feat: add slash command suggestions for provider configuration - #8

…RY, tests

Batch 1 — Compile Errors & Correctness:
- Add TOGGLE_KANBAN to KEYBIND_LABELS and KEYBIND_GROUPS (coder#1)
- Fix Ctrl+K conflict → Ctrl+Shift+K to avoid PREV_WORKSPACE collision (coder#2)
- Add TOGGLE_KANBAN dispatch handler in App.tsx (coder#3)
- Fix dirty-cache in promoteQueuedTask: move flag clearing after moveTask (coder#5)
- Wrap 4 IPC helpers in KanbanBoard with try/catch (coder#4)

Batch 2 — Consistency & DRY:
- Deduplicate KanbanColumnSchema: import from canonical orpc/schemas (coder#6)
- Remove unused KanbanBoardData import from kanbanStorage.test (coder#7)
- Remove dead KANBAN_STORAGE_KEYS constant (coder#8)
- Replace console.warn with log helper in kanbanStorage (coder#9)
- Replace useEffect form reset with key prop in TaskDetailModal (coder#10)
- Remove unsafe {} as KanbanTask cast, use EMPTY_FORM pattern (coder#11)
- Add loading state + error logging to KanbanView (coder#12, coder#13)
- Add KANBAN_STATUS_LABELS helper, use in TaskDetailModal (coder#14)
- Validate column membership in reorderTasks (coder#15)
- Use KanbanTaskPriority type import in TaskCreateModal (coder#16)

Batch 3 — Tests:
- Add 4 tests for updateColumn (coder#17)
- Add 3 tests for promoteQueuedTask (coder#18)
- Rename misleading test 'rejects empty title' → 'accepts empty title' (coder#19)

43 tests pass (was 36). No typecheck or lint regressions.

…ming aggregation (coder#4016)

## Summary

Splits src/browser/stores/WorkspaceStore.ts's three responsibilities at
two seams: the IPC subscription transport (reconnect loops, exponential
backoff, stall watchdogs, attempt-scoped aborts) moves into a new pure
`subscriptionTransport.ts` module, and high-frequency streaming deltas
(`stream-delta`, `reasoning-delta`, `tool-call-delta`, advisor live
output) move off whole-workspace state bumps onto a display-row-keyed
channel consumed by the leaf message renderer. WorkspaceStore becomes a
reactive cache; during streaming, only the row a delta changes
re-renders, not ChatPane and not sibling rows.

**Net LOC: production −100, tests −551 (total −651).**

## Background

Refactor coder#8 from the 2026-08-29 architecture review (evidence at main @
f04e0f8). The transport skeleton was copy-pasted across the onChat,
activity, and terminalActivity loops and only testable through mock
router harnesses. Every token delta called `states.bump(workspaceId)`,
re-rendering ChatPaneContent (1,991 lines) and re-running its 10 heavy
useMemos per delta, exactly what the repo's colocate-subscriptions rule
forbids.

## Implementation

- `runSubscriptionLoop` owns the generic attempt lifecycle (backoff
schedule + reset-on-event, stall watchdog with start-after-connect
semantics, attempt-scoped AbortController bound to the client generation
captured before the client await). It has no WorkspaceStore or React
knowledge and is fully driven by fake async iterators in tests.
Per-stream semantics (since-cursors, caught-up snapshots, replay
reconciliation, the single-active-onChat assertion) stay in
WorkspaceStore's callbacks.
- `subscribeToStats` / `subscribeToTimeline` intentionally stay as-is:
single-shot streams with a genuinely different lifecycle; forcing them
through the coordinator would add a shallow layer for no deletion win.
- Streaming deltas bump a channel keyed by the active message's trailing
displayed row, and `MessageRenderer` subscribes per row via
`useStreamingMessageDelta`, so exactly one subscriber wakes per delta
even for tool-heavy responses split into many rows. A delta that adds a
part creates a new display row, which no leaf has mounted yet, so that
case escalates to a workspace bump (plus a streaming-stats
invalidation); structural events (stream start/end, tool lifecycle,
init/caught-up) keep bumping whole WorkspaceState, preserving metadata
and hydration behavior.
- The live lookup only overrides the actively streaming message, so
completed rows keep transcript-level transforms (e.g. merged
stream-error counts). Channel keys are tracked per workspace and
released on terminal events, background activity stops, replay clears
(advisor), and workspace removal, so MapStore version maps do not grow
unbounded.
- At stream end the published final message and the live-read content
come from the same aggregator, so completion causes no visible content
swap.

## Irreducible additions (per-item)

- `subscriptionTransport.ts` (+151): the extracted seam itself; replaces
~480 lines of triplicated loop code.
- `subscriptionTransport.test.ts` (+279): table-driven scenarios over
fake iterators/timers (backoff, watchdog, abort ordering,
client-generation swap race); replaces ~1,180 lines of deleted
mock-router retry/watchdog tests in WorkspaceStore.test.ts.
- Behavioral tests in WorkspaceStore.test.ts: row-keyed delta routing
without whole-state bumps, row-creation escalation, text accumulation
equality, key release on terminal/background-stop paths, delta vs stale
bootstrap race.

## Validation

- Remote dogfood UAT (regression-focused, adversarial): long streaming
turns with tool calls, backend kill/restart mid-stream, window reload
mid-stream, workspace switch mid-stream, Escape interrupt, metadata
time-series vs baseline. A render-count probe confirmed the streaming
row updates without whole-pane re-renders.
- Reconnect forensics compared the branch against its fork-point
baseline with identical methodology (chat.jsonl + DOM message-id
inspection on both): behavior is structurally identical. Two
pre-existing main behaviors were documented and intentionally not
"fixed" here (resume-after-restart persists a partial + continuation
message pair, splitting one logical response across two rows; first-turn
aggregate stats populate late). Both reproduce identically at f04e0f8;
changing them would be behavior changes outside this refactor's scope.
- Five Codex review rounds hardened the delta channels (row mounting for
part-creating deltas, stats invalidation, key lifecycle across
background stops and replay clears, client-generation race, merged-error
row preservation); every fix landed with a discriminating test, most
red-green verified.
- `make static-check` green; whole-file `bun test` green for every
touched test file.

## Risks

Medium-sensitivity area: chat streaming and reconnect. The transport
extraction is structure-preserving (same backoff constants, watchdog
timings, and abort ordering, now asserted by deterministic unit tests),
and delta isolation only narrows which subscribers re-render; it does
not change event application. Highest residual risk is a subtle
re-render dependency on per-delta content outside the leaf path; UAT's
smoothness and metadata time-series probes targeted exactly that and
found parity with baseline.

---

_Generated with `xum` on Mike's behalf • Model:
`anthropic:claude-fable-5` • Thinking: `xhigh` • Cost: `$61.13`_

<!-- xum-attribution: model=anthropic:claude-fable-5 thinking=xhigh
costs=61.13 -->

## Stack

Layer 9/10 of the architecture refactor stack (net -5,101 LOC overall).
This PR's diff is only this layer, against `mike/arch-composer-hooks`.