Die Build-Sektion hat bisher ein Rezept mit allen sechs Testprojekten vorgegeben;
Implementer haben daraus "volle Suite vor fertig" gelesen. Bei bis zu 5
parallelen Tasks kostet das Stunden - und kann das nicht finden, wofuer es
teuer ist: jeder Task arbeitet in seinem eigenen Worktree mit nur seiner eigenen
Aenderung, eine Kollision zwischen zwei parallelen Tasks ist fuer beide
unsichtbar. Kollisionen zeigen sich nach dem Merge, und genau dort laeuft
verify_merges den Verify-Command einmal pro Batch.
Neue Regel: Solution-Build immer (42s, faengt projektuebergreifende
Compile-Brueche), Tests nur --no-build --filter auf das Beruehrte, und melden
welcher Filter benutzt wurde. Die Messwerte stehen dabei, damit die Regel nicht
als Geschmacksfrage gelesen wird.
Messung 2026-09-02, sequenziell nach einem Build: Worker.Tests 507s, Ui 33s,
Data 8s, Rest je 3s, Build 42s. Parallel ueber die Solution ist es schlechter,
nicht besser (Worker 801s, Ui 82s) - diese Tests legen echte Git-Worktrees an
und spawnen Prozesse, sie drangeln sich statt sich zu teilen.
Ausserdem korrigiert: die Zeile "dotnet build ClaudeDo.slnx requires .NET 9"
war veraltet.
Der Toggle sitzt unter den Gauges im Usage-Monitor, weil "das restliche Fenster
ausnutzen" eine Entscheidung ist, die man beim Blick auf das restliche Fenster
trifft. Er speichert sofort per Read-modify-write, gleiche Disziplin wie
SaveStagesAsync, damit er keine Felder ueberschreibt, die er nicht besitzt.
Zwei Stellen, die das Opt-in still ausgeschaltet haetten:
- WorkerHub.UpdateListConfig loescht die Config-Zeile, wenn alle Felder leer
sind. PowerUseAllowed muss in diese Leer-Pruefung, sonst verschwindet das
Opt-in sofort wieder, wenn es das einzige gesetzte Feld ist.
- SettingsModalViewModel baut das AppSettingsDto positional neu auf. Ohne
Round-trip haette jedes Speichern der Settings PowerUseEnabled auf den
Record-Default false zurueckgesetzt. Gleiches Muster wie _throttleStages.
Das Handoff-Budget ist als Spinner in Settings -> General editierbar, weil es
auch fuer normale (interaktive) Handler-Runs gilt und nicht an PowerUse haengt.
Das Addendum selbst ist ueber Settings -> Files anpassbar.
HandoffRoundTracker.Register nimmt den Cap jetzt als Parameter statt aus einer
Konstante; HandoffMcpTools liest ihn pro Aufruf aus app_settings. Werte unter 1
werden auf 1 geklemmt (in WorkerHub UND in Register): der Cap ist das Einzige,
was eine entlaufene Handoff-Kette begrenzt, "0 = unbegrenzt" darf kein Aufrufer
waehlen koennen.
InteractiveLaunchSpecService.ComposeHandlerPromptAsync haengt das PowerUse-
Addendum an den Phasen-Prompt, aber nur wenn globaler Schalter und Listen-Opt-in
beide gesetzt sind. Bewusst ein Addendum statt einer zweiten Prompt-Garnitur:
die gemeinsamen 95% bleiben an einer Stelle, und ein Run ohne PowerUse bekommt
einen byte-identischen Prompt wie vorher. Das Addendum ist als Override der
interaktiven Regeln formuliert (nie fragen, entscheiden und protokollieren;
Konflikte loesen oder abbrechen, nie im Tree liegen lassen; Arbeit ernten, wenn
eine Runde sonst leer laeuft).
Drei Tests pinnen alle vier Ecken des UND, weil die Sicherheitsaussage des
Features genau daran haengt.
app_settings bekommt power_use_enabled (Master-Schalter, wird im Usage-Monitor
gesetzt) und list_handler_max_handoffs (Default 4 = zwei Merge-Runden).
list_config bekommt power_use_allowed als Opt-in pro Liste.
Ein Handler-Run laeuft nur unbeaufsichtigt, wenn BEIDE Schalter gesetzt sind.
