fix: create global entrypoint for tui by Hweinstock · Pull Request #1365 · aws/agentcore-cli

Hweinstock · 2026-05-22T02:50:48Z

Description

Previously, commands like agentcore add, deploy, etc. rendered TUI screens inline via direct render() calls. This made it difficult to distinguish CLI vs TUI code paths, caused telemetry to be mislabeled as mode="cli" for interactive flows, and meant bare agentcore never initialized or shut down the telemetry client.

This PR creates a unified renderTUI() entrypoint that owns the telemetry lifecycle and replaces all inline render() calls. As part of this:

Extracted rendering logic into src/cli/tui/render.ts and shared notices into src/cli/notices.ts to break circular imports
Migrated add, deploy, create, remove, invoke to use renderTUI()
Made TelemetryClientAccessor.init() handle re-initialization safely
Preserved existing behavior (diffMode, isInteractive, actionOnBack)

This refactor fixes#895.

Type of Change

Bug fix
New feature

Testing

Typecheck, lint, unit tests pass
Manual E2E testing of affected commands

basic-flows.mov

Manually tested flows:

back on inline tui exists instead of going to help.
successful flow on tui exists instead of going back to help.
root level agentcore works as expected.
errors still propagate correctly in inline TUI.
telemetry is now emitted for both inline TUI and full screen.

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 · 2026-05-22T02:51:24Z

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

github-actions · 2026-05-22T02:51:40Z

Package Tarball

aws-agentcore-0.15.0.tgz

How to install

gh release download pr-1365-tarball --repo aws/agentcore-cli --pattern "*.tgz" --dir /tmp/pr-tarball
npm install -g /tmp/pr-tarball/aws-agentcore-0.15.0.tgz

agentcore-cli-automation

Thanks for unifying the TUI entrypoint — the RenderTUIOptions shape and InitialRoute typing are nice. I have a few concerns that I think need to be addressed before this merges.

Summary of issues

Telemetry regression for invoke TUI: The PR removes the withCommandRunTelemetry('invoke', …) wrapper that previously wrapped the TUI invoke flow. No replacement was added, so cli.command_run (with has_stream/has_session_id/auth_type/agent_protocol/duration/exit_reason) is no longer emitted for agentcore invoke interactive mode. See inline comment.
Behavior change: TUI flows no longer auto-exit on success when launched from CLI commands. Previously agentcore add, agentcore remove, etc. rendered their flows with isInteractive={false}, which makes screens like AddFlow, RemoveAllScreen, RemoveFlow, AddSuccessScreen, AddMemoryFlow, AddEvaluatorFlow, AddOnlineEvalFlow, AddIdentityFlow auto-exit after success (see e.g. AddFlow.tsx:206, RemoveAllScreen.tsx:30). Going through App now means they always receive isInteractive={true}, so the user has to press escape to leave after every successful run. See inline comment.
Double TelemetryClientAccessor.init() for command-routed paths. When agentcore add (or any of the migrated commands) runs, main() calls TelemetryClientAccessor.init(args[0], 'cli') (cli.ts:271) and then the action handler calls renderTUI() which calls TelemetryClientAccessor.init(initialRoute.name, 'tui') (cli.ts:127). The first client (mode=cli) is overwritten in the singleton and never shutdown()'d. See inline comment.
Circular import.src/cli/cli.ts imports registerInvoke/registerAdd/etc. from ./commands/*, and those command files now import renderTUI from '../../cli'. It works today because renderTUI is only referenced inside async action handlers, but it's fragile. See inline comment.

Items 1 and 3 are the most important — 1 is a clear telemetry regression and 3 partially undoes the telemetry-mode fix this PR is trying to land.

agentcore-devx-automation · 2026-05-22T03:25:32Z

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

agentcore-cli-automation

Thanks for the cleanup — the unified renderTUI() entrypoint is a nice improvement and the recent commits address most of the earlier review feedback (double init, isInteractive threading, render-tui module extraction).

Flagging two new issues below; both involve information being silently dropped on the way through renderTUI. The other pre-existing review comments still appear open but their underlying concerns look addressed in the latest commits — happy to leave those for the original reviewer to confirm.

agentcore-devx-automation · 2026-05-22T03:33:11Z

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

avi-alpert · 2026-05-26T17:32:38Z

+ const project = await configIO.readProjectSpec().catch(() => undefined);
+ const firstProtocol = project?.runtimes?.[0]?.protocol ?? 'unknown';
+
+ const result = await withCommandRunTelemetry(


When i tested this PR locally, i noticed that the value in the emitted telemetry was always 1 (likely how long it took to load the config).
Before this change, the value was the actual duration of my entire invoke session.

Discussed in person, this is to be consistent with dev, and to avoid tying the duration to arbitrary user sessions.

avi-alpert · 2026-05-26T17:35:32Z

+ process.stdout.write(SHOW_CURSOR);
+ }
+
+ await TelemetryClientAccessor.shutdown();


wouldnt shutting down TelemetryClientAccessor here make it so if telemetry is emitted during a post-exit action (i.e. launchBrowserDev) then it would stop working?

Yes, good call, this is important for when we add telemetry for the browser actions. I've swapped it to flush before the blocking process, then the shutdown should be handled by the global one in the finally block, if it reaches there. Shutdown is "best-effort" so its fine if it doesn't.

agentcore-devx-automation · 2026-05-26T17:37:39Z

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

avi-alpert · 2026-05-26T17:38:25Z

- }
+ enterAltScreen: false,
+ actionOnBack: 'exit',
+ isInteractive: false,


why was isInteractive switched to false here?

good question. I made this change thinking that isInteractive=true meant exit after first success (as it does for other commands), and I didn't want the tui to exit early. However, turns out its just dead code:
agentcore-cli/src/cli/tui/screens/invoke/InvokeScreen.tsx
Line 134 in 69b4ce6
isInteractive: _isInteractive,
. I think we can remove it as a followup.

agentcore-devx-automation · 2026-05-26T18:23:12Z

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

agentcore-devx-automation · 2026-05-26T20:00:15Z

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

agentcore-devx-automation · 2026-05-27T01:58:49Z

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

# Conflicts: # src/cli/cli.ts # src/cli/commands/invoke/command.tsx # src/cli/tui/screens/invoke/useInvokeFlow.ts

agentcore-devx-automation · 2026-05-27T02:10:18Z

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

Hweinstock · 2026-05-22T02:50:48Z

Description

Previously, commands like agentcore add, deploy, etc. rendered TUI screens inline via direct render() calls. This made it difficult to distinguish CLI vs TUI code paths, caused telemetry to be mislabeled as mode="cli" for interactive flows, and meant bare agentcore never initialized or shut down the telemetry client.

