The History mutation handlers already wait for backend success before changing visible list state, so failure recovery does not need to write back the request-time snapshot. Removing that restoration avoids overwriting a newer refresh result, and delete success now updates from the latest list state instead of the closure captured before the IPC call.
Constraint: PR #343 review identified stale snapshot writes during overlapping refresh and mutation requests.\nRejected: Add action-wide locking | more UI state than needed once failed mutations stop writing stale snapshots.\nConfidence: high\nScope-risk: narrow\nTested: cd openless-all/app && git diff --check && npm run build\nNot-tested: Manual overlapping refresh/delete in the Tauri app.
History actions previously assumed list, clear, and delete IPC calls would resolve. A rejected load could leave the page stuck in loading, while failed mutations gave no user-visible feedback. The page now clears loading in finally, shows retryable inline load errors, and reports clear/delete failures while restoring the prior local selection and items.
Constraint: Issue #310 asks for inline card feedback without changing Rust handlers or introducing a toast system.\nRejected: New shared error component | broader UI surface than needed for this single-page failure path.\nConfidence: high\nScope-risk: narrow\nTested: cd openless-all/app && git diff --check && npm run build\nNot-tested: Live Tauri IPC failure injection.\nRelated: https://github.com/appergb/openless/issues/310
Overview previously treated rejected history and credential IPC calls like empty data, which made backend read failures indistinguishable from a clean first run. The page now keeps separate error flags, shows provider status as unreadable when credential status cannot load, and replaces history-derived metrics/recent rows with retryable inline failure states.
Constraint: Issue #311 asks for per-card inline feedback without changing backend handler signatures or adding global toast UI.\nRejected: Global error boundary or toast | broader UI behavior than needed for the Overview failure path.\nConfidence: high\nScope-risk: narrow\nTested: cd openless-all/app && git diff --check && npm run build\nNot-tested: Live Tauri IPC failure injection.\nRelated: https://github.com/appergb/openless/issues/311
The Selection Ask page now disables its card toggles while a save is in flight, but the nested shortcut recorder still accepted new recordings. That left a small overlap window where a second QA hotkey save could race the current setQaHotkey/savePrefs pair.\n\nThis keeps the recorder default behavior unchanged for other pages by adding an optional disabled prop and only wiring it from Selection Ask during its saving state.\n\nConstraint: Review feedback on PR #338 requested preventing overlapping QA shortcut saves without broad settings-queue refactoring.\nRejected: Global serialization inside ShortcutRecorder | unnecessary for pages that already own their own persistence state.\nConfidence: high\nScope-risk: narrow\nTested: cd openless-all/app && npm run build\nTested: git diff --check\nNot-tested: Manual browser interaction against a slow Tauri IPC response.\nRelated: https://github.com/appergb/openless/issues/315\nRelated: https://github.com/appergb/openless/pull/338
Selection Ask reused the same QA hotkey and history persistence paths as other settings pages, but the page did not surface pending, success, or failure states. A rejected hotkey registration or settings write could therefore leave users believing the feature or history option was enabled when the backend had not accepted the change.\n\nThis keeps the existing persistence contract and mirrors the Translation page status treatment: register QA hotkeys before saving prefs, stop on registration errors, refresh after write failures, and add localized feedback for all supported UI languages.\n\nConstraint: Issue #315 requests minimal UI feedback and rollback behavior without refactoring the shared settings queue.\nRejected: Refactor HotkeySettingsContext persistence semantics | broader cross-page behavior change than needed for this bug.\nConfidence: high\nScope-risk: narrow\nTested: cd openless-all/app && npm run build\nTested: git diff --check\nNot-tested: Manual Tauri hotkey conflict flow on macOS/Windows.\nRelated: https://github.com/appergb/openless/issues/315
Style page writes previously updated local UI state before awaiting IPC, so rejected handlers left the page showing values that were never persisted. The change keeps the optimistic update, but centralizes rollback to the previous preferences and surfaces the backend error inline near the failing control.
Constraint: Issue #313 asks for minimal frontend-only handling; backend handlers remain unchanged.\nRejected: Global toast framework | outside the issue scope and heavier than an inline page error.\nConfidence: high\nScope-risk: narrow\nTested: cd openless-all/app && npx tsx src/lib/stylePrefs.test.ts\nTested: cd openless-all/app && npm run build\nTested: git diff --check\nRelated: https://github.com/appergb/openless/issues/313
Translation settings were optimistically updated without awaiting or surfacing write failures, so users could see enabled language and shortcut states that were never persisted. This keeps the fix inside the Translation page: all three save paths now show saving/saved/failed feedback, catch failed writes, and refresh preferences after failed preference persistence to roll back local state.
Constraint: Follow the new beta-first PR workflow from #327.
Rejected: Refactor HotkeySettingsContext queue behavior | issue explicitly excludes queue redesign.
Confidence: high
Scope-risk: narrow
Directive: Keep Translation page save feedback local unless SelectionAsk receives the same treatment in its own issue.
Tested: npm run build
Tested: git diff --check
Related: #314
Adds Casks/openless.rb + update-cask.yml workflow + README install docs. Closes#254#5.
Note: Formula/openless.rb known to fail in Homebrew sandbox (per author); will be removed in follow-up. Cask version will be bumped to current 1.2.23 in same follow-up.
Issue #323 asks for easier switching to more OpenAI-compatible LLM providers instead of forcing users through a single custom slot. The settings page already stores credentials per active LLM provider id, so the narrow fix is to add practical presets and make provider switching seed the model field only when it is empty.
Constraint: Keep provider storage in the existing system credential vault schema
Rejected: Build named custom profile management | larger UI and persistence surface than needed for the first issue slice
Confidence: medium
Scope-risk: narrow
Directive: New LLM preset ids rely on the generic active LLM credential routing; do not special-case them in backend code unless the provider stops being OpenAI-compatible
Tested: npm run build
Tested: git diff --check
Related: #323
The release matrix had kept Intel macOS disabled because the old macos-13 pool was deprecated and repeatedly stalled releases. GitHub now exposes an explicit Intel runner label, so the release lane can be restored without returning to the unstable macos-13 runner.
Constraint: Intel Mac release artifacts must be produced separately from Apple Silicon artifacts
Rejected: Re-enable macos-13 | that runner was explicitly disabled due to deprecation and queue instability
Confidence: high
Scope-risk: narrow
Directive: Keep macOS updater artifact names architecture-qualified when multiple darwin lanes are active
Tested: git diff --check
Tested: Parsed release-tauri.yml matrix and verified macOS arm64/x86_64 lanes plus arch-specific updater paths
Not-tested: Full GitHub Actions release run
