0.2.16#531
Merged
Merged
Conversation
…close fix(installer): stop bundled processes before windows upgrade
fix(ui): prevent jump rail number clipping
…povers fix(composer): anchor execution dropdowns under zoom
…rectly (v2) (#448) * Implement event throttling for SSE messages Added a message throttling queue and timer to batch events before sending to the frontend. * fix(desktop): throttle SSE events across IPC and add unit tests --------- Co-authored-by: xingyu <1736101137@qq.com>
feat(speech): add local whisper transcription
The SDD requirement status pills (planned/building/done/verified) only overrode the text color for the dark theme while keeping the light-mode background, which is a saturated translucent fill tuned for light surfaces. On the dark canvas the pill background stayed too bright and clashed with the already-adjusted light text color. Add matching dark-theme background overrides so the pill fill tracks the text color at a soft opacity, and add dark-theme segments for the SDD requirement progress bar so the filled segments use the lighter tailwind shades instead of the vivid base hues.
Add the Retroma parchment-inspired light palette from Talkcody as a built-in color mode in the mode workshop, toggled via a palette button on the default Kun card. Light-only: all Retroma selectors carry a :not([data-theme='dark']) guard so dark mode falls back to the Kun default. Activation mirrors iKun (data-retroma-mode DOM attribute + full --ds-* token block in base-shell.css), with no mascot figures.
feat(lsp): language server diagnostics (#466) + review fixes
feat(plugins): OAuth connector previews (#480) + URL validation
feat(cache): cache diagnostics (#485) + aggregation fixes
feat(terminal): monochrome + custom colors + CJK (#497) + fixes
feat(skills): import from GitHub (#464) + security hardening
Plan Mode was a single useState on the singleton Workbench, so toggling it in one chat affected every chat. Move the mode into useChatStore as composerMode, keyed per-thread via kun.threadComposerMode.v1 (mirroring the existing per-thread composerModel pattern), with a global kun.composerMode default for the no-active-thread case. - chat-store-helpers: add ComposerPlanMode type, normalize/read/persist helpers, per-thread remember/read map (capped at MAX_THREAD_COMPOSER_SELECTIONS), and composerModeForThread (prefers stored value, falls back to thread.mode === 'plan', else 'agent'). - chat-store-types/chat-store: add composerMode state + setComposerMode action; initialize from readStoredComposerMode. - chat-store-app-actions: setComposerMode writes the per-thread map when a thread is active, else the global default. - chat-store-thread-actions: selectThread hydrates composerMode via composerModeForThread. - Workbench/workbench-plan-controller: consume composerMode/setComposerMode from the store instead of local useState. Tests: add per-thread persistence and global-default isolation coverage in chat-store-helpers.test.ts and chat-store-app-actions.test.ts.
…mode feat(composer): make plan mode per-thread
fix(settings): 优化设置页布局
…icons fix(editor): load cross-platform picker icons
…58-rollback-safety
…78-schedule-queue
…56-mac-topbar # Conflicts: # src/renderer/src/components/chat/WorkbenchTopBar.tsx
The PR widened modelIdSchema to 512 chars (MAX_MODEL_ID_LENGTH) but two modelHint fields in clawTaskFromTextPayloadSchema and scheduleTaskFromTextPayloadSchema were still hard-capped at 128, which re-triggers issue #473 on the IM/fromText IPC route. Reuse modelIdSchema for both modelHint fields and add a regression test that exercises a 256-char id on each schema. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- Rewrite rollbackWorkspaceConfirmDetail (EN/ZH) to disclose that git reset --hard + git clean -fd will permanently destroy uncommitted edits, untracked files, and commits since the checkpoint. The old copy said 'only restores workspace files' which hid the destruction. - Surface rescueCheckpointId after a successful restore. Previously the action discarded it, leaving users with no path back if the rollback was a mistake. Now the id is console.info-logged for power users and shown in the transient store notice via the new rollbackWorkspaceSuccessWithRescue locale key. - Re-check busy AFTER the confirm dialog resolves. The user can type and send during the await; running git reset --hard mid-turn would wipe files the agent is actively editing. Reuse the existing rollbackWorkspaceBusyError key. - Gate rewindFileRollbackNotice on block.meta.workspaceCheckpointId. Users in non-git workspaces no longer see a 'file changes will be rolled back' notice when nothing actually rolls back. - Add store tests covering busy-after-confirm short-circuit and rescueCheckpointId surfacing (log + notice). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…, add tests Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Two HIGH-severity correctness fixes on top of #478's queued-task feature: 1. Worktree slot cleanup. Scheduled agent runs always leave changesCount>0, so findAvailablePoolIndex would permanently skip every used slot after a few runs. releaseWorktree now resets HEAD and runs git clean -fd in the slot before dropping the lease. Failure is best-effort: a slot that refuses to clean stays dirty (same as today) rather than blocking the release. Chose the worktree-service location so any future caller (renderer-side maintenance, removeWorktree, etc.) benefits — added an opt-out skipClean for callers that need it. 2. Concurrency cap race. runTask used to do `size < MAX_CONCURRENT` then await store.load() then enter runTaskInternal. Two concurrent IPC callers could both pass the size check before either incremented runningTaskIds. All paths now route through enqueueTask + drainQueue; the size check sits inside drainQueue (serialized by the drainingQueue flag), and the running slot is reserved synchronously inside drainQueue before runTaskInternal awaits anything. Immediate-run semantics are preserved via a per-task completion deferred so the caller still awaits the eventual ScheduleRunResult. Bonus: when a useWorktree task can't acquire a slot and other worktree tasks are running, the task is re-enqueued (with a short setTimeout backoff) instead of failing — once a slot frees, drainQueue picks it up. Tests: added (a) a four-runs-at-once race that verifies the cap holds, and (b) a slot-reuse test that confirms the same pool index is reusable across three back-to-back runs. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
fix(settings): accept long provider model ids (#475) + close modelHint gap
feat(memory): compact record details dialog (#502) + UX safety
feat(schedule): queued worktree task execution (#478) + correctness fixes
feat(chat): git rollback action (#458) + safety fixes
fix: macOS top-bar alignment + workflow sidebar back (#456) + develop merge
Plugin tokens were injected only on <html>, but in dark mode the `[data-theme='dark'] .ds-workbench-shell` rule re-declares the palette tokens locally, shadowing the injected values for the entire conversation subtree. Settings/sidebar live outside `.ds-workbench-shell`, so they themed correctly while the Workbench kept the default Kun blue. Light mode has no such local re-declaration, hence the dark-only symptom. Extend the generated token selector to also target `.ds-workbench-shell` (specificity 0,3,1 > the 0,2,0 base rule), mirroring the existing iKun `[data-ikun-mode='on'] .ds-workbench-shell` handling, so plugin tokens win inside the Workbench too. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Forking inherited the parent's `mode` and could carry a GUI plan context
bound to the source workspace. A forked "new conversation" then ran as a
plan turn against a stale plan context, producing both reported errors:
- "tool workspace does not match the active GUI plan workspace" from
create_plan (the plan context's workspace != the forked thread's), and
- a provider 400 "A parameter specified in the request is not valid"
(plan mode narrows the tool set to create_plan, which the cloned
history's other tool calls/results can't satisfy).
Root cure, two layers:
- fork() starts a fresh agent conversation (`mode: 'agent'`) instead of
inheriting plan mode; the plan artifact and its workspace belong to the
source thread.
- agent-loop ignores a plan context whose workspace doesn't match the
thread (isStalePlanContext) and no longer treats such a turn as a plan
turn, so it runs as a normal agent turn instead of hard-failing.
Adds tests for both layers.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- workbench-plan-controller: set in-flight thread marker *before* awaiting sendMessage to prevent fast-response race where create_plan block lands before the ref is tagged (new threads get re-tagged after send resolves) - workbench-plan-controller: extract shouldAutoOpenPlanPanel() pure helper + thread-id-tagged ref so a plan turn in thread A can't pop open thread B - Workbench.tsx: pass activeSkillWorkspace (thread's workspace, not the global workspaceRoot) to PlanPanel so guiPlanMatchesContext succeeds on the freshly-created plan and clearActivePlan() is not triggered - PlanPanel: temporary console.warn on both clear paths to diagnose the workspace/thread values when clearing does fire (remove after root cause is confirmed) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
7 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.