This PR creates a unified renderTUI() entrypoint that owns the telemetry lifecycle and replaces all inline render() calls. As part of this:

Extracted rendering logic into src/cli/tui/render.ts and shared notices into src/cli/notices.ts to break circular imports
Migrated add, deploy, create, remove, invoke to use renderTUI()
Made TelemetryClientAccessor.init() handle re-initialization safely
Preserved existing behavior (diffMode, isInteractive, actionOnBack)

This refactor fixes#895.

Type of Change

Bug fix
New feature

Testing

Typecheck, lint, unit tests pass
Manual E2E testing of affected commands

basic-flows.mov

Manually tested flows:

back on inline tui exists instead of going to help.
successful flow on tui exists instead of going back to help.
root level agentcore works as expected.
errors still propagate correctly in inline TUI.
telemetry is now emitted for both inline TUI and full screen.

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 · 2026-05-22T02:51:24Z

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

github-actions · 2026-05-22T02:51:40Z

Package Tarball

aws-agentcore-0.15.0.tgz

How to install

gh release download pr-1365-tarball --repo aws/agentcore-cli --pattern "*.tgz" --dir /tmp/pr-tarball
npm install -g /tmp/pr-tarball/aws-agentcore-0.15.0.tgz

agentcore-cli-automation

Thanks for unifying the TUI entrypoint — the RenderTUIOptions shape and InitialRoute typing are nice. I have a few concerns that I think need to be addressed before this merges.

Summary of issues

Telemetry regression for invoke TUI: The PR removes the withCommandRunTelemetry('invoke', …) wrapper that previously wrapped the TUI invoke flow. No replacement was added, so cli.command_run (with has_stream/has_session_id/auth_type/agent_protocol/duration/exit_reason) is no longer emitted for agentcore invoke interactive mode. See inline comment.
Behavior change: TUI flows no longer auto-exit on success when launched from CLI commands. Previously agentcore add, agentcore remove, etc. rendered their flows with isInteractive={false}, which makes screens like AddFlow, RemoveAllScreen, RemoveFlow, AddSuccessScreen, AddMemoryFlow, AddEvaluatorFlow, AddOnlineEvalFlow, AddIdentityFlow auto-exit after success (see e.g. AddFlow.tsx:206, RemoveAllScreen.tsx:30). Going through App now means they always receive isInteractive={true}, so the user has to press escape to leave after every successful run. See inline comment.
Double TelemetryClientAccessor.init() for command-routed paths. When agentcore add (or any of the migrated commands) runs, main() calls TelemetryClientAccessor.init(args[0], 'cli') (cli.ts:271) and then the action handler calls renderTUI() which calls TelemetryClientAccessor.init(initialRoute.name, 'tui') (cli.ts:127). The first client (mode=cli) is overwritten in the singleton and never shutdown()'d. See inline comment.
Circular import.src/cli/cli.ts imports registerInvoke/registerAdd/etc. from ./commands/*, and those command files now import renderTUI from '../../cli'. It works today because renderTUI is only referenced inside async action handlers, but it's fragile. See inline comment.

Items 1 and 3 are the most important — 1 is a clear telemetry regression and 3 partially undoes the telemetry-mode fix this PR is trying to land.

agentcore-devx-automation · 2026-05-22T03:25:32Z

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

agentcore-cli-automation

Thanks for the cleanup — the unified renderTUI() entrypoint is a nice improvement and the recent commits address most of the earlier review feedback (double init, isInteractive threading, render-tui module extraction).

Flagging two new issues below; both involve information being silently dropped on the way through renderTUI. The other pre-existing review comments still appear open but their underlying concerns look addressed in the latest commits — happy to leave those for the original reviewer to confirm.

agentcore-devx-automation · 2026-05-22T03:33:11Z

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

avi-alpert · 2026-05-26T17:32:38Z

+ const project = await configIO.readProjectSpec().catch(() => undefined);
+ const firstProtocol = project?.runtimes?.[0]?.protocol ?? 'unknown';
+
+ const result = await withCommandRunTelemetry(


When i tested this PR locally, i noticed that the value in the emitted telemetry was always 1 (likely how long it took to load the config).
Before this change, the value was the actual duration of my entire invoke session.

Discussed in person, this is to be consistent with dev, and to avoid tying the duration to arbitrary user sessions.

avi-alpert · 2026-05-26T17:35:32Z

+ process.stdout.write(SHOW_CURSOR);
+ }
+
+ await TelemetryClientAccessor.shutdown();


wouldnt shutting down TelemetryClientAccessor here make it so if telemetry is emitted during a post-exit action (i.e. launchBrowserDev) then it would stop working?

Yes, good call, this is important for when we add telemetry for the browser actions. I've swapped it to flush before the blocking process, then the shutdown should be handled by the global one in the finally block, if it reaches there. Shutdown is "best-effort" so its fine if it doesn't.

agentcore-devx-automation · 2026-05-26T17:37:39Z

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

avi-alpert · 2026-05-26T17:38:25Z

- }
+ enterAltScreen: false,
+ actionOnBack: 'exit',
+ isInteractive: false,


why was isInteractive switched to false here?

good question. I made this change thinking that isInteractive=true meant exit after first success (as it does for other commands), and I didn't want the tui to exit early. However, turns out its just dead code:
agentcore-cli/src/cli/tui/screens/invoke/InvokeScreen.tsx
Line 134 in 69b4ce6
isInteractive: _isInteractive,
. I think we can remove it as a followup.

agentcore-devx-automation · 2026-05-26T18:23:12Z

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

agentcore-devx-automation · 2026-05-26T20:00:15Z

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

agentcore-devx-automation · 2026-05-27T01:58:49Z

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

# Conflicts: # src/cli/cli.ts # src/cli/commands/invoke/command.tsx # src/cli/tui/screens/invoke/useInvokeFlow.ts

agentcore-devx-automation · 2026-05-27T02:10:18Z

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

Hweinstock · 2026-05-22T02:50:48Z

Description

Previously, commands like agentcore add, deploy, etc. rendered TUI screens inline via direct render() calls. This made it difficult to distinguish CLI vs TUI code paths, caused telemetry to be mislabeled as mode="cli" for interactive flows, and meant bare agentcore never initialized or shut down the telemetry client.

This PR creates a unified renderTUI() entrypoint that owns the telemetry lifecycle and replaces all inline render() calls. As part of this:

Extracted rendering logic into src/cli/tui/render.ts and shared notices into src/cli/notices.ts to break circular imports
Migrated add, deploy, create, remove, invoke to use renderTUI()
Made TelemetryClientAccessor.init() handle re-initialization safely
Preserved existing behavior (diffMode, isInteractive, actionOnBack)

This refactor fixes#895.

Type of Change

Bug fix
New feature

Testing

Typecheck, lint, unit tests pass
Manual E2E testing of affected commands

basic-flows.mov

Manually tested flows:

back on inline tui exists instead of going to help.
successful flow on tui exists instead of going back to help.
root level agentcore works as expected.
errors still propagate correctly in inline TUI.
telemetry is now emitted for both inline TUI and full screen.

