feat(web): make the terminal close confirmation optional

[pandec]

Upstream's terminal close confirmation (pingdotgg#7592) fires on every individual close with no way to turn it off, so closing a terminal always costs a dialog.

Adds a confirmTerminalClose client setting that defaults to true, keeping upstream behavior for anyone who doesn't touch it, plus a switch in Settings → General next to the archive and delete confirmations. The gate sits inside confirmTerminalClose() and reads the non-hook getClientSettings() snapshot, so all four call sites (drawer button, panel button, terminal.close keybinding, tab-strip close) stay untouched. Before client settings hydrate the snapshot holds the schema default, so the pre-hydration answer is still to ask.

Bulk tab closes and auto-exit cleanup never went through this path and still don't.

Verified with vp check, vp run typecheck, and focused tests for the confirm helper, settings search, client settings, and desktop client settings. The new "setting off" test was mutation-checked: removing the gate fails it.

Model: claude-opus-5, harness: Claude Code.

[pandec]

Review pass (gpt-5.6-sol) findings and dispositions:

Fixed — pre-hydration race in apps/web/src/lib/terminalCloseConfirm.ts. The snapshot holds the schema default until client settings hydrate, so a close early in a cold start prompted even with the setting off. Now awaits ensureClientSettingsHydrated() first, matching resolveBrowserDefaults. Covered by a new test, mutation-checked.

Declined — a docs/user/ page for the switch. None of the sibling confirmations (archive, delete, quit) are in docs/user/, upstream's dialog itself isn't documented there, and there is no terminals page to extend. Adding one for a single toggle is wider than this change warrants; the README fork section carries it.

Declined — extracting a shared boolean-setting row or a descriptor list for the three confirmation switches. Real duplication, but it would refactor upstream-owned code in SettingsPanels.tsx and raise merge cost against the fork policy.

Declined — rebuilding DesktopClientSettings.test.ts from DEFAULT_CLIENT_SETTINGS. Same reasoning: an unrelated upstream fixture.

Not this PR — cross-tab snapshot staleness and hydration overwriting an in-flight switch update. Both are properties of the shared client-settings store and affect every setting equally.

[pandec]

Upstream's terminal close confirmation (pingdotgg#7592) fires on every individual close with no way to turn it off, so closing a terminal always costs a dialog.

Adds a confirmTerminalClose client setting that defaults to true, keeping upstream behavior for anyone who doesn't touch it, plus a switch in Settings → General next to the archive and delete confirmations. The gate sits inside confirmTerminalClose() and reads the non-hook getClientSettings() snapshot, so all four call sites (drawer button, panel button, terminal.close keybinding, tab-strip close) stay untouched. Before client settings hydrate the snapshot holds the schema default, so the pre-hydration answer is still to ask.

Bulk tab closes and auto-exit cleanup never went through this path and still don't.

Verified with vp check, vp run typecheck, and focused tests for the confirm helper, settings search, client settings, and desktop client settings. The new "setting off" test was mutation-checked: removing the gate fails it.

Model: claude-opus-5, harness: Claude Code.

[pandec]

Review pass (gpt-5.6-sol) findings and dispositions:

Fixed — pre-hydration race in apps/web/src/lib/terminalCloseConfirm.ts. The snapshot holds the schema default until client settings hydrate, so a close early in a cold start prompted even with the setting off. Now awaits ensureClientSettingsHydrated() first, matching resolveBrowserDefaults. Covered by a new test, mutation-checked.

Declined — a docs/user/ page for the switch. None of the sibling confirmations (archive, delete, quit) are in docs/user/, upstream's dialog itself isn't documented there, and there is no terminals page to extend. Adding one for a single toggle is wider than this change warrants; the README fork section carries it.

Declined — extracting a shared boolean-setting row or a descriptor list for the three confirmation switches. Real duplication, but it would refactor upstream-owned code in SettingsPanels.tsx and raise merge cost against the fork policy.

Declined — rebuilding DesktopClientSettings.test.ts from DEFAULT_CLIENT_SETTINGS. Same reasoning: an unrelated upstream fixture.

Not this PR — cross-tab snapshot staleness and hydration overwriting an in-flight switch update. Both are properties of the shared client-settings store and affect every setting equally.

