clear_agent_cache falls back to renaming the agent cache directory
to <cache_dir>/.trash/<agent_id>-<nanos>-<counter>/ when
remove_dir_all fails — typically because a running <cmd>.exe or a
Defender scan holds the file. NTFS rename succeeds on directories
whose children are locked, so Upgrade and Uninstall no longer
surface ERROR_ACCESS_DENIED on Windows.
A detached OS thread sweeps .trash/ at startup with all errors
swallowed, panics caught, and no shared state — cannot block app
startup or leak threads. Still-locked entries are left for the
next launch.
Plain `npm install -g <pkg>@<version>` over an existing install does not
reliably re-resolve platform-specific optionalDependencies, leaving the
native CLI binary missing or stale (e.g. 'Native CLI binary for darwin-x64
not found' after the claude-agent-sdk upgrade). The Upgrade button now
runs npm uninstall -g first as a best-effort step, forcing npm to rebuild
the dependency graph from scratch on the subsequent install. If the clean
upgrade fails midway, the DB and the Settings UI resync to the actual
on-disk state instead of showing a phantom version.
- fix(chat-input): raise toolbar collapse threshold to 34rem so selectors stay visible at wider input widths.
- fix(acp): prevent Windows terminal hangs caused by orphaned pipes and .cmd shims.
- fix(acp): fix context loss when resuming a historical conversation by deferring auto-connect until the session id resolves.
- fix(sidebar): clarify empty state when completed conversations are hidden.
- chore(acp): bump claude-code to 0.31.0, codex to 0.12.0, opencode to 1.14.23, gemini to 0.39.1.
-----------------------------
# 发布版本 0.10.3
- 修复(聊天输入)：将工具栏折叠阈值提升到 34rem，更宽的输入区仍能保留选择器；
- 修复(ACP)：修复 Windows 下因管道残留与 .cmd shim 导致的终端挂起；
- 修复(ACP)：修复在历史会话中继续对话时偶发的上下文丢失问题，自动连接会等待历史会话 id 解析完成后再发起；
- 修复(侧边栏)：在隐藏已完成会话时显示更清晰的空态提示；
- 杂项(ACP)：升级 claude-code 至 0.31.0、codex 至 0.12.0、opencode 至 1.14.23、gemini 至 0.39.1。
On Windows the expert enable-status check compared `fs::read_link` output
against the central path, which silently failed for junctions: the link
state fell back to Broken/LinkedElsewhere, so enabled experts kept
reappearing as disabled and disappeared from the message-input expert
list. Switch the link verification to canonicalize both sides (which
transparently follows junctions and symlinks) and add a case-insensitive
path comparison on Windows. Also prefer `junction::get_target` for the
displayed target path.
Large raw_output snapshots from tool_call_update notifications caused
O(N²) traffic through the event pipeline, multi-GB transient allocations
and WKWebView crashes. The fix turns cumulative snapshots into bounded
incremental deltas and removes redundant payload copies.
- Add ToolCallOutputCache keyed by tool_call_id with a 8KB tail
fingerprint and total length. Detects cumulative extensions by
matching the cached tail at the expected offset in the incoming
snapshot, so it works even when the full output grows into the MB
range. Emits suffix deltas with raw_output_append=true; falls back
to a truncated replacement when content diverges.
- Cap any single emitted raw_output chunk at 64KB (MAX_SINGLE_EMIT_BYTES)
with a UTF-8 char-boundary-safe tail and ANSI-sequence-safe trimming.
Apply the same cap to emit_terminal_output_update.
- Bound the cache at 256 entries with FIFO eviction, and clear entries
when the tool call reaches completed / failed / cancelled / error.
- Seed the cache via a dedicated seed() method on SessionUpdate::ToolCall
so the initial event never emits an accidental append.
- Share emit payloads as Arc<serde_json::Value> across broadcast
receivers and skip the Tauri-side clone: serialize once and hand the
same Arc to both the webview emit and the WebSocket broadcaster.
- Add 14 unit tests covering boundary cases: identity, prefix extension
past the cached tail, divergence, oversized deltas, multibyte UTF-8
truncation, final-status cleanup, FIFO eviction, seed semantics, and
ANSI-safe trimming.
Scan ~/.codex/skills/.system so Codex CLI's bundled skills (imagegen,
skill-creator, etc.) appear in the $ autocomplete and Skills settings.
Mark them read_only on the API; the save/delete handlers refuse writes
to that path, and the UI shows a system badge with a tooltip and
disables edit/delete/save for those entries.
Adds a new "Quick Messages" settings page below Experts for managing reusable title/content snippets, backed by SQLite via SeaORM and exposed through both Tauri commands and the Axum web router. The list supports drag-to-reorder using the same motion/react Reorder pattern as the agent list, with translations provided across all 10 supported locales.
- Add `color` column to folder table with migration backfill and hash-based assignment on folder creation
- Expose `update_folder_color` via Tauri command and `/update_folder_color` HTTP route
- Render a color swatch before each folder name in the sidebar header; offer a 10-color palette (9 hues plus a theme-aware foreground sentinel) through the folder context menu
- Show the folder header "new conversation" button only on hover
- Drop the expanded-state tint on folder name and count badge; use a fixed neutral rail color for conversation items
Remove the parent-branch button shown beside the branch selector along
with its supporting API, command, service, model, and i18n strings.
Add migration m20260423 to drop the now-unused folder.parent_branch
column.