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 · 2026-05-22T02:51:24Z

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

github-actions · 2026-05-22T02:51:40Z

Package Tarball

aws-agentcore-0.15.0.tgz

How to install

gh release download pr-1365-tarball --repo aws/agentcore-cli --pattern "*.tgz" --dir /tmp/pr-tarball
npm install -g /tmp/pr-tarball/aws-agentcore-0.15.0.tgz

agentcore-cli-automation

Thanks for unifying the TUI entrypoint — the RenderTUIOptions shape and InitialRoute typing are nice. I have a few concerns that I think need to be addressed before this merges.

Summary of issues

Telemetry regression for invoke TUI: The PR removes the withCommandRunTelemetry('invoke', …) wrapper that previously wrapped the TUI invoke flow. No replacement was added, so cli.command_run (with has_stream/has_session_id/auth_type/agent_protocol/duration/exit_reason) is no longer emitted for agentcore invoke interactive mode. See inline comment.
Behavior change: TUI flows no longer auto-exit on success when launched from CLI commands. Previously agentcore add, agentcore remove, etc. rendered their flows with isInteractive={false}, which makes screens like AddFlow, RemoveAllScreen, RemoveFlow, AddSuccessScreen, AddMemoryFlow, AddEvaluatorFlow, AddOnlineEvalFlow, AddIdentityFlow auto-exit after success (see e.g. AddFlow.tsx:206, RemoveAllScreen.tsx:30). Going through App now means they always receive isInteractive={true}, so the user has to press escape to leave after every successful run. See inline comment.
Double TelemetryClientAccessor.init() for command-routed paths. When agentcore add (or any of the migrated commands) runs, main() calls TelemetryClientAccessor.init(args[0], 'cli') (cli.ts:271) and then the action handler calls renderTUI() which calls TelemetryClientAccessor.init(initialRoute.name, 'tui') (cli.ts:127). The first client (mode=cli) is overwritten in the singleton and never shutdown()'d. See inline comment.
Circular import.src/cli/cli.ts imports registerInvoke/registerAdd/etc. from ./commands/*, and those command files now import renderTUI from '../../cli'. It works today because renderTUI is only referenced inside async action handlers, but it's fragile. See inline comment.

Items 1 and 3 are the most important — 1 is a clear telemetry regression and 3 partially undoes the telemetry-mode fix this PR is trying to land.

agentcore-devx-automation · 2026-05-22T03:25:32Z

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

agentcore-cli-automation

Thanks for the cleanup — the unified renderTUI() entrypoint is a nice improvement and the recent commits address most of the earlier review feedback (double init, isInteractive threading, render-tui module extraction).

Flagging two new issues below; both involve information being silently dropped on the way through renderTUI. The other pre-existing review comments still appear open but their underlying concerns look addressed in the latest commits — happy to leave those for the original reviewer to confirm.

agentcore-devx-automation · 2026-05-22T03:33:11Z

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

avi-alpert · 2026-05-26T17:32:38Z

+ const project = await configIO.readProjectSpec().catch(() => undefined);
+ const firstProtocol = project?.runtimes?.[0]?.protocol ?? 'unknown';
+
+ const result = await withCommandRunTelemetry(


When i tested this PR locally, i noticed that the value in the emitted telemetry was always 1 (likely how long it took to load the config).
Before this change, the value was the actual duration of my entire invoke session.

Discussed in person, this is to be consistent with dev, and to avoid tying the duration to arbitrary user sessions.

avi-alpert · 2026-05-26T17:35:32Z

+ process.stdout.write(SHOW_CURSOR);
+ }
+
+ await TelemetryClientAccessor.shutdown();


wouldnt shutting down TelemetryClientAccessor here make it so if telemetry is emitted during a post-exit action (i.e. launchBrowserDev) then it would stop working?

Yes, good call, this is important for when we add telemetry for the browser actions. I've swapped it to flush before the blocking process, then the shutdown should be handled by the global one in the finally block, if it reaches there. Shutdown is "best-effort" so its fine if it doesn't.

agentcore-devx-automation · 2026-05-26T17:37:39Z

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

avi-alpert · 2026-05-26T17:38:25Z

- }
+ enterAltScreen: false,
+ actionOnBack: 'exit',
+ isInteractive: false,


why was isInteractive switched to false here?

good question. I made this change thinking that isInteractive=true meant exit after first success (as it does for other commands), and I didn't want the tui to exit early. However, turns out its just dead code:
agentcore-cli/src/cli/tui/screens/invoke/InvokeScreen.tsx
Line 134 in 69b4ce6
isInteractive: _isInteractive,
. I think we can remove it as a followup.

agentcore-devx-automation · 2026-05-26T18:23:12Z

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

agentcore-devx-automation · 2026-05-26T20:00:15Z

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

agentcore-devx-automation · 2026-05-27T01:58:49Z

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

# Conflicts: # src/cli/cli.ts # src/cli/commands/invoke/command.tsx # src/cli/tui/screens/invoke/useInvokeFlow.ts

agentcore-devx-automation · 2026-05-27T02:10:18Z

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

Hweinstock · 2026-05-22T02:50:48Z

Description

Previously, commands like agentcore add, deploy, etc. rendered TUI screens inline via direct render() calls. This made it difficult to distinguish CLI vs TUI code paths, caused telemetry to be mislabeled as mode="cli" for interactive flows, and meant bare agentcore never initialized or shut down the telemetry client.

This PR creates a unified renderTUI() entrypoint that owns the telemetry lifecycle and replaces all inline render() calls. As part of this:

Extracted rendering logic into src/cli/tui/render.ts and shared notices into src/cli/notices.ts to break circular imports
Migrated add, deploy, create, remove, invoke to use renderTUI()
Made TelemetryClientAccessor.init() handle re-initialization safely
Preserved existing behavior (diffMode, isInteractive, actionOnBack)

This refactor fixes#895.

Type of Change

Bug fix
New feature

Testing

Typecheck, lint, unit tests pass
Manual E2E testing of affected commands

basic-flows.mov

Manually tested flows:

back on inline tui exists instead of going to help.
successful flow on tui exists instead of going back to help.
root level agentcore works as expected.
errors still propagate correctly in inline TUI.
telemetry is now emitted for both inline TUI and full screen.

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 · 2026-05-22T02:51:24Z

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

github-actions · 2026-05-22T02:51:40Z

Package Tarball

aws-agentcore-0.15.0.tgz

How to install

gh release download pr-1365-tarball --repo aws/agentcore-cli --pattern "*.tgz" --dir /tmp/pr-tarball
npm install -g /tmp/pr-tarball/aws-agentcore-0.15.0.tgz

agentcore-cli-automation

Thanks for unifying the TUI entrypoint — the RenderTUIOptions shape and InitialRoute typing are nice. I have a few concerns that I think need to be addressed before this merges.

