🛋️ feat: Open Subagents Directly from Background Tasks

[lia-by-librechat]

Summary

The background-task popup's subagent arrow only scrolls to the dispatch call, which fails when that message is not loaded. It now opens the requested durable subagent task directly, without changing the visible chat. The action is labeled “Open subagent”; the popup gains no explanatory footer.

Durable activity retains its own parent context, draft, and accumulated live text/tool activity when the visible conversation changes. Navigation releases parent-stream ownership without clearing the selected task; leaving the chat host clears every invocation. A separate “Go to parent chat” control navigates only when requested. This first slice uses the existing resizable side panel; it does not introduce a second draggable-window implementation. Foreground-only activity still closes on navigation, and leaving the authenticated chat host clears the selection.

How it works

Popup row (thread + task)
  durableSubagentSelection
    existing panel host
      task-owned parent context
      authorized thread query and activity stream

Tool rows keep their existing scroll-to-call behavior. Event-origin subagents retain the actor timeline and pin the requested task. Opening transfers keyboard focus to the panel; closing can return to the task disclosure when the call is not rendered.

Type of change

• Feature
• Tests / tooling / CI

Testing

Latest head: 530b737474a8b91c1f90a01e9ef4eb4c39a5bb8b.

The permission-grant scenario records notification tags and asserts exactly one for its target conversation, rather than counting unrelated reply alerts from shared user history. It observes that reply’s unread dot before granting permission and confirms no notifications were constructed while permission was pending. It reuses the persisted reply timestamp. Production notification code is unchanged.

Passed locally:

• Real tracked Playwright permission-grant scenario, Chromium, normal polling cadence.
• Diagnostic browser run with older unread history outside the initial sidebar page. Constructor breakpoints confirmed distinct target and unrelated conversation tags; the target notified exactly once.
• 95 tests across notification-alert, unseen-conversation, and reply-watcher suites.
• Scenario formatting and whitespace checks. Frontend dependency packages and frontend app built successfully.

Client npx tsc --noEmit passed. Independent review of this exact head completed with no findings. CI client typechecking, package/Vite builds, API runtime smoke, and Lighthouse passed. The replacement browser shard 8/10 and other test/static lanes are still running, so the earlier CI failure is not yet confirmed cleared. Full Playwright suites and Lighthouse were not run locally.

[lia-by-librechat]

Head: 79440a1ef65b7a182ff9b6ca406d44c6095ae39f. Opens durable subagents directly from the task popup, preserves task-owned context across chat navigation, and separates parent-chat navigation. Uses the current resizable host; no popup footer or competing window shell. Independent review is running for this head.

[lia-by-librechat]

Head: 07032432c01d54a8a85cd88e60daa8f3f2b64cee. Uses route reconciliation for explicit parent navigation, keeps automatic task advancement from stealing chat focus, and reuses parent discovery until the panel and visible chat diverge. Adds transition coverage and fixes the initial CI type/lint diagnostics.

[lia-by-librechat]

Head: e28a064bfe9e9f9369c35287982053351d0bbb1f. Guards retained task identity against dispatch synchronization and delayed event refreshes, preserves mounted live-state readers across navigation cleanup, and skips hidden or disabled return-focus targets. Includes regression coverage for each transition.

[lia-by-librechat]

Head: e21c364ec98a843432916ac3bfbe243238403ebd. Focused verification passes: 416 tests across 11 suites, covering task opening, navigation retention, live progress, dispatch identity, and focus. Includes the corrected icon stub and direct-store observation in the test harness.

[lia-by-librechat]

Head: 2283119b258b4f266711a666df6b80be193d63e8. Preserves accumulated text and tool activity for the selected durable task during chat navigation, releases parent-stream ownership, and keeps final host teardown unconditional. Adds real drain regressions for tool/event tasks and caller coverage for navigation versus teardown.

[lia-by-librechat]

Head: e9bc679e21942126b37c35d8285d6d5fa17efd17, integrated with current dev (8829a3975c14). Retains selected task activity across navigation and restores the core progress-key import after the step-handler migration. Local verification: 568 tests across 22 focused suites, client tsc --noEmit, scoped ESLint, formatting/imports, JSON, and localization checks passed. Independent review will cover this exact integrated head.

[lia-by-librechat]

Reviewed head: e9bc679e21942126b37c35d8285d6d5fa17efd17. Independent review complete: no findings. Previous P1 navigation-settings, P2 task identity/activity, and P3 focus findings are fixed; no rejections. Local: 568 tests / 22 suites, client tsc --noEmit, scoped ESLint, formatting/imports, JSON, localization, and whitespace checks passed. CI typechecking, static checks, package/Vite builds, and Lighthouse passed; Ubuntu tests and memory/Redis e2e lanes remain in progress.

