tested the change? (full validation in PR comments after bug bash)

• I ran npm run test:unit and npm run test:integ
• I ran npm run typecheck
• I ran npm run lint
• If I modified src/assets/, I ran npm run test:update-snapshots and committed the updated snapshots — N/A, no src/assets/ changes

Checklist

• I have read the CONTRIBUTING document
• I have added any necessary tests that prove my fix is effective or my feature works
• I have updated the documentation accordingly
• I have added an appropriate example to the documentation to outline the feature, or no new docs are needed
• My changes generate no new warnings
• Any dependent changes have been merged and published

---

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the
terms of your choice.

[agentcore-devx-automation]

Claude Security Review: no high-confidence findings. (run)

[github-actions]

Package Tarball

aws-agentcore-0.21.0.tgz

How to install

gh release download pr-1659-tarball --repo aws/agentcore-cli --pattern "*.tgz" --dir /tmp/pr-tarball
npm install -g /tmp/pr-tarball/aws-agentcore-0.21.0.tgz

[agentcore-cli-automation]

Thanks for the fix — the deploy scoping itself is correct and the regression test is well-targeted.

I have two concerns about completeness that I think should be resolved before merging.

1. Post-deploy bookkeeping still operates on the wrong target (see inline on useDeployFlow.ts). The PR's stated dependency on #1302 is load-bearing: the description acknowledges that filtering context.awsTargets (so that awsTargets[0] reads in persist/teardown/transaction-search reflect the user's choice) is what #1302 does, but #1302 is still open and not merged. As of this PR alone, the deploy goes to the right target while the post-deploy bookkeeping does not — arguably a worse state than today's bug for affected users (silent state corruption rather than over-deployment).

Options:

1. Land #1302 first (or fold its awsTargets filtering into this PR) so the picker selection flows through preflight before this deploy-scope change goes in.
2. Combine the two PRs into one end-to-end fix for #1267.

2. Diff path is still unscoped. The pre-deploy cdkToolkitWrapper.diff() (around line 747) and the standalone diff-mode call (around line 1010) are still called without stacks. When a multi-target project's user picks one target the diff still runs against ALL synthesized stacks — same class of bug as the deploy issue, just without provisioning real infra. The PR title says "deploy" so this may be deliberate, but worth either fixing here (pass { stacks: deployStacks } to both diff() calls) or explicitly calling out as out-of-scope and tracked separately.

See inline comments for specifics.

[agentcore-cli-automation]

The deploy is now correctly scoped to the picked target's stack, but downstream code in this same effect still assumes a single-target world via awsTargets[0] / stackNames[0]:

• persistDeployedState() (line 205-206) reads stackNames[0] and ctx.awsTargets[0] to poll outputs and write deployed-state. For a [A, B] project where the user picks B, this polls A's stack in A's region (which was never deployed) and writes state under target A instead of B.
• The teardown branch (around line 399) uses context.awsTargets[0]?.name — runs teardown for the wrong target.
• The transaction-search setup (around lines 417-418) uses context.awsTargets[0]?.region / .account — wires CloudWatch in the wrong account/region.

Without #1302 (which filters preflightContext.awsTargets to the selected names), these post-deploy paths silently operate on the wrong target. The deploy bug is fixed but a new state-corruption bug is introduced for multi-target users who pick a subset.

Options:

1. Merge fix(deploy): respect TUI target selection in deploy flow #1302 first so context.awsTargets is already filtered to the selection by the time it reaches useDeployFlow.
2. Fold fix(deploy): respect TUI target selection in deploy flow #1302's filtering into this PR.
3. At minimum, in this hook resolve the active target(s) from selectedTargets (falling back to context.awsTargets) and use that for persistDeployedState, teardown, and transaction-search instead of context.awsTargets[0] / stackNames[0].

[aidandaly24]

Addressed in a37a9b5 by taking your Option 3 (resolve the active target in-hook, no dependency on the unmerged #1302).

Added a memoized activeTarget = selectedTargets?.[0] ?? context?.awsTargets[0] and routed the post-deploy bookkeeping through it:

• persistDeployedState() now derives currentStackName via toStackName(projectName, activeTarget.name) (instead of stackNames[0]) and polls activeTarget.region — so a non-first pick polls the stack that was actually deployed and writes deployed-state under the right target name.
• The teardown branch now resolves targetName/region from activeTarget instead of awsTargets[0].
• Transaction-search setup now reads activeTarget.region/.account.

