Headless Avalonia ListBox+VirtualizingStackPanel spike answers all five
questions from the Phase 2a gate task: containers stay bounded at 1000
items, the existing InputHitTest ghost-drag model survives recycling,
variable row heights don't drift the Extent (only naive index*height
math would), auto-scroll via Offset mutation is mechanically feasible,
and mixed header/task rows template correctly via implicit DataTemplates.
Recommendation: proceed to Phase 2b. Spike code itself is throwaway and
not checked in.
Every tool description now leads with what the tool does AND when to reach for
it, since MCP clients rank tools by that text. Per-parameter prose moved onto
the parameters as [Description], exhaustive result-shape enumerations and
design/history rationale dropped, and the repeated boilerplate clauses
(lean-task-ref, batch cap, refused-while-Running) pulled into McpToolDocs,
which also documents the style for future tools.
Tool-level description text: 20494 -> 13605 chars (-34%); combined with the new
parameter descriptions 18517 (-10%).
Closes gaps that caused wrong calls rather than just verbose ones:
- list_task_attachments returns metadata only, no file content
- run_task_now shares continue_task's single override slot and throws when busy
- list_runs is ordered oldest-first and feeds get_run
- workingDir on create_list/update_list is an existing local git repo path,
unvalidated until the first task run
- get_task_worktree's behind=0 also means the main ref was unreachable
Removes get_task_status_values: a whole tool entry for static reference text.
GetTask's description is now the canonical place for status meanings.
Both merge-helper ConPTY sessions loaded PromptKind.MergeHelper, so the
post-handoff session received the phase 0-2 dedupe/enhance instructions and was
told to ignore them by its brief alone.
Split into MergeHelperTriage (phases 0-2 + handoff) and MergeHelperExecute
(phases 3-5), so each session carries only its own phases. Consolidated the
generic ask-the-user rule to one place per prompt, scoped Phase 5's summary to
what the execute session actually knows, and moved the dedupe/enhance bilanz to
the triage handoff. Regression guards assert neither prompt carries the other's
phase headings and that the shared-checkout git rule stays in execute.
Submit for Review is now disabled while a ConPTY pane is starting, has
failed to launch, or has already exited, and MissionControlViewModel
guards against a rapid double-click racing two SubmitTaskForReviewAsync
calls. A failed launch no longer permanently occupies its TaskId dedupe
slot -- a Retry button re-fetches the launch spec and restarts the pane
in place. CloseConPtySession/Dispose now also unsubscribe
SubmitForReviewRequested, matching the other pane event handlers. Also
fixes the pre-existing nullable-dereference warning in
IslandsShellViewModel.SyncInteractiveSessionChips.
TaskStateService.CancelAsync allowed cancelling a WaitingForReview task
even while PlanningMergeOrchestrator was mid-drain on it: ApproveReview
awaits the whole multi-subtask merge synchronously, so a concurrent
CancelReview (UI or MCP) could flip the parent to Cancelled while the
orchestrator kept merging children onto the target branch, then
FinalizeParentDoneAsync would find the parent no longer WaitingForReview
and give up - leaving the merged diffs stranded with no rollback.
CancelAsync now rejects with a clear reason when HasActiveMerge(taskId)
is true. TaskStateService can't take PlanningMergeOrchestrator as a
direct constructor dependency (circular back to ITaskStateService), so
it takes a lazily-resolved Func<IActiveMergeState> instead, mirroring
the existing Func<ITaskStateService> cycle-break already used for
PlanningChainCoordinator.
UI polish: DetailsIslandViewModel.IsMergeDraining gates
CancelReviewCommand's CanExecute (same shape as
WorktreesOverviewModalViewModel.IsMerging), and the command's catch now
raises ErrorReported instead of swallowing the rejection silently.
Bumps the verified-against commit to 774f9d3 (the RunNow/picker
double-dispatch race fix) and documents the new claim-before-create
ordering and the RunCancellationRegistry double-registration guard.
Bump conpty-sessions.md's verified-against commit to d84607f and add a
section covering the CanSendToQueue / EnqueueTaskAsync interactive-session
gate and the switch to routing through the worker hub.
Deleting a child from the details pane hard-deleted straight from the UI
process via TaskRepository, bypassing TaskStateService.TryAdvanceParentAsync
entirely. Deleting the last child of a WaitingForChildren parent that way
left it wedged forever. WorkerHub.DeleteTask now mirrors the MCP delete_task
tool (running-task guard, FK-friendly error, advance-parent call), and the
UI goes through it.
TryAdvanceParentAsync also short-circuited when zero children remained,
treating "no children left" as "nothing to evaluate" instead of "all done" -
removed the early return so an empty child list (vacuously) counts as all
terminal.
review_task/continue_merge on a planning parent always leaves conflicts in
the tree, and the UI auto-opened the resolver on every PlanningMergeConflict
broadcast regardless of who started the merge -- so a running Claude session
resolving a unit-merge conflict could race a human editing the same shared
checkout in a resolver window neither of them asked for.
