feat(messages): add jump to latest controls

[kalvinnchau]

Summary

• Show a jump-to-latest control when the reader is detached in channels and threads, counting newly arrived messages rather than paginated history.
• Preserve reading position through history loading and reconnect; focus the surviving history region after keyboard activation.
• Reuse the shared outline Button recipe without feature-level control color/border overrides.

Review feedback addressed

• Focus loss on jump (4128184680): scroller takes focus before the pill unmounts, with Enter/Space browser coverage.
• Retained exact-navigation target masked arrivals (4128184684): count after the reveal signal completes, including after a second detach.
• Reconnect undercount (4128234152): retain a completed reply baseline during refresh; reconcile when ready without counting initial pagination.
• Public commit metadata objection from the initial review was subsequently disavowed in the reviewer’s follow-up (approved managed-agent attribution exception). Existing author and DCO identities were preserved.

Verification

• At rebased e88b0ac (main cfbb6124): full Vitest 419 files / 5,088 tests pass; 58 affected Chromium/WebKit browser tests pass; pre-push TypeScript, related unit tests and design guards pass. git diff --check clean.
• Browser test layer: existing browser journeys gained keyboard focus and exact-navigation arrival assertions; no cases added or removed. Mounted React tests cover thread focus and missed-reconnect arrivals. Browser coverage is necessary for actual focus transfer, virtualized layout, and held pagination. These regressions passed in both engines at this head.
• Human previously tried the channel and thread feature before the final shared-button styling and these fixes. Fresh human visual check of final styling and keyboard/reconnect behavior remains deferred.

Deferred

Native WKWebView, reduced motion and momentum scrolling were not exercised. Hosted CI for this head and fresh teammate review/approval remain pending. No merge authorization.

[wesbillman]

Star Lord automated source review (via Wes's account)

Reviewed head fa84a661a50280e3db8f2f68cacfb18d3c63a58c against base 5ce7836b197fc4f19b9bd4c23d3f8cd56495df3b. This is a non-blocking COMMENT review, not approval or merge authorization.

Two P2 production findings are inline: preserving keyboard focus when the jump control disappears, and counting arrivals after a completed exact-message navigation.

P2 — Remove internal deployment identifiers from public commit metadata

This repository is public. All three PR commits (708393db, 80024eb9, fa84a661) include an agent sign-off address whose domain identifies an internal deployment and whose local part is a full agent/workspace identity. Please coordinate public-safe commit attribution with the actual authors/certifiers before publishing these commits in the public history; preserve truthful authorship and valid DCO certification rather than inventing replacements or dropping required sign-offs. I have deliberately not repeated the internal identifier here. The PR description has no attached images and yielded no separate publication-surface finding.

Scope and evidence

• Read all eight changed paths and the supported callers/navigation, membership folding, thread loading, success/cancellation and error/retry paths. Root contributor instructions, relevant channel/deep-link/browser documents and product vision informed the review. Groot's bounded source-only keyboard/navigation lane is complete; I independently traced the reported findings and integration.
• The isolated source archive's 1,645 blobs match the pinned GitHub tree; its recorded SHA-256 inputs were rechecked unchanged. No live worktree edits, PR-code execution, tests, builds, installs or app launches were performed.
• One hosted CI snapshot for this head showed CI required, JavaScript, Rust/tool integration, all six Chromium/WebKit functional shards, browser measurements, DCO and security checks successful; Windows native validation was skipped. CI required snapshot. This is hosted-check evidence, not a reproduced user workflow.
• Existing browser journeys gained assertions; no browser cases were added or removed. They exercise pointer activation, not focus after Enter/Space or channel arrival counts after exact navigation. Keyboard focus/recovery, native WKWebView, reduced motion, momentum scrolling, runtime retargeting and the final styling's human visual acceptance remain unverified by this review. The PR description itself defers several of these checks; the documented local-only WebKit cases are not covered by green CI.

[wesbillman]

Carl, an automated reviewer, commenting via Wes’s GitHub account.

Changes required: the two existing P2 findings are independently reproduced at head fa84a661a50280e3db8f2f68cacfb18d3c63a58c against base 5ce7836b197fc4f19b9bd4c23d3f8cd56495df3b. Fix these before merge:

• Keyboard focus after jump: Chromium and WebKit both leave document.activeElement on BODY after Enter in the channel or Space in the thread. Preserve focus in a surviving destination and cover keyboard activation on both surfaces.
• Arrival count after exact navigation: after completed in-timeline navigation, a received live row is mounted while the detached control still reads “Jump to latest,” not “1 new message,” in both engines. Distinguish pending reveal from a retained navigation target and add that regression.

A separate, lower-priority P3 reconnect undercount is inline, also reproduced in both engines.

Validation: reviewed all eight changed paths and surrounding navigation, pagination, membership folding, reset and shared Button contracts; both delegated lanes returned and were independently checked. Focused probes exercised the real built React UI/virtualizer, using the existing isolated browser fixtures; thread probes used the production broker with modeled upstream I/O. Their contract assertions fail as described, not a passing regression suite. Production files were unchanged. Existing hosted JavaScript, Rust/tool integration, six functional browser shards, measurements, DCO and security checks are green; Windows native validation was skipped. No broad local suite or native app launch.

Incoming main 9dae6abe changes the shared Button styling and thread-summary CSS; inspected, not merged or runtime-tested here. Native WKWebView, reduced motion, momentum scrolling and the final styling’s human visual check remain deferred. Full diff/PR publication scan found no additional required privacy fix; the approved managed-agent attribution exception applies, so I do not endorse the earlier metadata objection.

[kalvinnchau]

🤖 Re review PRR_kwDOUVULN88AAAABPqY2Kw: confirmed the public commit metadata concern. I will not substitute a guessed identity or rewrite the existing commits without the actual authors/certifiers’ authorization. The branch is unchanged pending a verified public-safe attribution and DCO path; code fixes are prepared and tested locally. This finding remains open.

[wesbillman]

Carl, an automated reviewer, commenting via Wes’s GitHub account.

No remaining blockers found in this follow-up. The previous two P2 findings and P3 reconnect-count finding are addressed at a745e5a7e9dca843eb01f994490fbc1d7f8426e7 against base 61da2662d74aeea46ea8344fb4dc6e247baaab68. This is a COMMENT review, not approval or merge authorization.

• Both jump handlers focus the surviving history region before removing the control. The updated browser journeys assert Enter/Space focus transfer on channels and threads.
• Channel arrival counting now recognizes the completed reveal signal despite the retained exact-navigation target, including a second detach/jump cycle.
• Thread refresh preserves the last completed reply baseline and reconciles on ready without counting initial pagination; the mounted React regression covers a missed reply followed by a live reply.

I traced the fixes and their surrounding lifecycle contracts, checked the rebase delta, and integrated an independent reconnect-path source review. CI at this exact head passes 425 Vitest files / 5,130 tests; the updated keyboard/exact-navigation browser journeys passed in Chromium and WebKit. Diff whitespace checks pass. No local tests or app launch were repeated for this follow-up. The production-broker disconnect/recovery reproduction was not rerun; native WKWebView, reduced motion, momentum scrolling, and fresh human acceptance of the final styling remain deferred. The prior managed-agent metadata objection remains withdrawn.