Damit kann der globale Schalter ueber Nacht an bleiben, ohne dass der Handler
einer nicht zugestimmten Liste sein Verhalten aendert.
UpdateListConfigDto.PowerUseAllowed ist bewusst bool? (tri-state, null =
gespeicherten Wert behalten), analog zu TicketProjectId: SetConfigAsync kopiert
verbatim, ein non-nullable bool haette jedem fremden Writer erlaubt, das Opt-in
per Auslassung zu widerrufen. Der Test dazu nagelt genau das fest.
Migration ist drei reine ALTER TABLE ADD COLUMN, kein Table-Rebuild.
Conflict resolution for the #266 merge: #247 had moved PreviewMerge/PreviewMergeSet
and MergePreviewSetEntryDto out of ExternalMcpService into MergeMcpTools, so #266's
in-place edits to those had to be re-applied at the new location, together with the
PreviewMergeSet_ReportsExpectsNoDiffPerTask test.
## WO — verifiziert am 2026-09-02, nicht neu suchen
- Merge-Phasen-Prompt: `src/ClaudeDo.Data/PromptFiles.cs`, Konstante `MergeHelperMergeDefault` (ab Z. 495). Der neue Warte-Abschnitt gehört zwischen "## Verify once — after the last merge" (Z. 526) und "## Handoff / end" (Z. 542) — also nach dem Verify, vor der Handoff-Entscheidung.
- Brief-Template: `MergeHelperInitialDefault` (Z. 555) und `Mer
ClaudeDo-Task: 1788adc4-c671-4942-ab78-1ef1812da494
Data.Tests: default-false and true-roundtrip on TaskEntity.
Worker.Tests: add_task/update_task set/clear/leave-unchanged, and
preview_merge_set surfaces the flag per task.
`src/ClaudeDo.Worker/External/ExternalMcpService.cs` ist **1897 Zeilen** (nachgezählt 2026-09-02; die Datei ist seit dem Anlegen dieses Tasks von 1696 weiter gewachsen): ~30 DTO-Records oben plus ~25 Tool-Methoden — während **13** Geschwisterdateien `*McpTools.cs` im selben Ordner das Topic-Split-Muster schon tragen (`ListMcpTools`, `ConfigMcpTools`, `BatchMcpTools`, `NotesMcpTools`, `QueueStateMc
ClaudeDo-Task: 1862e160-ffde-4a50-aa03-bb7251f56088
An empty diff from a review/audit/verification task looks identical to
one that silently failed to commit. ExpectsNoDiff (TaskEntity, add_task/
update_task, preview_merge_set, review_task via TaskRefDto) lets the
merge helper tell them apart instead of treating every 0-changed-files
result as suspicious.
`src/ClaudeDo.Worker/External/ExternalMcpService.cs` ist **1897 Zeilen** (nachgezählt 2026-09-02; die Datei ist seit dem Anlegen dieses Tasks von 1696 weiter gewachsen): ~30 DTO-Records oben plus ~25 Tool-Methoden — während **13** Geschwisterdateien `*McpTools.cs` im selben Ordner das Topic-Split-Muster schon tragen (`ListMcpTools`, `ConfigMcpTools`, `BatchMcpTools`, `NotesMcpTools`, `QueueStateMc
ClaudeDo-Task: 1862e160-ffde-4a50-aa03-bb7251f56088
Guardrail for the upcoming ExternalMcpService.cs topic split: proves the
move is name-preserving by pinning the complete set of registered tool
names (computed via the SDK's own McpServerTool.Create, not a
reimplementation of the naming rule) before any file is moved.
Loc.Current (src/ClaudeDo.Ui/Localization/Loc.cs) is a process-wide mutable
static; which localizer is active depended on xUnit's parallel test
scheduling. Ui.Tests: disable collection parallelization (cheap, pure VM
suite) and install the real localizer in a module initializer so no test
can ever observe the uninitialized key-echo default. Worker.Tests: install
the same module initializer but leave parallelization on — disabling it
measured a 27% wall-clock regression on its real-SQLite/git suite, and the
initializer alone already removes the race. AgentConfigEditorViewModelTests
now installs its own localizer instead of relying on ambient state.
The tool description claimed Failed tasks carry failureReason, but the field
was missing from BatchTaskDetailDto and rejected when requested in `fields`.