Fall-throughs are preserved: the plan path (no project name / no picker) and the no-selection case still fall back to stackNames[0] / context.awsTargets[0], matching prior behavior. The TUI persists a single target; when several are picked we bookkeep the first selected one (the same target whose outputs we poll). New regression test asserts the persist path polls B's region + B's stack when B is the non-first pick.

[agentcore-cli-automation]

Because getStackOutputs is mocked to reject, the test never exercises the post-deploy persistDeployedState path — which is exactly the path that's broken when the picked target isn't awsTargets[0] (see the inline comment on useDeployFlow.ts around line 810). The test asserts deploy() is called with the right scope but doesn't catch the end-to-end regression for multi-target picks. Adding coverage that the persist path resolves outputs from the selected target's region/stack (not awsTargets[0] / stackNames[0]) would make the gap in the previous comment self-enforcing.

[aidandaly24]

Good catch — closed the coverage gap in a37a9b5. The getStackOutputs mock is now a hoisted spy (so the test can keep a handle while still short-circuiting the long poll), and a new test persist polls the SELECTED target stack/region, not awsTargets[0]/stackNames[0] renders with selectedTargets=[B] on an [A, B] project and asserts the persist path calls getStackOutputs(B.region, toStackName(project, B.name)) — and never A's region/stack. Before the hook fix this asserted on A's values, so the regression is now self-enforcing. Also added scopes the pre-deploy diff to the selected target stack to cover the diff-scoping fix.

[agentcore-devx-automation]

Claude Security Review: no high-confidence findings. (run)

[agentcore-cli-automation]

Reviewed this PR end-to-end. The existing review from agentcore-cli-automation already flags every serious issue I would have raised:

1. Post-deploy bookkeeping (useDeployFlow.ts ~L290-291, L831, L838, L870-871) — persistDeployedState, the teardown branch, and the transaction-search setup all still read context.awsTargets[0] / stackNames[0]. When a multi-target user picks a non-first target, the deploy now goes to the right stack but the post-deploy paths write deployed-state under the wrong target, polish outputs from a stack that wasn't deployed, and wire CloudWatch in the wrong account/region — a state-corruption regression. Already covered in the inline at L810 with concrete fix options.
2. Diff paths unscoped (useDeployFlow.ts L747 pre-deploy diff, L1010 standalone diff-mode) — same class of bug as the original (multi-target user picks one → diff runs against all). Already raised in the review summary.
3. Test gap on the persist path — getStackOutputs is mocked to reject so the regression test never exercises the persist code that issue build(deps): bump diff and @aws-cdk/cloudformation-diff #1 lives in. Already raised inline on the test file at L60.

No additional issues to flag — deferring to the existing review and not duplicating its comments. Approving the deploy-scoping change in isolation is correct, but I agree with the existing reviewer's CHANGES_REQUESTED state: the post-deploy bookkeeping needs to be fixed in this PR (or #1302 landed first) before merging, otherwise this trades over-deployment for silent state corruption on the exact same multi-target/non-first-pick path.

[aidandaly24]

Thanks for the thorough review. Both concerns are addressed in a37a9b5 — kept the change surgical and avoided depending on the unmerged #1302.

1. Post-deploy bookkeeping now tracks the deployed target. Added a memoized activeTarget (selectedTargets?.[0] ?? context?.awsTargets[0]) and routed persistDeployedState (stack name now via toStackName(project, activeTarget.name), region via activeTarget.region), the teardown branch, and transaction-search through it instead of awsTargets[0] / stackNames[0]. The plan path and no-selection case still fall back to the prior values, so single-target and pre-synthesized flows are unchanged. This removes the state-corruption window for multi-target users who pick a non-first target.

2. Diff path is now scoped. All three cdkToolkitWrapper.diff() call sites — the pre-deploy diff, the standalone diff-mode run, and the on-demand requestDiff — now pass { stacks: deployStacks }, so the diff mirrors exactly what will deploy. deployStacks is undefined (→ full assembly) when nothing is picked / on the plan path, preserving existing behavior.

Two new regression tests lock both fixes in: persist polls the selected target's region+stack (not awsTargets[0]), and the pre-deploy diff is scoped to the selected target's stack. Full local suite: prettier --check, eslint src/, typecheck, build, and the 7-test deploy unit file all green.

