TryInvokeAsync swallowed all exceptions including HubException, so a
blocked merge (uncommitted changes in target, mid-merge state, inactive
worktree) returned null silently — the task stayed WaitingForReview with
no feedback shown. Switch to a direct _hub.InvokeAsync so both VM
catch blocks (TasksIsland ErrorReported, DetailsIsland ShowErrorAsync)
actually fire.
Add regression tests for both call sites verifying that a throwing
worker causes the error to be reported.
Adds Moss / Peat / Sea preset swatches to the General settings tab.
Selecting a preset mutates the live SolidColorBrush objects in the
Application resource dictionary so all StaticResource consumers update
instantly; the choice is persisted to ui.config.json and re-applied
at startup. Missing or unknown saved value falls back to Moss.
External tasks finished outside a ClaudeDo run had no way to close out
their tracking task; update_task_status now permits Done alongside
Idle/Queued/Cancelled, refusing it when the task has an active
worktree so review_task stays the only path that merges.
## Problem
Nach "Neuen Task hinzufügen" (Eingabefeld im Tasks-Header, `Ctrl+N`) wird der Task angelegt und in die
Liste eingefügt, aber **nicht ausgewählt**. Um Beschreibung/Config zu ergänzen, muss man ihn erst
manuell anklicken.
## Ist-Zustand
`src/ClaudeDo.Ui/ViewModels/Islands/TasksIslandViewModel.cs`, `AddAsync()` (ab Zeile ~424):
legt die `TaskEntity` an, speichert über `TaskRepository.AddA
ClaudeDo-Task: 5af6ac9acb4447929e3e550c3c233a25
## Problem
Im Task-Kontextmenü gibt es bereits ein Untermenü "Markieren als …" (`tasks.ctxMarkAs`) mit
"Erledigt" und "Abgebrochen". Die beiden Manuell-Einträge hängen dagegen ganz unten flach im
Hauptmenü — inkonsistent, obwohl sie semantisch dasselbe "markieren als"-Verb sind.
## Ist-Zustand
`src/ClaudeDo.Ui/Views/Islands/TaskRowView.axaml`:
- Zeile ~47-50: `<MenuItem Header="{loc:Tr tasks.ctxM
ClaudeDo-Task: c0630785eb8e47c8bf1da4108028a3c4
Reminders written down as todos had no home: every task looked like Claude work.
A manual task now shows a MANUAL badge and hides send-to-queue, refine and the
planning session; the queue picker, daily prep and the list handler all skip it,
with a TaskStateService guard so the MCP surface and hub cannot start one either.
Opening a hand-driven ConPTY session stays available on purpose.
A list can be marked manual in its settings, which makes tasks created there
(UI and MCP add_task) start out manual. Toggle per task from its context menu.
ClaudeDo never passed --effort, so every session inherited whatever effortLevel
the user's Claude Code config happened to carry. Settings -> General now holds one
row per model alias (haiku medium/20, sonnet high/30, opus high/40, fable high/25)
supplying the global effort and turn defaults; list- and task-level max-turns
overrides still win, and the agent editor's inherited badge follows the model.
--effort is applied to autonomous runs and to every ConPTY spec (task session,
planning start/resume, ad-hoc, list handler). The model itself is deliberately not
forced on interactive sessions. The single global 'Max turns' field is replaced by
the table, and 'fable' joins ModelRegistry.Aliases.
The migration also adds the is_manual columns used by the next commit.
A task driven by hand in Mission Control is Idle with an Active worktree, so its
lifecycle chip read "Parked" — indistinguishable from a task genuinely set aside.
The shell now mirrors Mission Control's open panes onto the rows, which show an
accent "Interactive" chip instead; tapping it surfaces Mission Control and
focuses that pane (the open command already dedupes by task id).
Both actions previously gave no feedback: opening a ConPTY session only created
the tile after the launch-spec roundtrip (which may build a worktree), and the
refine button just disappeared while the run was in flight.
Add a shared Ellipse.spinner style, and let ConPtyPaneViewModel resolve its own
launch spec so the tile shows up immediately with a starting overlay. A failed
launch now keeps the tile with its inline error banner instead of never
appearing — Start() is separated from the ctor so the host can subscribe to
ErrorReported before the launch begins.
OemQuestion is the '#' key on a German layout, so the window-level binding both
swallowed '#' app-wide and left Ctrl+K dead — even though the search box already
advertises 'Ctrl K' as its shortcut.
EditableTitle had no save handler at all (only EditableDescription did), so
retitling a task in the details island was silently discarded on rebind. Mirror
the debounced description save, capturing the row so a task switch mid-debounce
cannot retitle the wrong task.
The update pipeline moved app/ and worker/ to .bak before extraction with no
retry, so a just-killed worker whose file handles had not been released yet
(WaitForExit returns before the OS flushes them) caused Directory.Move to throw
ERROR_SHARING_VIOLATION, stopping the pipeline on a cryptic error screen.
- DownloadAndExtractStep: stash/rollback moves+deletes now retry through
transient IO/access errors (~5s); a persistent lock returns an actionable
message instead of the raw error.
- StopWorkerStep: reading MainModule no longer skips Kill on failure, and a
short settle follows the kill so handles are released before extraction.
An embedded ConPTY session leaves its worktree changed but never touches
task status, so hand-driven work had no path into the review/merge flow.
Add SubmitTaskForReview: commit the worktree (same auto-commit as a headless
run), then transition Idle/Failed -> WaitingForReview via the new
TaskStateService.SubmitInteractiveForReviewAsync. Approve then merges it.