Add failureReason to the DTO, KnownDetailFields, and BuildDetail so list
handlers and other callers can read failure reasons without per-task get_task
calls. Tested: 31 BatchMcpToolsTests pass, including two new cases for
failureReason retrieval on Failed and Idle tasks.
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Worker.Tests carried its own hand-rolled FakeWorkerClient implementing the
whole IWorkerClient surface, duplicating StubWorkerClient in Ui.Tests. Every
hub-signature change had to be applied twice. Link StubWorkerClient.cs into
Worker.Tests via a csproj Compile Include and subclass it there; add
SetTaskStatusCalls tracking to the shared stub (the one behavior Worker.Tests
needed that Ui.Tests didn't have).
Triage phase now audits dependsOnTaskId chains before handoff: distinguishes real
from artificial (plan-numbered) dependencies, resolves artificial ones via
update_task, groups survivors into a wave plan, and checks the wave count against
the 4-handoff cap (2 merge rounds) so a chain too long to finish says so explicitly
instead of stranding mid-run.
Merge phase now reads a task's own diff when preview_merge_set's changedFileCount
is below 5, reserving the sonnet review subagent for larger diffs where the
context-protection cost is actually worth it.
Updates DefaultFor_merge_helper_merge_never_reads_diffs_itself, renamed to
DefaultFor_merge_helper_merge_gates_the_diff_subagent_on_a_file_count_threshold,
to match the new threshold behavior.
src/ClaudeDo.Ui/ViewModels/Islands/TaskRowViewModel.cs enthält ~107 handgeschriebene OnPropertyChanged(nameof(...))-Aufrufe in partial void OnXChanged-Methoden. Das Projekt nutzt an anderen Stellen den Generator: [NotifyPropertyChangedFor] in DetailsIslandViewModel.cs, ConflictResolverViewModel.cs, AgentConfigEditorViewModel.cs, Installer/WizardViewModel.cs.
Das ist die eine Datei, in der eine ve
ClaudeDo-Task: 87976462-5845-4e01-8233-32afadca02bd
From `docs/open.md` (re-verified against the code 2026-09-02, still open): nothing reconciles `git worktree list` against the `worktrees` table. `ExternalMcpService.ListWorktrees` and the Worktrees overview read DB rows only; `WorktreeManager` calls git solely inside its self-heal path (`ListWorktreePathsForBranchAsync`). Observed once: 22 git-registered worktrees + 26 `claudedo/*` branches in `C:
ClaudeDo-Task: 765fed13-db66-44e4-9285-c4b2d415b443
Sibling migrations branched off the same parent have silently dropped
each other's columns during a SQLite table rebuild before, and every
other Data.Tests fixture uses EnsureCreated, which skips the migration
chain entirely and would never catch that.
LocaleStore.Load silently skips a *.json that fails to parse, so a broken
de.json left store.Available with only "en" and the parity loop compared
English against itself, always green. Add two guards before the loop:
loaded-locale count must match the number of *.json files in the folder,
and "de" must be present via TryGet, both with failure messages naming the
missing file. Also independently JsonDocument.Parse every *.json so a
syntax error surfaces as a syntax error with the file name, not as
"locale missing". LocaleStore's catch-and-skip is untouched -- a broken
translation must still not crash the app at runtime.
"New list" pre-creates the DB row before opening List Settings.
Cancelling or closing that modal left an untitled empty list behind;
CreateListAsync now rolls the row back unless the modal reports Saved.
Also moves the new-list/import-repo buttons up next to the "My Lists"
label instead of a row below the list, and puts Notes above Important
in the smart-list nav order.
Reuses the existing drag-drop move (MoveTaskToListAsync) so the same
running/worktree guards apply — only ListId changes.
Also count WaitingForReview tasks (not Done-with-a-branch) for the
review portion of the status pill, and localize the pill and open-task
count strings that were still hardcoded English.
Sdk.Web defaults to Server GC, one heap per core with a ~350 MB
gen0 budget each — an idle worker sat at ~230 MB private. Workstation
GC drops that to ~40 MB with no server-side concurrency to lose.
The banner compared the running worker's build SHA against the
selected list's HEAD via a git ancestor check on every list switch
and reconnect. It added a git-rev-parse MSBuild target, a hub call,
and a poll loop for a corner case that hasn't come up in practice —
not worth the moving parts.
.git as a file (worktrees, submodules) isn't a standalone repo — the
scanner was flooding the import list with worktree entries alongside
their real repo.