[pandec]

Upstream's terminal close confirmation (pingdotgg#7592) fires on every individual close with no way to turn it off, so closing a terminal always costs a dialog.

Adds a confirmTerminalClose client setting that defaults to true, keeping upstream behavior for anyone who doesn't touch it, plus a switch in Settings → General next to the archive and delete confirmations. The gate sits inside confirmTerminalClose() and reads the non-hook getClientSettings() snapshot, so all four call sites (drawer button, panel button, terminal.close keybinding, tab-strip close) stay untouched. Before client settings hydrate the snapshot holds the schema default, so the pre-hydration answer is still to ask.

Bulk tab closes and auto-exit cleanup never went through this path and still don't.

Verified with vp check, vp run typecheck, and focused tests for the confirm helper, settings search, client settings, and desktop client settings. The new "setting off" test was mutation-checked: removing the gate fails it.

Model: claude-opus-5, harness: Claude Code.

[pandec]

Review pass (gpt-5.6-sol) findings and dispositions:

Fixed — pre-hydration race in apps/web/src/lib/terminalCloseConfirm.ts. The snapshot holds the schema default until client settings hydrate, so a close early in a cold start prompted even with the setting off. Now awaits ensureClientSettingsHydrated() first, matching resolveBrowserDefaults. Covered by a new test, mutation-checked.

Declined — a docs/user/ page for the switch. None of the sibling confirmations (archive, delete, quit) are in docs/user/, upstream's dialog itself isn't documented there, and there is no terminals page to extend. Adding one for a single toggle is wider than this change warrants; the README fork section carries it.

Declined — extracting a shared boolean-setting row or a descriptor list for the three confirmation switches. Real duplication, but it would refactor upstream-owned code in SettingsPanels.tsx and raise merge cost against the fork policy.

Declined — rebuilding DesktopClientSettings.test.ts from DEFAULT_CLIENT_SETTINGS. Same reasoning: an unrelated upstream fixture.

Not this PR — cross-tab snapshot staleness and hydration overwriting an in-flight switch update. Both are properties of the shared client-settings store and affect every setting equally.

[pandec]

Upstream's terminal close confirmation (pingdotgg#7592) fires on every individual close with no way to turn it off, so closing a terminal always costs a dialog.

Adds a confirmTerminalClose client setting that defaults to true, keeping upstream behavior for anyone who doesn't touch it, plus a switch in Settings → General next to the archive and delete confirmations. The gate sits inside confirmTerminalClose() and reads the non-hook getClientSettings() snapshot, so all four call sites (drawer button, panel button, terminal.close keybinding, tab-strip close) stay untouched. Before client settings hydrate the snapshot holds the schema default, so the pre-hydration answer is still to ask.

Bulk tab closes and auto-exit cleanup never went through this path and still don't.

Verified with vp check, vp run typecheck, and focused tests for the confirm helper, settings search, client settings, and desktop client settings. The new "setting off" test was mutation-checked: removing the gate fails it.

Model: claude-opus-5, harness: Claude Code.

[pandec]

Review pass (gpt-5.6-sol) findings and dispositions:

Fixed — pre-hydration race in apps/web/src/lib/terminalCloseConfirm.ts. The snapshot holds the schema default until client settings hydrate, so a close early in a cold start prompted even with the setting off. Now awaits ensureClientSettingsHydrated() first, matching resolveBrowserDefaults. Covered by a new test, mutation-checked.

Declined — a docs/user/ page for the switch. None of the sibling confirmations (archive, delete, quit) are in docs/user/, upstream's dialog itself isn't documented there, and there is no terminals page to extend. Adding one for a single toggle is wider than this change warrants; the README fork section carries it.

Declined — extracting a shared boolean-setting row or a descriptor list for the three confirmation switches. Real duplication, but it would refactor upstream-owned code in SettingsPanels.tsx and raise merge cost against the fork policy.

Declined — rebuilding DesktopClientSettings.test.ts from DEFAULT_CLIENT_SETTINGS. Same reasoning: an unrelated upstream fixture.

Not this PR — cross-tab snapshot staleness and hydration overwriting an in-flight switch update. Both are properties of the shared client-settings store and affect every setting equally.