Summary of issues

Telemetry regression for invoke TUI: The PR removes the withCommandRunTelemetry('invoke', …) wrapper that previously wrapped the TUI invoke flow. No replacement was added, so cli.command_run (with has_stream/has_session_id/auth_type/agent_protocol/duration/exit_reason) is no longer emitted for agentcore invoke interactive mode. See inline comment.
Behavior change: TUI flows no longer auto-exit on success when launched from CLI commands. Previously agentcore add, agentcore remove, etc. rendered their flows with isInteractive={false}, which makes screens like AddFlow, RemoveAllScreen, RemoveFlow, AddSuccessScreen, AddMemoryFlow, AddEvaluatorFlow, AddOnlineEvalFlow, AddIdentityFlow auto-exit after success (see e.g. AddFlow.tsx:206, RemoveAllScreen.tsx:30). Going through App now means they always receive isInteractive={true}, so the user has to press escape to leave after every successful run. See inline comment.
Double TelemetryClientAccessor.init() for command-routed paths. When agentcore add (or any of the migrated commands) runs, main() calls TelemetryClientAccessor.init(args[0], 'cli') (cli.ts:271) and then the action handler calls renderTUI() which calls TelemetryClientAccessor.init(initialRoute.name, 'tui') (cli.ts:127). The first client (mode=cli) is overwritten in the singleton and never shutdown()'d. See inline comment.
Circular import.src/cli/cli.ts imports registerInvoke/registerAdd/etc. from ./commands/*, and those command files now import renderTUI from '../../cli'. It works today because renderTUI is only referenced inside async action handlers, but it's fragile. See inline comment.

Items 1 and 3 are the most important — 1 is a clear telemetry regression and 3 partially undoes the telemetry-mode fix this PR is trying to land.

agentcore-devx-automation · 2026-05-22T03:25:32Z

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

agentcore-cli-automation

Thanks for the cleanup — the unified renderTUI() entrypoint is a nice improvement and the recent commits address most of the earlier review feedback (double init, isInteractive threading, render-tui module extraction).

Flagging two new issues below; both involve information being silently dropped on the way through renderTUI. The other pre-existing review comments still appear open but their underlying concerns look addressed in the latest commits — happy to leave those for the original reviewer to confirm.

agentcore-devx-automation · 2026-05-22T03:33:11Z

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

avi-alpert · 2026-05-26T17:32:38Z

+ const project = await configIO.readProjectSpec().catch(() => undefined);
+ const firstProtocol = project?.runtimes?.[0]?.protocol ?? 'unknown';
+
+ const result = await withCommandRunTelemetry(


When i tested this PR locally, i noticed that the value in the emitted telemetry was always 1 (likely how long it took to load the config).
Before this change, the value was the actual duration of my entire invoke session.

Discussed in person, this is to be consistent with dev, and to avoid tying the duration to arbitrary user sessions.

avi-alpert · 2026-05-26T17:35:32Z

+ process.stdout.write(SHOW_CURSOR);
+ }
+
+ await TelemetryClientAccessor.shutdown();


wouldnt shutting down TelemetryClientAccessor here make it so if telemetry is emitted during a post-exit action (i.e. launchBrowserDev) then it would stop working?

Yes, good call, this is important for when we add telemetry for the browser actions. I've swapped it to flush before the blocking process, then the shutdown should be handled by the global one in the finally block, if it reaches there. Shutdown is "best-effort" so its fine if it doesn't.

agentcore-devx-automation · 2026-05-26T17:37:39Z

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

avi-alpert · 2026-05-26T17:38:25Z

- }
+ enterAltScreen: false,
+ actionOnBack: 'exit',
+ isInteractive: false,


why was isInteractive switched to false here?

good question. I made this change thinking that isInteractive=true meant exit after first success (as it does for other commands), and I didn't want the tui to exit early. However, turns out its just dead code:
agentcore-cli/src/cli/tui/screens/invoke/InvokeScreen.tsx
Line 134 in 69b4ce6
isInteractive: _isInteractive,
. I think we can remove it as a followup.

agentcore-devx-automation · 2026-05-26T18:23:12Z

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

agentcore-devx-automation · 2026-05-26T20:00:15Z

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

agentcore-devx-automation · 2026-05-27T01:58:49Z

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

# Conflicts: # src/cli/cli.ts # src/cli/commands/invoke/command.tsx # src/cli/tui/screens/invoke/useInvokeFlow.ts

agentcore-devx-automation · 2026-05-27T02:10:18Z

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

Hweinstock · 2026-05-22T02:50:48Z

Description

Previously, commands like agentcore add, deploy, etc. rendered TUI screens inline via direct render() calls. This made it difficult to distinguish CLI vs TUI code paths, caused telemetry to be mislabeled as mode="cli" for interactive flows, and meant bare agentcore never initialized or shut down the telemetry client.

This PR creates a unified renderTUI() entrypoint that owns the telemetry lifecycle and replaces all inline render() calls. As part of this:

Extracted rendering logic into src/cli/tui/render.ts and shared notices into src/cli/notices.ts to break circular imports
Migrated add, deploy, create, remove, invoke to use renderTUI()
Made TelemetryClientAccessor.init() handle re-initialization safely
Preserved existing behavior (diffMode, isInteractive, actionOnBack)

This refactor fixes#895.

Type of Change

Bug fix
New feature

Testing

Typecheck, lint, unit tests pass
Manual E2E testing of affected commands

basic-flows.mov

Manually tested flows:

back on inline tui exists instead of going to help.
successful flow on tui exists instead of going back to help.
root level agentcore works as expected.
errors still propagate correctly in inline TUI.
telemetry is now emitted for both inline TUI and full screen.

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 · 2026-05-22T02:51:24Z

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

github-actions · 2026-05-22T02:51:40Z

Package Tarball

aws-agentcore-0.15.0.tgz

How to install

gh release download pr-1365-tarball --repo aws/agentcore-cli --pattern "*.tgz" --dir /tmp/pr-tarball
npm install -g /tmp/pr-tarball/aws-agentcore-0.15.0.tgz

agentcore-cli-automation

Thanks for unifying the TUI entrypoint — the RenderTUIOptions shape and InitialRoute typing are nice. I have a few concerns that I think need to be addressed before this merges.