The status PATCH needs statusName alongside statusId or the API 400s,
so every write-back had been silently failing. The board's flat
assigneeName is null on every live ticket; the real assignee list is
ticketAssignees[]. Import now also picks up In Bearbeitung tickets,
not just Offen ones.
Both issues are live-verified against the real API, so the feature
flag flips back on.
Listen bekommen eine Icon-Spalte (Migration AddListIcon) und einen Icon-Picker
in den Listeneinstellungen; 34 kuratierte Geometrien aus der Icon-Bibliothek,
Fallback auf "Folder" bei unbekanntem Key.
Der Metadata-Footer im Task-Detail entfällt; created-at wandert in die
Header-Zeile neben das ID-Badge, der Close-Button (und CloseDetailsCommand)
fällt weg.
- "+" öffnet ein Menü mit allen Listen (working dir) statt direkt den
Ordner-Dialog; "Ordner wählen…" bleibt als Fallback erste Zeile, damit
das Menü auch bei fehlgeschlagener Listen-Abfrage nutzbar bleibt
- Rail-Rows: Kontextmenü "Sitzung öffnen" (gesperrt für Running — der
Worktree gehört dort dem autonomen Lauf)
- Aktivitäts-Strip statt reiner Queue: WaitingForReview/Failed/Roadblock
kommen dazu, Attention zuerst, unterschieden nur über Farbe (Aktionen
bleiben in der Task-Detailansicht)
ApproveAndMergeAsync treated any non-Active worktree (including Kept, the
state RevertMergeAsync leaves behind) the same as "never had a worktree",
so re-approving a reverted task ran only the verify gate and marked the
task Done without merging anything back.
Now a Kept worktree is distinguished: if its branch is gone (the common
case, since approve normally deletes it), approve returns Blocked instead
of a fake "merged". If the branch survives, its tip is already an ancestor
of the target (from the original merge), so a plain `git merge` would
silently no-op ("Already up to date."); cherry-picking the original merge
commit's own diff (-m 1) is what actually restores the content, recorded
as a fresh MergeCommit.
CancelAsync only checked PlanningMergeOrchestrator's in-memory HasActiveMerge, which is
empty after a restart. IActiveMergeState gains an async fallback that checks on-disk
(GitService.IsMidMergeAsync) for a WaitingForReview parent with children, so the guard
still blocks cancel until the merge is continued or aborted.
ScopeGlobs auf Tasks und SerializeOnFileOverlap auf der Listen-Config waren
ungenutzt: der Scope wurde nie befuellt, also hat der Queue-Picker nie
serialisiert. ScopeOverlap, das Picker-Gate, die DTO-Felder, die UI-Option und
die Spalten fallen weg (Migration DropFileScopeSerialization).
Die Bandel-API ist unter dem eingetragenen Hostnamen nicht erreichbar
(Verbindung aktiv abgelehnt), das Feature bliebe also eine Einrichtung,
die nicht funktionieren kann. TicketSettingsTabViewModel.FeatureEnabled
versteckt alle drei Einstiegspunkte: Settings-Tab, Projekt-Dropdown in
den Listen-Einstellungen, Kontextmenue-Eintrag.
Gespeicherte Base-URL, PAT und Projektverknuepfungen bleiben erhalten
und greifen beim Umlegen sofort wieder.
Ein blanker Hostname ist eine relative URI; HttpClient lehnt sie mit
InvalidOperationException ab. Die lief an dem zu engen catch-Filter
vorbei, flog roh aus dem Hub und wurde von TryInvokeAsync zu null
verschluckt - die UI meldete "Worker not reachable", obwohl der Worker
lief und die Ursache eine Eingabe war.
- TicketSystemConfig ergaenzt fehlendes http://
- BandelTicketClient uebersetzt jede Transport-Exception in eine
lesbare TicketApiException
- TestTicketConnection faengt alles und antwortet mit dem Grund
- GetTicketSettings liefert die effektive URL, damit die Ergaenzung
sichtbar ist
Der Footer-Strip trug schon immer ein Level, es gab nur keinen
Einstiegspunkt jenseits von FlashFooterError. "3 neue Tasks" in Rot
liest sich als Fehlschlag.