[pandec]

Upstream's terminal close confirmation (pingdotgg#7592) fires on every individual close with no way to turn it off, so closing a terminal always costs a dialog.

Adds a confirmTerminalClose client setting that defaults to true, keeping upstream behavior for anyone who doesn't touch it, plus a switch in Settings → General next to the archive and delete confirmations. The gate sits inside confirmTerminalClose() and reads the non-hook getClientSettings() snapshot, so all four call sites (drawer button, panel button, terminal.close keybinding, tab-strip close) stay untouched. Before client settings hydrate the snapshot holds the schema default, so the pre-hydration answer is still to ask.

Bulk tab closes and auto-exit cleanup never went through this path and still don't.

Verified with vp check, vp run typecheck, and focused tests for the confirm helper, settings search, client settings, and desktop client settings. The new "setting off" test was mutation-checked: removing the gate fails it.

Model: claude-opus-5, harness: Claude Code.

[pandec]

Review pass (gpt-5.6-sol) findings and dispositions:

Fixed — pre-hydration race in apps/web/src/lib/terminalCloseConfirm.ts. The snapshot holds the schema default until client settings hydrate, so a close early in a cold start prompted even with the setting off. Now awaits ensureClientSettingsHydrated() first, matching resolveBrowserDefaults. Covered by a new test, mutation-checked.

Declined — a docs/user/ page for the switch. None of the sibling confirmations (archive, delete, quit) are in docs/user/, upstream's dialog itself isn't documented there, and there is no terminals page to extend. Adding one for a single toggle is wider than this change warrants; the README fork section carries it.

Declined — extracting a shared boolean-setting row or a descriptor list for the three confirmation switches. Real duplication, but it would refactor upstream-owned code in SettingsPanels.tsx and raise merge cost against the fork policy.

Declined — rebuilding DesktopClientSettings.test.ts from DEFAULT_CLIENT_SETTINGS. Same reasoning: an unrelated upstream fixture.

Not this PR — cross-tab snapshot staleness and hydration overwriting an in-flight switch update. Both are properties of the shared client-settings store and affect every setting equally.

[pandec]

Upstream's terminal close confirmation (pingdotgg#7592) fires on every individual close with no way to turn it off, so closing a terminal always costs a dialog.

Adds a confirmTerminalClose client setting that defaults to true, keeping upstream behavior for anyone who doesn't touch it, plus a switch in Settings → General next to the archive and delete confirmations. The gate sits inside confirmTerminalClose() and reads the non-hook getClientSettings() snapshot, so all four call sites (drawer button, panel button, terminal.close keybinding, tab-strip close) stay untouched. Before client settings hydrate the snapshot holds the schema default, so the pre-hydration answer is still to ask.

Bulk tab closes and auto-exit cleanup never went through this path and still don't.

Verified with vp check, vp run typecheck, and focused tests for the confirm helper, settings search, client settings, and desktop client settings. The new "setting off" test was mutation-checked: removing the gate fails it.

Model: claude-opus-5, harness: Claude Code.

[pandec]

Review pass (gpt-5.6-sol) findings and dispositions:

Fixed — pre-hydration race in apps/web/src/lib/terminalCloseConfirm.ts. The snapshot holds the schema default until client settings hydrate, so a close early in a cold start prompted even with the setting off. Now awaits ensureClientSettingsHydrated() first, matching resolveBrowserDefaults. Covered by a new test, mutation-checked.

Declined — a docs/user/ page for the switch. None of the sibling confirmations (archive, delete, quit) are in docs/user/, upstream's dialog itself isn't documented there, and there is no terminals page to extend. Adding one for a single toggle is wider than this change warrants; the README fork section carries it.

Declined — extracting a shared boolean-setting row or a descriptor list for the three confirmation switches. Real duplication, but it would refactor upstream-owned code in SettingsPanels.tsx and raise merge cost against the fork policy.

Declined — rebuilding DesktopClientSettings.test.ts from DEFAULT_CLIENT_SETTINGS. Same reasoning: an unrelated upstream fixture.

Not this PR — cross-tab snapshot staleness and hydration overwriting an in-flight switch update. Both are properties of the shared client-settings store and affect every setting equally.

