Approved source pins: Thaw 73d275284a34a95c02416086f891b4fb4179bbab; Ice d567f1ecac68936a090d6b47d6f0b882c489c6f4.
Tasks: (1) shared identity/layout/deadline policies and upstream ports; (2) authenticated discovery/capture worker; (3) serialized native move/click/restore engine; (4) UI, migration and lifecycle integration; (5) fixtures, packaging, docs, GitHub and verification.
The v1.5.0 release was published on 2026-10-22 after the user waived the outstanding physical validation matrix. The native fixture evidence and checks that were not run are distinguished in the final release report; the incomplete matrix is not a passing result.
-
Source archives fetched at exact approved commits; current workspace clean at 6757f6f.
-
Shared policies, migration, exact/unique identity matching, native-response checks, layout planner and upstream circuit breaker implemented. RED tests observed before implementation; policy suite passed.
-
Both helpers authenticate current packaged ad-hoc designated requirements. Discovery initially found no windows: upstream JoinExistingSession/NSRunLoop configuration was essential and is now retained. A nonempty live-window assertion and mismatched-signature rejection now pass.
-
Discovery and capture are separate embedded processes. Capture supports on-screen SCK, offscreen dynamic SkyLight, actual pixel scale, byte/count limits and attempt-based recycling; only the panel requests up to 2 batches/s.
-
Store/UI/lifecycle switched to native selection. Normal quit no longer clears requested visibility. Native event transport and temporary-menu restoration are implemented but still need physical permission-enabled validation.
-
Native fixture, provenance, GPL license and release-gate documentation added. No release published.
-
Fresh review found unresolved-row blocking, notch visibility, premature panel dismissal, stale restoration, custom-popover rejection and discovery starvation. Fixed these paths before installation. Divider recovery now moves only our own divider past unresolved/protected anchors; those anchors are never activation targets. Discovery retains window/display association until display geometry changes, and freshly scanned owners lose stale AX entries.
-
Installed candidate grants required removing/readding the exact ad-hoc bundle in both permission lists; granted helper discovery resolves QQ/Teams/Mac Duo/Snapzy/WeChat and the four fixture controls. Captured fixture frames returned, but this does not yet prove recognizable panel presentation or native activation.
-
Live hide failed: Control Center blocked OverflowBar’s own empty status hosts. One host restart and scoped system visibility refresh did not recover them. The strengthened real-control-window test failed before the startup-order repair. SwiftUI panel initialization now precedes host registration and images are assigned immediately; notch height validation allows the observed one-point vertical rounding. Physical retest pending.
-
Sparkle key is intentionally only in the existing GitHub Actions secret and unavailable locally; local archive signing is blocked. At this validation stage, publication remained withheld.
-
Startup-order and host recreation alone did not clear the installed Control Center block. Investigation found the directly matching CodexBar issue #1440: a stale foreign trackedApplications/menuItemLocations entry can override an enabled app. macOS denied reading the protected file; a backup-first scoped repair procedure was tested historically with temporary Python Full Disk Access. That procedure is not shipped or supported by this candidate.
-
The last control-recovery test failed on helper timeout, not a control assertion. Worker discovery now coalesces concurrent requests to avoid redundant queued two-second scans. No claim of physical hiding success yet.
-
Historical validation found and removed one stale foreign Control Center mapping after retaining a backup; both native controls subsequently appeared. That temporary repair procedure and its test script are not included in this candidate and are not supported user tools.
-
Hosted status windows include 14 points of horizontal padding around their AX buttons. Discovery now matches unique centers, as upstream does, instead of rejecting the different widths. Correctly attributed fixture and application identities resolve after refreshing grants for the installed signature.
-
Actual hiding on macOS 26.4 is verified: the three selected fixture windows have negative offscreen coordinates while normal icons remain accessible. Relaunch restores desired hiding. The second-row visual and native-activation gates remain pending.
-
Hosted XCTest now suppresses automatic application startup work; the complete suite passes with one explicitly skipped Accessibility-dependent physical control test. Panel keyboard/Escape coverage also passes. The panel now follows upstream persistence across key resignation and permits keyboard focus for explicit opening only.
-
Rebuilt installed panel is keyboard accessible; Escape and genuine fixture captures are verified through Computer Use. First offscreen native-menu activation failed its move verification and safely revealed the group. Comparison with pinned upstream found source-anchored/fidelity releases must target the source, and native clicks must omit the move-only 0x33 field and use click-up state zero. A focused test failed on the old click fields and passes after correction. Failed operations now suspend automatic reapplication until explicit hiding, preserving the saved desired state. Live activation retest pending.
-
Physical macOS 26.4 retest: advertised native menu opened from a revealed second row (fixture nativeMenus 2 to 3), then from an actually hidden group after temporary divider reveal (3 to 4). The group returned offscreen after native-menu closure. Direct dragging from the expanded divider remains unreliable on this host, so temporary reveal precedes fresh resolution and remaining notch moves.
-
A clean fixture relaunch while hidden exposed fresh AX geometry remaining unresolved until reveal; lifecycle observation now reveals before settling discovery. The fresh AppKit popover did not respond to native events, so advertised freshly resolved AXPress is preferred for left activation, with native events otherwise. A helper timeout cannot cause a second uncertain click. Live popover/toggle retest pending.
-
Review recovery fixes: single active native interface/restoration; known status-window exclusion allows one-row menus; restoration uncertainty latches revealed state and suspends automatic management; automatic control repairs count toward the shared move circuit breaker.
-
Latest complete XCTest result: 47 passed, zero failed, one explicitly skipped permission-dependent own-control test on macOS 26.4. Debug, universal Release, nested ad-hoc signatures, documentation links and Homebrew temporary-tap validation passed.
-
Installed reviewed candidate hid all four native fixtures at real negative coordinates. Hidden popover activation incremented its real counter 0 to 1 and created a native owner window; Escape closed it and the native group returned offscreen. Hidden toggle activation incremented 0 to 1 and fresh AX title became OB-T1. Observable changes remain revealed until explicit hiding; their success feedback is now a quiet status instead of an alert.
-
At this validation stage, physical macOS 15, real Electron tray coverage, 100 full cycles and an instrumented two-hour soak remained outstanding; the release report records them as untested.
-
Repeated relaunch exposed a forced checkbox-order swap of two already grouped fixtures across the notch. Hiding now preserves observed relative order and only changes section membership; regression verifies no moves for a correctly grouped selection with different checkbox ordering. Discovery carries actual AX button bounds so invisible hosted padding cannot falsely demand a notch move; outside-host AX bounds are rejected by the visibility policy. Full tests pass; repeat installed validation follows.
-
Oct 11: approval-review quota failure cleared after fresh usage verification; latest candidate installed and both existing grants refreshed. Relaunch with new system indicators exposed remaining move failures. Direct comparison with pinned Thaw found the adapter's drag relay direction reversed: it must enter at the owner PID, post to the session tap, then confirm the scrombled event back at the owner PID. The transport now uses that fixed route and recomputes release geometry after the press reflows the bar. Automated suite passes; installed native move regression retest pending.