Summary of issues

Telemetry regression for invoke TUI: The PR removes the withCommandRunTelemetry('invoke', …) wrapper that previously wrapped the TUI invoke flow. No replacement was added, so cli.command_run (with has_stream/has_session_id/auth_type/agent_protocol/duration/exit_reason) is no longer emitted for agentcore invoke interactive mode. See inline comment.
Behavior change: TUI flows no longer auto-exit on success when launched from CLI commands. Previously agentcore add, agentcore remove, etc. rendered their flows with isInteractive={false}, which makes screens like AddFlow, RemoveAllScreen, RemoveFlow, AddSuccessScreen, AddMemoryFlow, AddEvaluatorFlow, AddOnlineEvalFlow, AddIdentityFlow auto-exit after success (see e.g. AddFlow.tsx:206, RemoveAllScreen.tsx:30). Going through App now means they always receive isInteractive={true}, so the user has to press escape to leave after every successful run. See inline comment.
Double TelemetryClientAccessor.init() for command-routed paths. When agentcore add (or any of the migrated commands) runs, main() calls TelemetryClientAccessor.init(args[0], 'cli') (cli.ts:271) and then the action handler calls renderTUI() which calls TelemetryClientAccessor.init(initialRoute.name, 'tui') (cli.ts:127). The first client (mode=cli) is overwritten in the singleton and never shutdown()'d. See inline comment.
Circular import.src/cli/cli.ts imports registerInvoke/registerAdd/etc. from ./commands/*, and those command files now import renderTUI from '../../cli'. It works today because renderTUI is only referenced inside async action handlers, but it's fragile. See inline comment.

Items 1 and 3 are the most important — 1 is a clear telemetry regression and 3 partially undoes the telemetry-mode fix this PR is trying to land.

agentcore-devx-automation · 2026-05-22T03:25:32Z

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

agentcore-cli-automation

Thanks for the cleanup — the unified renderTUI() entrypoint is a nice improvement and the recent commits address most of the earlier review feedback (double init, isInteractive threading, render-tui module extraction).

Flagging two new issues below; both involve information being silently dropped on the way through renderTUI. The other pre-existing review comments still appear open but their underlying concerns look addressed in the latest commits — happy to leave those for the original reviewer to confirm.

agentcore-devx-automation · 2026-05-22T03:33:11Z

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

avi-alpert · 2026-05-26T17:32:38Z

+ const project = await configIO.readProjectSpec().catch(() => undefined);
+ const firstProtocol = project?.runtimes?.[0]?.protocol ?? 'unknown';
+
+ const result = await withCommandRunTelemetry(


When i tested this PR locally, i noticed that the value in the emitted telemetry was always 1 (likely how long it took to load the config).
Before this change, the value was the actual duration of my entire invoke session.

Discussed in person, this is to be consistent with dev, and to avoid tying the duration to arbitrary user sessions.

avi-alpert · 2026-05-26T17:35:32Z

+ process.stdout.write(SHOW_CURSOR);
+ }
+
+ await TelemetryClientAccessor.shutdown();


wouldnt shutting down TelemetryClientAccessor here make it so if telemetry is emitted during a post-exit action (i.e. launchBrowserDev) then it would stop working?

Yes, good call, this is important for when we add telemetry for the browser actions. I've swapped it to flush before the blocking process, then the shutdown should be handled by the global one in the finally block, if it reaches there. Shutdown is "best-effort" so its fine if it doesn't.

agentcore-devx-automation · 2026-05-26T17:37:39Z

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

avi-alpert · 2026-05-26T17:38:25Z

- }
+ enterAltScreen: false,
+ actionOnBack: 'exit',
+ isInteractive: false,


why was isInteractive switched to false here?

good question. I made this change thinking that isInteractive=true meant exit after first success (as it does for other commands), and I didn't want the tui to exit early. However, turns out its just dead code:
agentcore-cli/src/cli/tui/screens/invoke/InvokeScreen.tsx
Line 134 in 69b4ce6
isInteractive: _isInteractive,
. I think we can remove it as a followup.

agentcore-devx-automation · 2026-05-26T18:23:12Z

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

agentcore-devx-automation · 2026-05-26T20:00:15Z

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

agentcore-devx-automation · 2026-05-27T01:58:49Z

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

# Conflicts: # src/cli/cli.ts # src/cli/commands/invoke/command.tsx # src/cli/tui/screens/invoke/useInvokeFlow.ts

agentcore-devx-automation · 2026-05-27T02:10:18Z

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

Hweinstock · 2026-05-22T02:50:48Z

Description

Previously, commands like agentcore add, deploy, etc. rendered TUI screens inline via direct render() calls. This made it difficult to distinguish CLI vs TUI code paths, caused telemetry to be mislabeled as mode="cli" for interactive flows, and meant bare agentcore never initialized or shut down the telemetry client.

This PR creates a unified renderTUI() entrypoint that owns the telemetry lifecycle and replaces all inline render() calls. As part of this:

Extracted rendering logic into src/cli/tui/render.ts and shared notices into src/cli/notices.ts to break circular imports
Migrated add, deploy, create, remove, invoke to use renderTUI()
Made TelemetryClientAccessor.init() handle re-initialization safely
Preserved existing behavior (diffMode, isInteractive, actionOnBack)

This refactor fixes#895.

Type of Change

Bug fix
New feature

Testing

Typecheck, lint, unit tests pass
Manual E2E testing of affected commands

basic-flows.mov

Manually tested flows:

back on inline tui exists instead of going to help.
successful flow on tui exists instead of going back to help.
root level agentcore works as expected.
errors still propagate correctly in inline TUI.
telemetry is now emitted for both inline TUI and full screen.

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 · 2026-05-22T02:51:24Z

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

github-actions · 2026-05-22T02:51:40Z

Package Tarball

aws-agentcore-0.15.0.tgz

How to install

gh release download pr-1365-tarball --repo aws/agentcore-cli --pattern "*.tgz" --dir /tmp/pr-tarball
npm install -g /tmp/pr-tarball/aws-agentcore-0.15.0.tgz

agentcore-cli-automation

Thanks for unifying the TUI entrypoint — the RenderTUIOptions shape and InitialRoute typing are nice. I have a few concerns that I think need to be addressed before this merges.

Summary of issues

Telemetry regression for invoke TUI: The PR removes the withCommandRunTelemetry('invoke', …) wrapper that previously wrapped the TUI invoke flow. No replacement was added, so cli.command_run (with has_stream/has_session_id/auth_type/agent_protocol/duration/exit_reason) is no longer emitted for agentcore invoke interactive mode. See inline comment.
Behavior change: TUI flows no longer auto-exit on success when launched from CLI commands. Previously agentcore add, agentcore remove, etc. rendered their flows with isInteractive={false}, which makes screens like AddFlow, RemoveAllScreen, RemoveFlow, AddSuccessScreen, AddMemoryFlow, AddEvaluatorFlow, AddOnlineEvalFlow, AddIdentityFlow auto-exit after success (see e.g. AddFlow.tsx:206, RemoveAllScreen.tsx:30). Going through App now means they always receive isInteractive={true}, so the user has to press escape to leave after every successful run. See inline comment.
Double TelemetryClientAccessor.init() for command-routed paths. When agentcore add (or any of the migrated commands) runs, main() calls TelemetryClientAccessor.init(args[0], 'cli') (cli.ts:271) and then the action handler calls renderTUI() which calls TelemetryClientAccessor.init(initialRoute.name, 'tui') (cli.ts:127). The first client (mode=cli) is overwritten in the singleton and never shutdown()'d. See inline comment.
Circular import.src/cli/cli.ts imports registerInvoke/registerAdd/etc. from ./commands/*, and those command files now import renderTUI from '../../cli'. It works today because renderTUI is only referenced inside async action handlers, but it's fragile. See inline comment.