Tickets/-Ordner, TaskStateService.NotifyAsync, ticket_api_base_url und TicketRef/
TicketProjectId dokumentiert. Verbatim-Copy-Warnung um TicketProjectId ergänzt und
den PermissionMode-Vorbestandsbug (SetConfigAsync liess das Feld im Update-Zweig
aus, gefixt in f0a3a186) vermerkt. docs/open.md um die fuenf offenen Punkte aus der
manuellen Verifikation des Specs ergänzt, inkl. Hinweis dass die Bandel-Response-
Shapes nie live verifiziert wurden.
Inline Foreground/Stroke an Icons entfernt — ein lokaler Wert schlaegt jeden
Style, dadurch erreichten Hover/Checked/Disabled das Icon nicht. Farbe wird
jetzt per Klasse gewaehlt (.danger, .star-btn.on, .active), alle Zustaende
liegen in einem geordneten Block in IslandStyles.axaml.
- Path.plan-icon -> Path.icon-stroke (gilt jetzt fuer Button und ToggleButton)
- Icon.Activity als gefuellte Kontur (PathIcon fuellt, Stroke-Polyline gab Blobs)
- BloodBrightBrush als Hover-Ton fuer destruktive Icons
- icon-btn:disabled bleibt transparent statt Fluent-Grau
Bisher nur per Hand in worker.config.json. Neuer Abschnitt CLAUDE CLI mit
eigenem Save-Button (die Datei gehoert dem Worker, nicht app_settings) und einer
Zeile, wohin der Wert per ExecutableResolver tatsaechlich aufloest — 'nicht im
PATH gefunden' war bisher erst am fehlgeschlagenen Run zu sehen. WorkerConfig ist
DI-Singleton und ClaudeProcess loest pro Spawn auf, also greift die Aenderung ab
dem naechsten Run ohne Neustart.
SaveOnlineInbox/SaveClaudeBin teilen jetzt ein SaveKey(), damit beide dieselbe
read-modify-write-Semantik haben (alle anderen Keys bleiben unberuehrt).
Die zugehoerigen Locale-Keys sind im vorigen Commit mitgelaufen.
Die vier UsageThrottle-Prozente waren nur per Gauge-Drag im Usage-Monitor
erreichbar und wurden im Settings-Modal unsichtbar durchgeschleift. Jetzt
read-only angezeigt plus Button in den Usage-Monitor. Der Button schliesst die
Settings absichtlich: Save() schreibt die beim Oeffnen gelesenen Throttle-Werte
zurueck und wuerde ein zwischenzeitliches Draggen ueberschreiben.
Bisher gab es nur AppSettings.DefaultPermissionMode global — ein Task, der plan
oder acceptEdits braucht, erzwang das Umstellen der globalen Einstellung. Neue
Spalten tasks.permission_mode und list_config.permission_mode, Auflösung
task -> list -> global im EffectiveRunConfigResolver (den TaskRunner und
get_effective_run_config gemeinsam nutzen), ComboBox mit Inherited-Badge im
geteilten Agent-Editor.
MigrationBaselineTests: das Fixture baute per EnsureCreated das heutige Schema
und stempelte Legacy-History darauf — jede Migration nach dem Squash lief damit
in 'duplicate column name'. Es migriert jetzt gezielt bis InitialCreate und
prueft 'nichts pending' statt 'InitialCreate ist die einzige Zeile'.
Das Flag war bisher nur ueber MCP set_list_config erreichbar. UpdateListConfigDto
fuehrt es tri-state (null = gespeicherten Wert behalten), damit nur das
Listen-Modal es setzen/loeschen kann und kein anderer Aufrufer es per Omission
verliert.
Icon.GitBranch ist Line-Art und wurde per PathIcon gefuellt (nur Knoten sichtbar).
Zusaetzlich brach {Binding Task.HasBranch} bei leerer Auswahl (Task=null) und
IsVisible fiel auf den Default true zurueck — FallbackValue=False blendet die
Zeile jetzt aus, solange kein Task gebunden ist.
Neuer Button in der virtuellen Review-Liste: öffnet das Merge-Helper-Auswahlmodal
im Review-Modus (alle WaitingForReview-Tasks repo-verknüpfter Listen, gruppiert
unter Listen-Headern, alle vorgetickt) und startet pro gewählter Liste eine
ConPTY-Handler-Session direkt in der Merge-Phase — via bestehendem Handoff-Spec
(nextPhase "merge"), kein Triage/Wait davor. Handler-Host-Task wie beim vollen
"Let Claude handle it"-Lauf (HandlerBaseCommit, IsManual).
