fix: pr-bug-scan validated finding from #4658

[buf0-bot]

Automated fix-PR from pr-bug-scan for parent #4658.

Fixer status: FIXED_WITH_CODE_PROOF
Summary: Gate the three unread-clear calls in onTerminalKeyDown so modifier-only/repeat/copy-chord keydowns no longer dismiss sibling-pane attention.
Blocked proof link: src/renderer/src/components/terminal-pane/pty-connection.ts:719 onTerminalKeyDown unconditionally calls deps.clearTerminalTabUnread/clearTerminalPaneUnread/clearWorktreeUnread before any guard
Typecheck: skipped

Proof (from validator)

• C1 — proof_type: code_analysis
  • trigger: User has a tab with split panes A and B in the same worktree; pane B emits BEL or completion attention while hidden, then user types/presses any key (e.g. holds Shift, presses Cmd+C to copy a selection, autorepeats arrow keys) inside pane A.
  • observable: Tab-level and worktree-level attention dots disappear for sibling pane B as soon as the user types into pane A, even when no real character was sent to any terminal (e.g. Cmd+C with selection, Shift held, Escape pressed).
  • path:
    • src/renderer/src/components/terminal-pane/pty-connection.ts:719 onTerminalKeyDown unconditionally calls deps.clearTerminalTabUnread(deps.tabId), deps.clearTerminalPaneUnread(cacheKey), deps.clearWorktreeUnread(deps.worktreeId) before any guard
    • src/renderer/src/components/terminal-pane/pty-connection.ts:744 listener is registered on pane.terminal.element with capture:true so every key — including modifier-only, repeat, copy, plain Escape — reaches the handler
    • src/renderer/src/components/terminal-pane/pty-connection.ts:1309 onData (post-diff) intentionally no longer clears unread, so this is the only clear path for keyboard interaction
• C2 — proof_type: code_analysis
  • trigger: Settings: experimentalTerminalAttention=true, notifications.agentTaskComplete=false, notifications.terminalBell=true. An agent emits BEL during the same burst as its working→idle transition (typical Claude/Codex completion).
  • observable: No terminal-bell OS notification is delivered for the BEL that the user explicitly enabled; only the silent attention dot is set.
  • path:
    • src/renderer/src/components/terminal-pane/pty-connection.ts onBell sets pendingTerminalBellNotification=true and schedules a bell OS notification because hasPendingAgentTaskCompleteNotification() short-circuits on isAgentTaskCompleteNotificationEnabled()=false
    • src/renderer/src/components/terminal-pane/pty-connection.ts:1068 scheduleAgentTaskCompleteNotification now runs because syncAgentTaskCompleteTrackingEnabled() returns true (attention is on)
    • src/renderer/src/components/terminal-pane/pty-connection.ts:1090-1099 dispatch() sets pendingTerminalBellNotification=false and calls clearTerminalBellNotificationTimer() before invoking deps.dispatchNotification with suppressOsNotification:true
    • src/renderer/src/components/terminal-pane/use-notification-dispatch.ts:286-288 dispatchTerminalNotification returns early on suppressOsNotification before window.api.notifications.dispatch — no OS notification fires

---

Generated by pr-bug-scan (proof-machine architecture). Human approval required before merge.

[pullfrog]

ℹ️ No critical issues — minor test-coverage observation inline.

Reviewed changes — gates onTerminalKeyDown unread-clearing so that modifier-only key presses, autorepeat, and copy chords (Cmd/Ctrl+C with active selection) no longer dismiss sibling-pane attention dots. This addresses the validated finding from #4658 where typing into pane A would prematurely clear attention for pane B.

• Gate unread clearing in onTerminalKeyDown — moves the three clear*Unread calls after the escape/Ctrl-C intent checks and adds two early-return guards: one for modifier-only/repeat keydowns, one for copy chords with an active selection.

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using DeepSeek Pro (free via Pullfrog for OSS) ｜ 𝕏}

[pullfrog]

ℹ️ The modifier-only/repeat guard and copy-chord guard added here have no test coverage. The existing test at pty-connection.test.ts:4860 (clears tab and worktree unread on real keydown) only exercises the happy path with key='a', no modifiers, repeat=false. Consider adding tests for:

• Modifier-only presses (key='Shift', key='Alt', key='Control', key='Meta') — assert unread is not cleared
• Repeat events (key='a', repeat=true) — assert unread is not cleared
• Copy chord with selection (key='c', ctrlKey=true or metaKey=true, hasSelection()=true) — assert unread is not cleared

Technical details

# Missing test coverage for `onTerminalKeyDown` guard conditions

## Affected sites
- `src/renderer/src/components/terminal-pane/pty-connection.ts:736-752` — new modifier-only/repeat guard and copy-chord guard
- `src/renderer/src/components/terminal-pane/pty-connection.test.ts:4860-4884` — existing happy-path test, no guard-coverage tests

## Required outcome
- Each guard condition path (modifier-only, repeat, copy-chord) should have a dedicated test asserting that `clearTerminalTabUnread`, `clearTerminalPaneUnread`, and `clearWorktreeUnread` are NOT called.

## Suggested approach
- Dispatch `keydown` events with the guard-triggering properties (e.g. `key='Shift'`, `repeat=true`, or `key='c'` + `ctrlKey=true` + mock `hasSelection()`) and assert the three clear functions were not called, mirroring the pattern of the existing test at line 4860.