Deep directories (beyond the workspace snapshot's depth limit) relied on
a lazy-loaded override cache that was never invalidated, so agent-created
files, in-app deletes / renames / rollbacks, and external changes inside
expanded deep folders stayed invisible until the folder was reopened.
Propagate watcher `changed_paths` through the delta envelope and fire a
Meta event whenever FS activity doesn't alter the tree/git snapshots, so
the frontend can surgically invalidate affected cache entries and
re-fetch. Manual refresh (Reload from disk) clears the cache and
re-hydrates still-expanded deep dirs through the same path. Replayed
deltas after reconnect are forwarded to the same listeners.
Also split the combined workspace-state effect into tree / git / status
slices so unrelated state transitions (e.g. the 'resyncing' flip during
a refresh) no longer rebuild the entire node tree and cause a flash.
Replace the legacy folder + welcome routes with a unified /workspace route
that hosts all folders, conversations, tabs, and terminals in one window.
- Persist opened tabs to the database (opened_tabs entity + migration)
so tab layout survives restarts and deep-link bootstrap restores state
- Replace FolderContext shim with AppWorkspaceProvider, ActiveFolderProvider,
and TabProvider; expose both opened (folders) and full DB (allFolders)
listings via list_all_folder_details
- Return conversations across all non-deleted folders from list_all when
no folder filter is given, so the sidebar can show every folder's history
- Add ConversationContextBar above the chat input with folder picker
(auto-opens unopened folders on select), branch picker, and commit /
push / merge / stash entries to restore BranchDropdown functionality
- Rework sidebar with stats header, search, flat / folder-grouped view
modes (localStorage-persisted), reveal-in-sidebar event subscriber,
and per-folder context menu (focus, close tabs, remove from workspace);
indent conversations under folder headers in grouped mode
- Gate terminal creation on active folder and show folder context
- Remove deprecated BranchDropdown, FolderNameDropdown, welcome route,
and per-folder window commands
- Localize all new strings across 10 locales
Previously LANG/LC_ALL were only set on Windows, so git stderr was
localized on macOS/Linux and substring checks for English messages
(e.g. 'unknown revision or path not in the working tree') silently
failed — empty-repo/missing-branch cases surfaced as red error
banners in the Git log panel instead of rendering as empty lists.
Reintroduce a local not-a-git-repo fallback state in the git log panel so non-repo errors still render the dedicated empty state when workspace state streaming is degraded.
Improve git repository preflight classification by distinguishing missing paths, permission issues, and non-directory targets before checking .git presence.
Add not_a_git_repository to the frontend AppErrorCode union for explicit typed handling.
Consolidate `.git` presence detection into a shared `git_repo` module used by both the workspace state watcher and the command preflight helper, replacing duplicated local definitions.
Introduce `AppErrorCode::NotAGitRepository` (HTTP 422) and preflight eleven frontend-callable git commands (log, status, list-branches, diff, diff-with-branch, show-diff, show-file, push-info, list-remotes, list-all-branches, commit-branches) so non-git folders short-circuit with a structured error instead of leaking locale-dependent git stderr.
Frontend `isNotAGitRepoError` checks the error code first and falls back to a multi-language regex list centralized in `src/i18n/git-error-patterns.ts`, covering the nine languages git actually translates into.
Wire the git log panel to `workspaceState.isGitRepo` rather than a local cached flag, so running `git init` or deleting `.git` externally propagates through the watcher and refreshes the panel automatically.
- Replace hardcoded "Claude Code" in agent_responding with each session's actual AgentType label; store agent_type on ActiveSession and populate it in both task-start and resume paths
- Remove silent ClaudeCode fallback in resolve_agent_type; /task now prompts the user to pick an agent when none is selected and the folder has no default
- Move session command strings out of session_commands.rs into i18n.rs with full 10-language coverage (en/zh-CN/zh-TW/ja/ko/es/de/fr/pt/ar)
- Fix French accent and wrap Latin agent name with Arabic bidi isolates (FSI/PDI) for consistent rendering across Telegram/Lark/WeiXin clients
- Release the bridge lock before the get_lang DB lookup in the content_delta flush path to avoid cross-session contention
- Log unknown agent_type fallback in conversation_service::parse_agent_type for observability
Reconcile the cached is_git_repo flag against the filesystem on every watch flush so `git init` or deletion of .git is reflected immediately: sync the stored flag, drop stale git_snapshot data when the repo goes away, emit a meta delta when presence flips without any data change, and mark the event as requires_resync so the frontend re-fetches the snapshot to pick up the new flag.
Replace the title-bar branch polling interval with a self-adjusting setTimeout chain that backs off to 60s when get_git_branch returns null or throws and drops back to 10s once a branch is detected, so branches created externally recover within one slow tick without hammering the backend on non-git folders.
Gate git refresh on .git presence so file churn in non-git workspaces no longer produces endless resync_hint events, and silently log tree/git refresh errors during watch flushing instead of flagging requires_resync, which turned transient failures into self-reinforcing loops.
Degrade gracefully when the filesystem watcher fails to attach (e.g. permission denied, inotify quota): keep the initial snapshot, surface a degraded flag, and expose a store-level restart that the banner uses to retry attachment after the root cause is fixed.
Propagate is_git_repo through the snapshot so the git log and changes tabs render a dedicated "Not a Git repository" empty state instead of raw git stderr with a useless retry button.
Stop polling get_git_branch from the title bar once it returns null and re-arm on visibility change.
Add translations for the new banner, empty-state, and retry keys across all ten locales.