Delete (row context menu + detail header) now gates on HasInteractiveSession
the same way SendToQueue already does, so an open ConPTY session's claude
process can't be orphaned by deleting its task out from under it.
Row removal (in-UI delete and the delta-refresh path for externally deleted
tasks) now clears SelectedTask via SelectFrom(null, "row-removed") when the
removed row was selected, so the detail pane no longer keeps showing a
deleted task with live buttons.
The DB's global UtcConverter only tags DateTimes as Utc on read
(SpecifyKind), it never converts on write. SetScheduledForAsync persisted
the ThemedDatePicker's Local/Unspecified wall-clock value verbatim, so
QueuePicker's comparison against DateTime.UtcNow fired scheduled tasks late
by the local UTC offset (e.g. 2h in CEST). Convert to UTC at the write
boundary, and ToLocalTime() at the read/compare sites (overdue checks in
TaskRowViewModel/TasksIslandViewModel, the date-picker's edit seed) so
existing scheduled/overdue display doesn't shift.
Existing DB rows hold local wall-clock values mistagged as Utc; no
migration added (few rows, one-time 2h-class shift accepted per the
originating audit finding).
Per-merge verify made an N-task batch pay N x ~7 min, each run testing
the same moving main. review_task/merge_task/continue_merge now take
skipVerify: the merge lands but the gate AND the Done transition are
deferred (status merged_verify_pending, task stays WaitingForReview).
The new verify_merges tool runs the list verify command once for the
explicitly listed tasks and promotes them to Done on success - explicit
ids so the handler own submitted task and parked verify_failed tasks
are never swept up; an Active worktree is refused per entry.
PlanningMergeOrchestrator threads the flag through the unit merge (no
more per-child verify) and skips FinalizeParentDoneAsync when verifies
were deferred. The merge-helper Merge prompt approves with
skipVerify=true and calls verify_merges once after the last merge. UI
approve is unchanged; the no-Done-without-green-verify invariant stays
server-enforced.
ConflictMarkerParser only recognizes text-marker conflicts, so a modify/delete or
rename/delete conflict (git never writes markers for those) was invisible to the
resolver: GetConflictDocumentsAsync read it as an already-resolved stable file, and
ContinueMergeAsync's marker scan let it through, so `git add` on the unmerged path
silently kept whichever side wasn't deleted.
GetConflictDocumentsAsync now tells a modify/delete-style conflict (one index stage
missing) apart from an ordinary already-resolved file (both stages present) and
synthesizes a real whole-file conflict block from the index stages so it shows up in
the resolver. ContinueMergeAsync compares such a path against git's own default
checkout content before treating it as resolved, so an untouched file blocks the
merge instead of being silently staged. Resolving to the deleted side (empty content)
now runs `git rm` instead of staging a tracked, zero-byte file.
update_task_status Idle/Done, WorkerHub.SetTaskStatus, and the UI's Mark
Done/Cancelled menu items could all force a Running task's status without
going through TaskStateService's guards, leaving its CLI process untracked
and its worktree writes silently dropped. Idle now routes through the
already-guarded TaskStateService.ResetToIdleAsync instead of the repo's
unconditional ResetToManualAsync; Done and SetTaskStatus explicitly reject
Running; the context menu disables both entries with a tooltip while
running. TaskRunner also now logs a warning whenever a terminal transition
(SubmitForChildren/SubmitForReview/Complete/Fail) is rejected instead of
silently discarding the run result. DeleteTask's MCP path gets the same
friendly foreign-key message WorkerHub.DeleteTask already had for a task
with children.
IBaseDirtyChecker, IInteractiveLaunchSpecService and the LaunchSpec wrapper
had one implementation and one caller each; ProcessRunnerAdapter existed only
to give a static class an interface, and InstallArtifactLocator used
inheritance for two constructor arguments. The external MCP container now
shares its singletons through a Share<T> helper instead of 17 near-identical
registrations.
WorkerHub and IWorkerClient each carried their own copy of every record and
had already drifted — MergePreviewDto lost its verify fields on the client
side, two records disagreed on their name. They now live in Data/Wire.cs and
reach both sides via a <Using> item, so a hub signature change is a compile
error instead of a silently dropped JSON field.