PlanningMergeOrchestrator.StartAsync now takes an externallyDriven flag (set
by ExternalMcpService's MCP-driven review_task path, left false for the UI's
ApproveReview) that rides along on the PlanningMergeConflict broadcast. The
UI only auto-opens the resolver when it's false; otherwise it shows a
persistent banner with a manual "Open resolver" button, cleared on
PlanningMergeAborted/PlanningCompleted. A new GetActiveExternalConflictsAsync
query (checked against GitService.IsMidMergeAsync rather than the in-memory
flag alone) lets the UI resync the banner on reconnect instead of trusting a
one-shot broadcast that isn't replayed after a restart.
The childless single-task conflict path was checked and needed no change --
it only broadcasts the generic TaskUpdated, never PlanningMergeConflict.
preview_merge/preview_merge_set now report isEmpty (ahead==0, or
HandlerBaseCommit==HandlerHeadCommit for a worktree-less list-handler
task) instead of leaving an empty branch indistinguishable from a small
one. preview_merge also stops throwing for worktree-less handler tasks,
falling back to their fixed commit range. review_task's parent approve
returns emptyChildren, naming the Done children whose review range
contributed nothing before the unit merge lands. TaskRefDto/TaskDto now
expose roadblockCount so a CLAUDEDO_BLOCKED child is identifiable over
MCP, since it still reaches Done per the unified parent model.
WaitingForChildren already counted as "changed" since it's outside Queued/Running,
so waiting on a planning parent returned immediately even though its children were
still running. The new opt-in flag (default false, unchanged behavior) keeps polling
through WaitingForChildren and only reports changed once the parent reaches
WaitingForReview or a terminal status.
list_tasks on a list of ~100 verbosely-described tasks could return
390k+ characters in one call, blowing past the caller's token limit.
Both tools now default to lean TaskRefDto references (no
Description/Result) and take an includeDescription flag to opt back
into the full TaskDto payload — same flag-alongside-nullable-payload
idiom already used by BatchGetTaskResult/TaskConfigResult. get_task is
unchanged.
add_task and batch_add_tasks now report up to 3 open (non-terminal)
tasks in the same list with a strongly overlapping title, so a
parallel agent can notice and mention a likely duplicate instead of
silently creating one. The task is always created regardless. Uses a
cheap normalized-word overlap heuristic (no embeddings/LLM call),
robust to German umlaut/digraph spelling variants. Breaking change:
AddTask now returns AddTaskResult { task, possibleDuplicates } instead
of a bare TaskRefDto; BatchAddTaskResult gained a PossibleDuplicates
field.
review-merge: worktree-less approvals run the gate too, and all three UI merge
entry points handle verify_failed explicitly.
Ui gotchas: NumericUpDown.Value is decimal? and nulls while the box is empty.
Documents the Checks/ + SystemCheckPage feature (gating rule, check list,
FreshInstall-only placement) and the ExecutableResolver/.cmd-shim wiring in
ClaudeDo.Worker's ClaudeCliPreflight/ClaudeProcess. Adds docs/explore-notes/installer-preflight.md
(bumped/corrected against the actual implementation) and links it from the
explore-notes README and root CLAUDE.md.
The underlying code lives only on unmerged task branches (06aca9b3.../40272c0b...),
not on main yet, and two follow-up features ("Claude Help Me" button, Config-mode
Diagnose section) were never implemented because they blocked on that same missing
merge. Both gaps are called out explicitly in the new docs and in docs/open.md's
verification checklist, instead of being asserted as done.
MaxTimeoutSeconds was 170s against runs that take tens of minutes, forcing
a dozen full-context wait rounds per long-running batch. Raise it to 900s
and raise MCP_TOOL_TIMEOUT in lockstep (ClaudeProcess + every
InteractiveLaunchSpecService launch spec) to 930000ms so the client
connection actually stays open that long instead of aborting first.
Add get_queue_state (QueueStateMcpTools): configured vs. effective
parallel-slot count (via QueueService.GetSlotCountsAsync, extracted from
the former GetEffectiveMaxParallelAsync), active slots with taskId +
startedAt including the run_task_now override slot, and queued tasks in
pick order -- so a caller can observe queue occupancy instead of inferring
it from maxParallelExecutions.
Adds a read-only get_effective_run_config(taskId) tool that reports the
model/max-turns/effort/permission-mode/agent-path/system-prompt/skills a
task will actually run with, each tagged with its source (task/list/
preset/global), plus max-turns' raw requested value and clamp status.
Extracted the model/max-turns/agent-path resolution out of
TaskRunner.ResolveConfigAsync into EffectiveRunConfigResolver so the run
path and the new reporting tool share one codepath instead of risking
drift, per docs/explore-notes/worker-task-pipeline.md's max-turns trap.
## Der Zweifel
Ein autonomer Agent und eine interaktive Session sollen sich unterschiedlich verhalten. Unklar war, ob beide denselben System-Prompt mitbekommen.
## Vorbefund (2026-08-05, beim Verfeinern erhoben — **selbst nachpruefen, nicht uebernehmen**)
Sie teilen ihn **nicht**. Pro Startweg:
| Startweg | Einstieg | System-Prompt |
|---|---|---|
| Autonomer Run | `ClaudeArgsBuilder.cs:69-73`
ClaudeDo-Task: b259a1c9a73846ce833f0b555fb1de79
mika kuns