Surfaces: a 'Submit for review' button in the detail work console (shown for
an Idle/Failed task with a worktree) and on the ConPTY Command Center pane
header (task-based panes; closes the pane on success). Tests cover the new
transition (Idle/Failed accepted, Running/Queued/Done/Review rejected).
Approve & Merge is disabled until the pending changes have been inspected:
DetailsIslandViewModel tracks ReviewDiffViewed (reset on task switch and on
every state change), MergeSectionViewModel raises DiffViewed when a diff or
combined diff is opened and exposes HasReviewableDiff, and a hint sits next
to the button. A review with nothing to inspect (sandbox run, no worktree)
approves straight through.
ClaudeDo-Task: f9809a93
Interactive planning sessions launched with --permission-mode plan, which
gates EVERY MCP tool call regardless of --allowedTools (verified: even a
read-only mcp__claudedo__list_task_lists is denied under plan mode). So the
session prompted the user on the first CreateChildTask -- the whole point of
a planning session.
Switch BuildPlanningStartArgs/BuildPlanningResumeArgs to --permission-mode
default, which honours the allowlist. File edits stay blocked via the planning
system prompt + AllowedTools omitting Write/Edit/Bash. Resume also re-passes
--allowedTools, since the CLI does not restore it across --resume.
The earlier 'glob does not match' hypothesis was empirically falsified:
mcp__claudedo__*, the bare server name, and the explicit tool name all allow
the tool with zero permission_denials in default mode.
Planning start/resume opened an external Windows Terminal (wt) window.
Route them through the embedded ConPTY Command Center pane instead, matching
the existing interactive-session UX (no external window).
- Extract bare planning arg builders (BuildPlanningStartArgs/ResumeArgs) from
WindowsTerminalLauncher; the wt path still uses them (kept, not removed).
- InteractiveLaunchSpecService.BuildPlanningStart/Resume map a planning
context into a LaunchSpec (planning args + env: MAX_THINKING_TOKENS,
CLAUDEDO_PLANNING_TOKEN). Hub GetPlanningStart/ResumeLaunchSpec run
StartAsync/ResumeAsync then return the spec.
- UI: OpenPlanningSession + the resume branch raise OpenPlanningConPtyRequested;
the shell opens Mission Control and hosts a planning ConPTY pane.
Env is process-global by design (sequential human-paced sessions). wt planning
code retained. Tests added for the arg/env mapping.
THe update notification should be in the Bottom Rught hand corner as a Small Popup card. IT should have the Update and Dismiss Butotns. It should only show when there is a new Update available, not if everything is up to date
ClaudeDo-Task: c14606f009e44e4ba04bfe5bcbc884ad
The inline answer banner for a running task's AskUser question only existed
in Mission Control. Surface the same banner in the detail pane, bound to the
detail island's existing TaskMonitorViewModel (shared state — no new state,
no duplication) so a question can be answered without opening Mission
Control. Mirrors MC: live TaskQuestionAsked events for the selected task.
A planning parent stays Status=Idle while its session is Active (or
Finalized-but-not-queued), so the lifecycle chip read "Idle"/"Parked"
next to the PLANNING/PLANNED badge. Hide the chip in that case; the badge
already conveys the state.
A --json-schema run can echo the structured {"summary":...} JSON as the
result text, which then landed verbatim in task.Result and rendered raw in
the OUTCOME card. Unwrap the summary in StreamAnalyzer when the result text
is itself such a JSON object (a plain markdown result is kept as-is), and
add a UI safety net in ApplyOutcome for rows already persisted as raw JSON.
Finalize/Discard broadcast only the parent's TaskUpdated; the delta path
updated the parent row but never recomputed child-derived flags or dropped
discarded children, so subtasks stayed "Draft" after finalize and deleted
rows lingered after discard until a manual reload. Reconcile the whole
list when the updated task is (or owns) a planning subtree.
Opening the dropped IStorageFile stream ran outside the try, so a
first-drop failure escaped the async-void handler as an unobserved fault
(generic "An error occurred"). Wrap the stream-open loop and surface the
exception type via DropStatus; include the exception type in the generic
add-file failure too, so the intermittent case is analyzable.
ResumePlanningSessionAsync wrapped everything in an empty catch, so a
failed resume vanished silently. Report it via the footer error channel
with a dedicated planningResumeFailed message.
Continue was already gated on CanContinue (all files resolved); surface a
footer hint counting the still-unresolved conflicts/files so the disabled
state is explained, not just for binary files.
Show an explanatory empty-state under the install row when no skills are
installed. Rename the unified-parent status/labels from "Improvements" to
neutral "Subtasks" (waitingForChildren, agentStatus.children,
childOutcomesLabel) since WaitingForChildren now covers planning too.
Icon.Plus was stroke-only geometry, invisible when used in a PathIcon
(fills). Author it as a filled cross, fixing every PathIcon use at once
(New-session, add-list, add-task). Replace the header bar's raw gear
TextBlock with the shared Icon.Settings PathIcon for consistency.
- CancelAsync now signals the running Claude process of the cancelled task and
its cascaded children via the new RunCancellationRegistry (queue + override
slots register their CTS there) instead of only flipping DB state.
- external MCP review_task 'approve' now mirrors the hub's ApproveReview:
unit merge for parents, ApproveAndMergeAsync for childless tasks, optional
targetBranch; ReviewTaskResult carries mergeStatus/conflicts.
mika kuns