[agentcore-devx-automation]

Claude Security Review: no high-confidence findings. (run)

[agentcore-cli-automation]

Re-reviewed after the a37a9b5 fixup. Both prior concerns are properly resolved:

1. Post-deploy bookkeeping — activeTarget = selectedTargets?.[0] ?? context?.awsTargets[0] is now threaded through persistDeployedState (stack name via toStackName(project, target.name), region via target.region), the teardown branch (cleanupPaymentCredentialProviders + performStackTeardown(targetName)), and the transaction-search setup. Fallbacks preserve single-target and pre-synthesized plan paths. Verified by grepping all remaining awsTargets[0] / stackNames[0] references — they're all in comments or fallback branches now.
2. Diff scoping — all three cdkToolkitWrapper.diff() call sites (pre-deploy at L768, standalone diff-mode at L1034, on-demand at L284) now pass { stacks: deployStacks }. deployStacks correctly resolves to undefined when nothing is picked / on the plan path, so the assembly-wide fallback is preserved.
3. Test coverage — the new persist polls the SELECTED target stack/region test (rendering with selectedTargets=[B] on an [A, B] project and asserting getStackOutputs is called with B's region + toStackName(project, B.name), never A's) closes the gap on the previously-untested persist path. The hoisted getStackOutputsSpy pattern is clean. The new scopes the pre-deploy diff to the selected target stack test exercises the isFirstDeploy=false branch.

Mocking surface stays at real I/O boundaries (preflight hook, telemetry wrapper, CloudFormation SDK, operations/deploy filesystem reads) — not excessive.

No new merge-blocking issues from me.

One latent follow-up (not a blocker, pre-existing, not introduced by this PR): when the user picks multiple targets in the multi-select (e.g. 2 of 3), the deploy correctly scopes to those 2 via PATTERN_MUST_MATCH, but activeTarget = selectedTargets[0] means persistDeployedState / teardown / transaction-search only run for the first picked target. The second deployed stack's outputs never make it into deployed-state.json. The code comment on the activeTarget memo ("if the user picked several, we bookkeep the first selected one") explicitly acknowledges this. This was true before the PR too — back then the deploy hit all configured targets but persist wrote awsTargets[0] only — so it isn't a regression. Worth tracking as a separate issue since the multi-pick path is now more reachable in practice (users can deliberately pick 2 targets and have both actually deploy).

LGTM to merge.

[agentcore-cli-automation]

Approving. The two prior CHANGES_REQUESTED issues (post-deploy bookkeeping operating on awsTargets[0] and the unscoped diff paths) are properly addressed in a37a9b5 via activeTarget + { stacks: deployStacks } on all three diff sites, and the new persist regression test closes the coverage gap on the post-deploy path. Details in the comment thread. One latent multi-pick persist gap noted as a follow-up — pre-existing, not a regression, not a blocker.

[agentcore-devx-automation]

Claude Security Review: no high-confidence findings. (run)

[agentcore-cli-automation]

Reviewed the PR end-to-end (all commits through d899da3d) after reading the prior review threads. Every serious concern I would have raised has already been flagged and addressed:

• Post-deploy bookkeeping on awsTargets[0] / stackNames[0] — raised inline at useDeployFlow.ts L810 in the first review, fixed in a37a9b5 via the activeTarget memo threaded through persistDeployedState, the teardown branch (cleanupPaymentCredentialProviders + performStackTeardown), and transaction-search setup.
• Unscoped diff paths — raised in the first review summary, fixed in a37a9b5 by passing { stacks: deployStacks } to all three cdkToolkitWrapper.diff() call sites (pre-deploy L776, standalone diff-mode L1057, on-demand requestDiff).
• Persist path not exercised by the regression test — raised inline on the test file, closed in a37a9b5 via the hoisted getStackOutputsSpy + persist polls the SELECTED target stack/region… test.
• Success summary listing every synthesized stack instead of the scoped set — raised at L927, fixed in d899da3d (deployStacks?.patterns ?? stackNames) with a regression test.
• Empty picker selection silently falling back to ALL_STACKS — raised at L165, fixed in d899da3d by distinguishing undefined (unscoped) from [] (PATTERN_MUST_MATCH with no patterns → toolkit-lib throws NoStacksMatched) with a regression test.
• Multi-target pick leaves non-first targets without persist/transaction-search — raised at L183, addressed in d899da3d with a user-visible warning and tracked for full per-target bookkeeping under a follow-up issue.

Mocking surface in the new test file stays at real I/O boundaries (CDK toolkit wrapper, preflight hook, CloudFormation SDK, telemetry wrapper, filesystem-reading operations). Not excessive given useDeployFlow orchestrates half a dozen subsystems that would each need substantial fixtures to run for real; mocking at the hook boundary is the pragmatic choice here.

No new blockers or additional issues to flag from me. LGTM to merge.