Items 1 and 3 are the most important — 1 is a clear telemetry regression and 3 partially undoes the telemetry-mode fix this PR is trying to land.

agentcore-devx-automation · 2026-05-22T03:25:32Z

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

agentcore-cli-automation

Thanks for the cleanup — the unified renderTUI() entrypoint is a nice improvement and the recent commits address most of the earlier review feedback (double init, isInteractive threading, render-tui module extraction).

Flagging two new issues below; both involve information being silently dropped on the way through renderTUI. The other pre-existing review comments still appear open but their underlying concerns look addressed in the latest commits — happy to leave those for the original reviewer to confirm.

agentcore-devx-automation · 2026-05-22T03:33:11Z

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

avi-alpert · 2026-05-26T17:32:38Z

+ const project = await configIO.readProjectSpec().catch(() => undefined);
+ const firstProtocol = project?.runtimes?.[0]?.protocol ?? 'unknown';
+
+ const result = await withCommandRunTelemetry(


When i tested this PR locally, i noticed that the value in the emitted telemetry was always 1 (likely how long it took to load the config).
Before this change, the value was the actual duration of my entire invoke session.

Discussed in person, this is to be consistent with dev, and to avoid tying the duration to arbitrary user sessions.

avi-alpert · 2026-05-26T17:35:32Z

+ process.stdout.write(SHOW_CURSOR);
+ }
+
+ await TelemetryClientAccessor.shutdown();


wouldnt shutting down TelemetryClientAccessor here make it so if telemetry is emitted during a post-exit action (i.e. launchBrowserDev) then it would stop working?

Yes, good call, this is important for when we add telemetry for the browser actions. I've swapped it to flush before the blocking process, then the shutdown should be handled by the global one in the finally block, if it reaches there. Shutdown is "best-effort" so its fine if it doesn't.

agentcore-devx-automation · 2026-05-26T17:37:39Z

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

avi-alpert · 2026-05-26T17:38:25Z

- }
+ enterAltScreen: false,
+ actionOnBack: 'exit',
+ isInteractive: false,


why was isInteractive switched to false here?

good question. I made this change thinking that isInteractive=true meant exit after first success (as it does for other commands), and I didn't want the tui to exit early. However, turns out its just dead code:
agentcore-cli/src/cli/tui/screens/invoke/InvokeScreen.tsx
Line 134 in 69b4ce6
isInteractive: _isInteractive,
. I think we can remove it as a followup.

agentcore-devx-automation · 2026-05-26T18:23:12Z

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

agentcore-devx-automation · 2026-05-26T20:00:15Z

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

agentcore-devx-automation · 2026-05-27T01:58:49Z

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

# Conflicts: # src/cli/cli.ts # src/cli/commands/invoke/command.tsx # src/cli/tui/screens/invoke/useInvokeFlow.ts

agentcore-devx-automation · 2026-05-27T02:10:18Z

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

Hweinstock · 2026-05-22T02:50:48Z

Description

Previously, commands like agentcore add, deploy, etc. rendered TUI screens inline via direct render() calls. This made it difficult to distinguish CLI vs TUI code paths, caused telemetry to be mislabeled as mode="cli" for interactive flows, and meant bare agentcore never initialized or shut down the telemetry client.

This PR creates a unified renderTUI() entrypoint that owns the telemetry lifecycle and replaces all inline render() calls. As part of this:

Extracted rendering logic into src/cli/tui/render.ts and shared notices into src/cli/notices.ts to break circular imports
Migrated add, deploy, create, remove, invoke to use renderTUI()
Made TelemetryClientAccessor.init() handle re-initialization safely
Preserved existing behavior (diffMode, isInteractive, actionOnBack)

This refactor fixes#895.

Type of Change

Bug fix
New feature

Testing

Typecheck, lint, unit tests pass
Manual E2E testing of affected commands

basic-flows.mov

Manually tested flows:

back on inline tui exists instead of going to help.
successful flow on tui exists instead of going back to help.
root level agentcore works as expected.
errors still propagate correctly in inline TUI.
telemetry is now emitted for both inline TUI and full screen.

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 · 2026-05-22T02:51:24Z

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

github-actions · 2026-05-22T02:51:40Z

Package Tarball

aws-agentcore-0.15.0.tgz

How to install

gh release download pr-1365-tarball --repo aws/agentcore-cli --pattern "*.tgz" --dir /tmp/pr-tarball
npm install -g /tmp/pr-tarball/aws-agentcore-0.15.0.tgz

agentcore-cli-automation

Thanks for unifying the TUI entrypoint — the RenderTUIOptions shape and InitialRoute typing are nice. I have a few concerns that I think need to be addressed before this merges.

Summary of issues

Telemetry regression for invoke TUI: The PR removes the withCommandRunTelemetry('invoke', …) wrapper that previously wrapped the TUI invoke flow. No replacement was added, so cli.command_run (with has_stream/has_session_id/auth_type/agent_protocol/duration/exit_reason) is no longer emitted for agentcore invoke interactive mode. See inline comment.
Behavior change: TUI flows no longer auto-exit on success when launched from CLI commands. Previously agentcore add, agentcore remove, etc. rendered their flows with isInteractive={false}, which makes screens like AddFlow, RemoveAllScreen, RemoveFlow, AddSuccessScreen, AddMemoryFlow, AddEvaluatorFlow, AddOnlineEvalFlow, AddIdentityFlow auto-exit after success (see e.g. AddFlow.tsx:206, RemoveAllScreen.tsx:30). Going through App now means they always receive isInteractive={true}, so the user has to press escape to leave after every successful run. See inline comment.
Double TelemetryClientAccessor.init() for command-routed paths. When agentcore add (or any of the migrated commands) runs, main() calls TelemetryClientAccessor.init(args[0], 'cli') (cli.ts:271) and then the action handler calls renderTUI() which calls TelemetryClientAccessor.init(initialRoute.name, 'tui') (cli.ts:127). The first client (mode=cli) is overwritten in the singleton and never shutdown()'d. See inline comment.
Circular import.src/cli/cli.ts imports registerInvoke/registerAdd/etc. from ./commands/*, and those command files now import renderTUI from '../../cli'. It works today because renderTUI is only referenced inside async action handlers, but it's fragile. See inline comment.

Items 1 and 3 are the most important — 1 is a clear telemetry regression and 3 partially undoes the telemetry-mode fix this PR is trying to land.

agentcore-devx-automation · 2026-05-22T03:25:32Z

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

agentcore-cli-automation

Thanks for the cleanup — the unified renderTUI() entrypoint is a nice improvement and the recent commits address most of the earlier review feedback (double init, isInteractive threading, render-tui module extraction).

