fix(hook): remap aux buttons that arrive without HID attribution on macOS 26

[tims2002]

Problem

On macOS 26, all OS-hook button remaps and hold+swipe gestures silently stop working, while HID++-diverted single-action bindings keep working. This makes the breakage look device-specific (it isn't) — observed with an ERGO M575 over both Bluetooth LE and a Unifying receiver, and it should affect every device whose Middle/Back/Forward go through the CGEventTap hook.

Root cause

macOS 26 delivers other-mouse (aux button) CGEvents without a backing IOHIDEvent — CGEventCopyIOHIDEvent returns null — so event_sender_id yields None and button_source_may_remap fails closed, passing every Middle/Back/Forward press through untouched.

Verified with a listen-only diagnostic tap on macOS 26.6.2 (arm64), pressing buttons on the same physical device:

event type=1  button=0 hidEvent=yes senderID=0x10001e970 Product=USB Receiver VendorID=1133   ← left click: attribution intact
event type=25 button=4 hidEvent=NULL                                                          ← aux button: no HID backing
event type=26 button=4 hidEvent=NULL

Left clicks still attribute correctly, so the regression is specific to aux-button events in the macOS 26 input pipeline.

Fix

Allow an unattributed event through the gate only for the OS-hook buttons (Middle/Back/Forward). The fail-closed rule exists so the built-in trackpad can never be swallowed — but trackpads never emit those buttons, so that rationale doesn't apply to them. Attributed events keep the existing Logitech/non-trackpad check, and everything else keeps failing closed.

Testing

• macOS 26.6.2 (arm64), ERGO M575 via Bluetooth LE and via Unifying receiver (slot 2)
• Click remaps and all five per-direction gesture slots on Back/Forward work again; Middle unaffected as a single-action HID++ divert and works as a gesture owner too
• Trackpad input unaffected (left/right/primary pass-through preserved)

Possible follow-up (not in this PR)

The M575 (and likely other non-MX devices) reports divertable, raw-xy on CIDs 0x52/0x53/0x56 (openlogi diag controls), so per-direction gestures for these buttons could be delivered over HID++ raw-XY diversion like the MX gesture button — transport-independent and immune to CGEvent pipeline changes like this one. Happy to file a separate issue if useful.

🤖 Generated with Claude Code

[greptile-apps]

Greptile Summary

The PR restores macOS 26 remapping for auxiliary mouse events that arrive without HID attribution while retaining source checks for attributed events and fail-closed behavior for primary buttons.

• Passes the button ID into the source-remapping gate.
• Allows unattributed Middle, Back, and Forward events on macOS.
• Preserves the existing Logitech/non-trackpad validation for attributed events.

Confidence Score: 5/5

The PR appears safe to merge, with no concrete blocking or independently actionable issue identified.

The relaxed path applies only to Middle, Back, and Forward events lacking device attribution; attributed events retain the existing source validation, and primary buttons remain excluded before remapping.

Important Files Changed

Filename	Overview
crates/openlogi-agent-core/src/hook_runtime.rs	Narrows the macOS missing-attribution exception to the three OS-hook auxiliary buttons without exposing primary-button events to suppression.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Mouse button event] --> B{Middle, Back, or Forward?}
  B -- No --> P[Pass through]
  B -- Yes --> C{Device attributed?}
  C -- Yes --> D{Logitech non-trackpad?}
  D -- No --> P
  D -- Yes --> R[Resolve configured remap]
  C -- No --> E{Running on macOS?}
  E -- No --> R
  E -- Yes --> R

Loading

_{Reviews (1): Last reviewed commit: "fix(hook): remap aux buttons that arrive..." | Re-trigger Greptile}