[pandec]

Upstream's terminal close confirmation (pingdotgg#7592) fires on every individual close with no way to turn it off, so closing a terminal always costs a dialog.

Adds a confirmTerminalClose client setting that defaults to true, keeping upstream behavior for anyone who doesn't touch it, plus a switch in Settings → General next to the archive and delete confirmations. The gate sits inside confirmTerminalClose() and reads the non-hook getClientSettings() snapshot, so all four call sites (drawer button, panel button, terminal.close keybinding, tab-strip close) stay untouched. Before client settings hydrate the snapshot holds the schema default, so the pre-hydration answer is still to ask.

Bulk tab closes and auto-exit cleanup never went through this path and still don't.

Verified with vp check, vp run typecheck, and focused tests for the confirm helper, settings search, client settings, and desktop client settings. The new "setting off" test was mutation-checked: removing the gate fails it.

Model: claude-opus-5, harness: Claude Code.

[pandec]

Review pass (gpt-5.6-sol) findings and dispositions:

Fixed — pre-hydration race in apps/web/src/lib/terminalCloseConfirm.ts. The snapshot holds the schema default until client settings hydrate, so a close early in a cold start prompted even with the setting off. Now awaits ensureClientSettingsHydrated() first, matching resolveBrowserDefaults. Covered by a new test, mutation-checked.

Declined — a docs/user/ page for the switch. None of the sibling confirmations (archive, delete, quit) are in docs/user/, upstream's dialog itself isn't documented there, and there is no terminals page to extend. Adding one for a single toggle is wider than this change warrants; the README fork section carries it.

Declined — extracting a shared boolean-setting row or a descriptor list for the three confirmation switches. Real duplication, but it would refactor upstream-owned code in SettingsPanels.tsx and raise merge cost against the fork policy.

Declined — rebuilding DesktopClientSettings.test.ts from DEFAULT_CLIENT_SETTINGS. Same reasoning: an unrelated upstream fixture.

Not this PR — cross-tab snapshot staleness and hydration overwriting an in-flight switch update. Both are properties of the shared client-settings store and affect every setting equally.

[pandec]

Upstream's terminal close confirmation (pingdotgg#7592) fires on every individual close with no way to turn it off, so closing a terminal always costs a dialog.

Adds a confirmTerminalClose client setting that defaults to true, keeping upstream behavior for anyone who doesn't touch it, plus a switch in Settings → General next to the archive and delete confirmations. The gate sits inside confirmTerminalClose() and reads the non-hook getClientSettings() snapshot, so all four call sites (drawer button, panel button, terminal.close keybinding, tab-strip close) stay untouched. Before client settings hydrate the snapshot holds the schema default, so the pre-hydration answer is still to ask.

Bulk tab closes and auto-exit cleanup never went through this path and still don't.

Verified with vp check, vp run typecheck, and focused tests for the confirm helper, settings search, client settings, and desktop client settings. The new "setting off" test was mutation-checked: removing the gate fails it.

Model: claude-opus-5, harness: Claude Code.

[pandec]

Review pass (gpt-5.6-sol) findings and dispositions:

Fixed — pre-hydration race in apps/web/src/lib/terminalCloseConfirm.ts. The snapshot holds the schema default until client settings hydrate, so a close early in a cold start prompted even with the setting off. Now awaits ensureClientSettingsHydrated() first, matching resolveBrowserDefaults. Covered by a new test, mutation-checked.

Declined — a docs/user/ page for the switch. None of the sibling confirmations (archive, delete, quit) are in docs/user/, upstream's dialog itself isn't documented there, and there is no terminals page to extend. Adding one for a single toggle is wider than this change warrants; the README fork section carries it.

Declined — extracting a shared boolean-setting row or a descriptor list for the three confirmation switches. Real duplication, but it would refactor upstream-owned code in SettingsPanels.tsx and raise merge cost against the fork policy.

Declined — rebuilding DesktopClientSettings.test.ts from DEFAULT_CLIENT_SETTINGS. Same reasoning: an unrelated upstream fixture.

Not this PR — cross-tab snapshot staleness and hydration overwriting an in-flight switch update. Both are properties of the shared client-settings store and affect every setting equally.