Flagging two new issues below; both involve information being silently dropped on the way through renderTUI. The other pre-existing review comments still appear open but their underlying concerns look addressed in the latest commits — happy to leave those for the original reviewer to confirm.

agentcore-devx-automation · 2026-05-22T03:33:11Z

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

avi-alpert · 2026-05-26T17:32:38Z

+ const project = await configIO.readProjectSpec().catch(() => undefined);
+ const firstProtocol = project?.runtimes?.[0]?.protocol ?? 'unknown';
+
+ const result = await withCommandRunTelemetry(


When i tested this PR locally, i noticed that the value in the emitted telemetry was always 1 (likely how long it took to load the config).
Before this change, the value was the actual duration of my entire invoke session.

Discussed in person, this is to be consistent with dev, and to avoid tying the duration to arbitrary user sessions.

avi-alpert · 2026-05-26T17:35:32Z

+ process.stdout.write(SHOW_CURSOR);
+ }
+
+ await TelemetryClientAccessor.shutdown();


wouldnt shutting down TelemetryClientAccessor here make it so if telemetry is emitted during a post-exit action (i.e. launchBrowserDev) then it would stop working?

Yes, good call, this is important for when we add telemetry for the browser actions. I've swapped it to flush before the blocking process, then the shutdown should be handled by the global one in the finally block, if it reaches there. Shutdown is "best-effort" so its fine if it doesn't.

agentcore-devx-automation · 2026-05-26T17:37:39Z

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

avi-alpert · 2026-05-26T17:38:25Z

- }
+ enterAltScreen: false,
+ actionOnBack: 'exit',
+ isInteractive: false,


why was isInteractive switched to false here?

good question. I made this change thinking that isInteractive=true meant exit after first success (as it does for other commands), and I didn't want the tui to exit early. However, turns out its just dead code:
agentcore-cli/src/cli/tui/screens/invoke/InvokeScreen.tsx
Line 134 in 69b4ce6
isInteractive: _isInteractive,
. I think we can remove it as a followup.

agentcore-devx-automation · 2026-05-26T18:23:12Z

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

agentcore-devx-automation · 2026-05-26T20:00:15Z

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

agentcore-devx-automation · 2026-05-27T01:58:49Z

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

# Conflicts: # src/cli/cli.ts # src/cli/commands/invoke/command.tsx # src/cli/tui/screens/invoke/useInvokeFlow.ts

agentcore-devx-automation · 2026-05-27T02:10:18Z

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

Hweinstock · 2026-05-22T02:50:48Z

Description

Previously, commands like agentcore add, deploy, etc. rendered TUI screens inline via direct render() calls. This made it difficult to distinguish CLI vs TUI code paths, caused telemetry to be mislabeled as mode="cli" for interactive flows, and meant bare agentcore never initialized or shut down the telemetry client.

This PR creates a unified renderTUI() entrypoint that owns the telemetry lifecycle and replaces all inline render() calls. As part of this:

Extracted rendering logic into src/cli/tui/render.ts and shared notices into src/cli/notices.ts to break circular imports
Migrated add, deploy, create, remove, invoke to use renderTUI()
Made TelemetryClientAccessor.init() handle re-initialization safely
Preserved existing behavior (diffMode, isInteractive, actionOnBack)

This refactor fixes#895.

Type of Change

Bug fix
New feature

Testing

Typecheck, lint, unit tests pass
Manual E2E testing of affected commands

basic-flows.mov

Manually tested flows:

back on inline tui exists instead of going to help.
successful flow on tui exists instead of going back to help.
root level agentcore works as expected.
errors still propagate correctly in inline TUI.
telemetry is now emitted for both inline TUI and full screen.

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 · 2026-05-22T02:51:24Z

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

github-actions · 2026-05-22T02:51:40Z

Package Tarball

aws-agentcore-0.15.0.tgz

How to install

gh release download pr-1365-tarball --repo aws/agentcore-cli --pattern "*.tgz" --dir /tmp/pr-tarball
npm install -g /tmp/pr-tarball/aws-agentcore-0.15.0.tgz

agentcore-cli-automation

Thanks for unifying the TUI entrypoint — the RenderTUIOptions shape and InitialRoute typing are nice. I have a few concerns that I think need to be addressed before this merges.

Summary of issues

Telemetry regression for invoke TUI: The PR removes the withCommandRunTelemetry('invoke', …) wrapper that previously wrapped the TUI invoke flow. No replacement was added, so cli.command_run (with has_stream/has_session_id/auth_type/agent_protocol/duration/exit_reason) is no longer emitted for agentcore invoke interactive mode. See inline comment.
Behavior change: TUI flows no longer auto-exit on success when launched from CLI commands. Previously agentcore add, agentcore remove, etc. rendered their flows with isInteractive={false}, which makes screens like AddFlow, RemoveAllScreen, RemoveFlow, AddSuccessScreen, AddMemoryFlow, AddEvaluatorFlow, AddOnlineEvalFlow, AddIdentityFlow auto-exit after success (see e.g. AddFlow.tsx:206, RemoveAllScreen.tsx:30). Going through App now means they always receive isInteractive={true}, so the user has to press escape to leave after every successful run. See inline comment.
Double TelemetryClientAccessor.init() for command-routed paths. When agentcore add (or any of the migrated commands) runs, main() calls TelemetryClientAccessor.init(args[0], 'cli') (cli.ts:271) and then the action handler calls renderTUI() which calls TelemetryClientAccessor.init(initialRoute.name, 'tui') (cli.ts:127). The first client (mode=cli) is overwritten in the singleton and never shutdown()'d. See inline comment.
Circular import.src/cli/cli.ts imports registerInvoke/registerAdd/etc. from ./commands/*, and those command files now import renderTUI from '../../cli'. It works today because renderTUI is only referenced inside async action handlers, but it's fragile. See inline comment.

Items 1 and 3 are the most important — 1 is a clear telemetry regression and 3 partially undoes the telemetry-mode fix this PR is trying to land.

agentcore-devx-automation · 2026-05-22T03:25:32Z

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

agentcore-cli-automation

Thanks for the cleanup — the unified renderTUI() entrypoint is a nice improvement and the recent commits address most of the earlier review feedback (double init, isInteractive threading, render-tui module extraction).

Flagging two new issues below; both involve information being silently dropped on the way through renderTUI. The other pre-existing review comments still appear open but their underlying concerns look addressed in the latest commits — happy to leave those for the original reviewer to confirm.

agentcore-devx-automation · 2026-05-22T03:33:11Z

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

avi-alpert · 2026-05-26T17:32:38Z

+ const project = await configIO.readProjectSpec().catch(() => undefined);
+ const firstProtocol = project?.runtimes?.[0]?.protocol ?? 'unknown';
+
+ const result = await withCommandRunTelemetry(


When i tested this PR locally, i noticed that the value in the emitted telemetry was always 1 (likely how long it took to load the config).
Before this change, the value was the actual duration of my entire invoke session.