The incremental chain was ~31k lines of generated Designer snapshots for a
schema no shipped database steps through anymore. MigrateAndConfigure now
baselines an existing DB onto the squashed id (pre-EF, or history ending at
AddPrimeActionKind) and throws with a reinstall message for a mid-chain DB.
MigrationBaselineTests covers all three paths with a real Migrate().
Set Stateless = true on the internal (/mcp on the SignalR port) and external
MCP HTTP transports. Two effects:
- No Mcp-Session-Id, so a worker restart can no longer 404 a session that
outlives it -- ConPTY tiles in the UI process and externally registered
claude sessions keep working across a restart.
- A 2026-07-28 client is no longer refused back to the initialize handshake.
A stateful server rejects that revision on purpose (see the SDK's
StreamableHttpHandler.s_sessionSupportingProtocolVersions), so the new
per-request protocol path was unreachable regardless of SDK version.
Nothing here used the stateful-only features (sampling, elicitation, resource
subscriptions, unsolicited notifications). In-tool progress notifications ride
the POST's own response stream and are unaffected -- which matters, since ~20
tools use them to hold off the client's 300s idle abort. All three consumers
(TaskRunner, PlanningSessionManager, the installer's RegisterMcpStep) already
register type: "http", so dropping the legacy SSE endpoint breaks nothing.
Verified against the running worker across two restarts: tools/list returns
57 tools over a bare POST with no initialize and no session id, and a full
2026-07-28 tools/call round-trip returns real data.
MarkDoneAsync was Idle-only, so a finished task (typically a list-handler
run with no worktree to merge) could not be ticked off. Guard on Running
instead and cover the other statuses with tests.
The re-templated TabControl dropped PART_ItemsPresenter, so no TabItem
containers were realized and SelectedContent stayed null -> empty body.
Keep the presenter collapsed instead. Also let the last-run label trim
with a tooltip rather than reserving a fixed MinWidth.
Nine interfaces had exactly one implementation and no test double — they existed
only to be named twice in a DI registration: IFindingsStore, IFindingsStoreLocator,
IPrimeScheduleSignal, IRefineRunner, IWeekReportService, IMergeCoordinator,
IMissionControlPane, IOnlineLoginService, ITaskListFilter. Consumers now depend on
the concrete type; the DTO records that shared those files moved next to their
implementation. IInteractiveLaunchSpecService stays — it carries 54 lines of
contract documentation, which is not ceremony.
IMergeCoordinator in particular had a redundant null object: MergeCoordinator with
a null Handler already no-ops, and every test used the real class with Handler set.
Filtering/ collapses from 8 files to 1. ITaskListFilter and TaskListFilterBase were
a double abstraction over four predicates, with MatchesAsContext => false declared
in both. SmartFlagFilter also compiled its expression twice (its own _flag plus the
inherited Matches cache) — it now uses the cache.
StaticTokenAuthProvider was in src but production uses ZitadelAuthProvider; it is
a test double, so it moves to the test project. Its own test goes away with it.
TestDbFactory was redeclared as a private nested class in 40 test files,
NullServiceProvider and StubNotesApi in 14 each, DefaultStub in 5 — 73
declarations, all semantically identical (StubNotesApi differed only in
formatting and type qualification). They now live in TestDoubles.cs next to the
existing StubWorkerClient, which was already the shared-double pattern in this
project.
Eight public members had no caller anywhere in src: GitService.GetFileDiffAsync,
SubtaskRepository.DeleteByTaskIdAsync, TaskRepository.GetByListAsync (a
backwards-compat alias for GetByListIdAsync) and .GetByCreatorAsync,
WorktreeRepository.GetByStatesAsync, TaskMonitorViewModel.SetPendingQuestion
(a duplicate of the live-event lambda), PrimeClaudeTabViewModel.ApplyFiredEvent,
and StreamLineFormatter.FormatFile. TaskAttachmentRepository.DeleteAllForTaskAsync
was reachable only from its own test; the ON DELETE CASCADE on task_attachments
already covers it. Tests for the deleted members go with them.
Dropped two package refs the platform already provides: EntityFrameworkCore.Design
in Worker (the design-time factory and the migrations live in Data, which has its
own ref) and System.IO.FileSystem.AccessControl in Installer.Tests (net8.0-windows
ships the ACL APIs in the shared framework).
Mika Kuns