[danny-avila]

@codex review the latest head, final review

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-16T22:03:03.248811Z	b408793	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

Codex Review: Something went wrong. Try again later by commenting “@codex review”.

An unknown error occurred

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[danny-avila]

@codex review the latest head, final review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e9bc679e21

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[lia-by-librechat]

Head: ed55a3f089d842b0008842abc70d48439f3fa154. Refreshes the saved external focus-return target on every explicit subagent open. Added mounted-panel regressions for switching to an unloaded task and reopening the same task. Local focused suites: 181 tests across 5 suites passed; client tsc --noEmit and scoped static checks passed. Independent review is running for this head.

[lia-by-librechat]

Reviewed head: ed55a3f089d842b0008842abc70d48439f3fa154. Independent review complete: no findings. Codex’s P2 focus-return finding is fixed and its thread resolved; none rejected. Local: 181 tests across five suites, client tsc --noEmit, scoped lint/static checks, and frontend dependency builds passed. CI typechecking, package/Vite builds, and Lighthouse passed; remaining checks are still running.

[danny-avila]

@codex review the latest head, final review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ed55a3f089

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[lia-by-librechat]

Head: b408793f58d16bcef7b5eaad657d1ce977e9f623. Prefers the actual opener on close even when its message card remains loaded. Desktop/mobile coverage verifies opener priority and unavailable-opener fallbacks. Local: 254 tests / seven suites, client tsc --noEmit, scoped ESLint/Prettier/import/JSON/localization checks passed. Independent review is running for this exact head.

[danny-avila]

@codex review the latest head, final review

[lia-by-librechat]

Reviewed head: b408793f58d16bcef7b5eaad657d1ce977e9f623. Independent review complete: no findings. Both Codex P2 focus-return findings are fixed and resolved; none rejected. Local verification: 254 tests across seven suites, client tsc --noEmit, scoped lint/static checks, and frontend dependency builds passed. CI client typechecking, package/Vite builds, and Lighthouse passed. Static checks, Ubuntu tests, runtime smoke, and memory/Redis e2e remain in progress.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Can't wait for the next one!

Reviewed commit: b408793f58

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[lia-by-librechat]

Head: dbc848a9b1d89182130558a591f2b8359e084264. Fixes the failed permission-grant notification scenario without changing production behavior or weakening its one-notification assertion. The fixture now reuses the real completed reply timestamp instead of advancing lastResponseAt twice. The failed browser trace showed distinct reply stamps; a disposable check of the actual alert hook produced one notification with a stable stamp and two when it advanced. Local notification/read-state/watcher suites: 95 tests passed; client tsc --noEmit and scenario formatting/lint passed. Fresh CI and independent review are running for this head.

[lia-by-librechat]

Reviewed head: dbc848a9b1d89182130558a591f2b8359e084264. Independent review complete: no findings. The CI-fixture correction keeps the persisted reply stamp stable and preserves the one-notification assertion; production notification behavior is unchanged. Local: 95 focused tests plus two real-hook timestamp checks, client tsc --noEmit, scenario lint/formatting, and focused Playwright discovery passed. CI client typechecking, package/Vite builds, and Lighthouse passed. The replacement memory shard 8/10 is still running; the original browser failure is not yet confirmed cleared.

[lia-by-librechat]

Head: 530b737474a8b91c1f90a01e9ef4eb4c39a5bb8b. The remaining permission-grant failure counted notifications for different conversations. A real browser reproduction with older unread history confirmed distinct notification tags at constructor breakpoints. The scenario now counts only its target tag and waits for that reply’s unread indicator before granting permission, so it verifies an observed withheld arrival and still rejects duplicate notifications for the target. The tracked scenario passed locally at its normal cadence; a diagnostic browser run with unrelated backlog also passed. Production notification code is unchanged. Fresh CI and independent review cover this head.

[lia-by-librechat]

Reviewed head: 530b737474a8b91c1f90a01e9ef4eb4c39a5bb8b. Independent review complete: no findings. Previous findings are fixed; none rejected. The tracked permission scenario passed locally at normal cadence, and browser stress coverage with unrelated unread history also passed. Local notification suites: 95 tests passed; client tsc --noEmit and scenario lint/formatting passed. CI typechecking, builds, runtime smoke, and Lighthouse passed. Replacement memory shard 8/10 remains in progress; the earlier CI failure is not yet confirmed cleared.