Discussed in person, this is to be consistent with dev, and to avoid tying the duration to arbitrary user sessions.

avi-alpert · 2026-05-26T17:35:32Z

+ process.stdout.write(SHOW_CURSOR);
+ }
+
+ await TelemetryClientAccessor.shutdown();


wouldnt shutting down TelemetryClientAccessor here make it so if telemetry is emitted during a post-exit action (i.e. launchBrowserDev) then it would stop working?

Yes, good call, this is important for when we add telemetry for the browser actions. I've swapped it to flush before the blocking process, then the shutdown should be handled by the global one in the finally block, if it reaches there. Shutdown is "best-effort" so its fine if it doesn't.

agentcore-devx-automation · 2026-05-26T17:37:39Z

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

avi-alpert · 2026-05-26T17:38:25Z

- }
+ enterAltScreen: false,
+ actionOnBack: 'exit',
+ isInteractive: false,


why was isInteractive switched to false here?

good question. I made this change thinking that isInteractive=true meant exit after first success (as it does for other commands), and I didn't want the tui to exit early. However, turns out its just dead code:
agentcore-cli/src/cli/tui/screens/invoke/InvokeScreen.tsx
Line 134 in 69b4ce6
isInteractive: _isInteractive,
. I think we can remove it as a followup.

agentcore-devx-automation · 2026-05-26T18:23:12Z

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

agentcore-devx-automation · 2026-05-26T20:00:15Z

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

agentcore-devx-automation · 2026-05-27T01:58:49Z

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

# Conflicts: # src/cli/cli.ts # src/cli/commands/invoke/command.tsx # src/cli/tui/screens/invoke/useInvokeFlow.ts

agentcore-devx-automation · 2026-05-27T02:10:18Z

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

Hweinstock temporarily deployed to e2e-testing May 22, 2026 02:50 — with GitHub Actions Inactive

github-actionsBot added the size/m PR size: M label May 22, 2026

agentcore-devx-automationBot added the claude-security-reviewing Claude Code /security-review in progress label May 22, 2026

github-actionsBot added the agentcore-harness-reviewing AgentCore Harness review in progress label May 22, 2026

agentcore-devx-automationBot removed the claude-security-reviewing Claude Code /security-review in progress label May 22, 2026

agentcore-cli-automation suggested changes May 22, 2026
View reviewed changes

Comment threadsrc/cli/commands/invoke/command.tsx Outdated
Comment threadsrc/cli/tui/App.tsx Outdated
Comment threadsrc/cli/cli.ts Outdated
Comment threadsrc/cli/commands/invoke/command.tsx Outdated

github-actionsBot added size/l PR size: L and removed agentcore-harness-reviewing AgentCore Harness review in progress size/m PR size: M labels May 22, 2026

Hweinstock temporarily deployed to e2e-testing May 22, 2026 03:25 — with GitHub Actions Inactive

Hweinstock closed this May 22, 2026

Hweinstock reopened this May 22, 2026

agentcore-devx-automationBot added the claude-security-reviewing Claude Code /security-review in progress label May 22, 2026

github-actionsBot added size/l PR size: L agentcore-harness-reviewing AgentCore Harness review in progress and removed size/l PR size: L labels May 22, 2026

agentcore-devx-automationBot removed the claude-security-reviewing Claude Code /security-review in progress label May 22, 2026

agentcore-cli-automation suggested changes May 22, 2026
View reviewed changes

Comment threadsrc/cli/commands/deploy/command.tsx
Comment threadsrc/cli/tui/screens/invoke/useInvokeFlow.ts Outdated

github-actionsBot removed the agentcore-harness-reviewing AgentCore Harness review in progress label May 22, 2026

Hweinstock temporarily deployed to e2e-testing May 22, 2026 03:30 — with GitHub Actions Inactive

Hweinstock force-pushed the fix/tui-telemetry-init branch from 9456078 to bce4cf9Compare May 22, 2026 03:32

github-actionsBot added size/l PR size: L and removed size/l PR size: L labels May 22, 2026

agentcore-devx-automationBot added the claude-security-reviewing Claude Code /security-review in progress label May 22, 2026

agentcore-devx-automationBot removed the claude-security-reviewing Claude Code /security-review in progress label May 22, 2026

Hweinstock force-pushed the fix/tui-telemetry-init branch from bce4cf9 to 8d0f3b3Compare May 22, 2026 03:33

avi-alpert reviewed May 26, 2026
View reviewed changes

Comment threadsrc/cli/commands/remove/command.tsx Outdated

avi-alpert reviewed May 26, 2026
View reviewed changes

Hweinstock force-pushed the fix/tui-telemetry-init branch from 7f6cf1a to 9bad38dCompare May 26, 2026 17:36

github-actionsBot added size/l PR size: L and removed size/l PR size: L labels May 26, 2026

Hweinstock temporarily deployed to e2e-testing May 26, 2026 17:36 — with GitHub Actions Inactive

agentcore-devx-automationBot added the claude-security-reviewing Claude Code /security-review in progress label May 26, 2026

agentcore-devx-automationBot removed the claude-security-reviewing Claude Code /security-review in progress label May 26, 2026

avi-alpert reviewed May 26, 2026
View reviewed changes

fix: maintain remove all entrypoint
6600562

github-actionsBot removed the size/l PR size: L label May 26, 2026

Hweinstock temporarily deployed to e2e-testing May 26, 2026 18:22 — with GitHub Actions Inactive

github-actionsBot added the size/l PR size: L label May 26, 2026

agentcore-devx-automationBot added the claude-security-reviewing Claude Code /security-review in progress label May 26, 2026

agentcore-devx-automationBot removed the claude-security-reviewing Claude Code /security-review in progress label May 26, 2026

fix(telemetry): avoid early client shutdown
4b78fc1

github-actionsBot added size/l PR size: L and removed size/l PR size: L labels May 26, 2026

Hweinstock temporarily deployed to e2e-testing May 26, 2026 19:59 — with GitHub Actions Inactive

agentcore-devx-automationBot added the claude-security-reviewing Claude Code /security-review in progress label May 26, 2026

avi-alpert previously approved these changes May 26, 2026
View reviewed changes

Merge branch 'main' into fix/tui-telemetry-init
b4d2a09
# Conflicts: # src/cli/cli.ts # src/cli/commands/invoke/command.tsx # src/cli/tui/screens/invoke/useInvokeFlow.ts

avi-alpert approved these changes May 27, 2026
View reviewed changes

aidandaly24 mentioned this pull request Jun 25, 2026
fix(cli): CLI-side defensive exit only. Did not bump ink 6.8.0 -> 7... (#204) aidandaly24/agentcore-cli#53
Draft