commandId crosses three boundaries (session log, wire admission response,
client flow pairing), so per the branded-id rule it becomes
Branded<'CommandId'>, declared in a new pure @deepseek-ai/dsh-commands/brand
outlet (the dsh-llm/brand shape: type + constructor, no Context merges, so
wire and client programs can name it without loading the host plugin). The
event payloads, CommandExecution, and the executor mint carry the brand; the
wire schema gains commandIdSchema as the domain's single brand-cast point
(the approvals precedent); CommandNode and the fixture's fabrication cast
follow type-only.
The manager's titleSnapshots Map and its session/title frame consumption
dissolve into resident per-session ProjectionValueStores (create-on-demand,
outliving instantiation — the same role the snapshot map played): a
session/projection frame lands whether or not the Session exists, list rows
read the store's 'title' key, subscribed baselines truncate phantom rows, and
session-removed drops the store. The fixture converts to the host parallel:
a projections block on the tail page (title + todos units), push frames on
unit-advancing events, and a post-subscribe projection baseline replacing the
bespoke title control frame.
TodoDock reads useProjection('todos') (whole list or null pre-first-write;
absent renders nothing) and declare-merges the todos key through the pure-type
outlet — the identical member the tool-todo host unit owns, drift rejected by
any program holding both. The core Session's todos field, its todo/write case,
and the snapshot member retire; the client folds nothing. Session specs for
the retired client fold move to the value-store spec's seq coverage; snapshot
literals across component specs drop the field.
Client-side domain folding is gone (RFC final: the host is the only
computation site): ProjectionCellSpec/fromEvent, ProjectionCellSet, the
SessionsService cell roster, and the Session event-dispatch projection hooks
all delete; Session.projections becomes the generic value store (manager-owned
via SessionOptions so frames landing before instantiation and the history
baseline converge on one row set), and installWindow only seeds the store from
a carried block. The cell specs retire with the machinery — the value store's
own spec owns the seq semantics now.
The push-model client base (session-projection RFC final): ProjectionValueStore
holds key → {value, seq} per session, seeded by the tail page's projections
block and updated by session/projection frames under one rule — higher seq
wins on both paths (stale baseline cannot overwrite a newer frame; replayed
frames cannot regress; an omitting fresh baseline clears = capability absent);
truncate() drops phantom rows past a subscribed durable baseline. Per-key
identity-stable faces (always defined; absence is an undefined snapshot) feed
useProjection; the renderer contract's projections member becomes faceOf.
12 store specs cover both seq directions, absence, truncation, and batching.
CommandService.execute now returns a CommandExecution — the normalized
result plus the commandId minted for its command/run/command/done records —
and the wire admission value carries commandId exactly when matched, so the
issuing client can correlate its RPC acknowledgment with the flow node the
lifecycle events produce. apiproxy api/schema/handler, the connection
fixture, and the TUI/plan/goal consumers follow the new shape.
The line field is deleted (pre-release, no shim): name and args are
parseCommand's own split — name plus verbatim rawInput with its separator
whitespace — so a consumer (a projection unit folding its own command
records, a rich command card) never re-parses a line. CommandNode mirrors
the split (name/args, both null on a run-less cross-window node); the
generic card rebuilds its display line as /name + args. The connection
fixture logs the same structured payload.
The FoldAdapter folds the log-only command/run + command/done pair (paired
by commandId) into a CommandNode outside the surface fold and merges the
nodes into the flow by seq; cross-window cuts soft-fall like tool pairs (a
done-only window builds the node from the done, a run with no done renders
as still executing). ChatView renders command nodes through the new keyed
'conversation.chat.commandview' hole (key = command name) with
GenericCommandCard — a stripped-down GenericToolCard showing the command
line and outcome text — as the render-site fallback, so any slash command
renders durably with zero registration and survives refresh, other tabs,
and resume via the mux-broadcast events.
The wire type now carries projections?: SessionProjectionsBlock (host-base
landed), so the structural projectionsOf narrowing and its TODO(gui) go away —
Session reads result.value.projections directly at all three installWindow
sites. ProjectionsBaseline stays as the cell framework's structural twin
(React-free layer keeps depending on the type table only) with values typed
Partial<SessionProjectionMap>; the erased walk moves inside resetBaseline
where per-key typing is re-established by schema.parse.
Swap the client runtime's parallel-construction placeholder for
import type from @deepseek-ai/dsh-session-projection/types — the zero-import
outlet, never the package root, whose dsh-agent → dsh-session chain would drag
the host Context.sessions merge into the client program. One type table end to
end (host provider, wire block, client cell, React hook); the spec's test key
now declare-merges the real module. Adds the workspace dep and the tsconfig
project reference.
Object layer of the session-projection RFC client base: ProjectionCellSpec/
ProjectionCell/ProjectionCellSet with the single seq-watermark rule (live and
window-replace events share one filter; baseline reset re-seeds value+watermark
unless a newer commit applied; absent key = capability absent), Session
dispatch at appendLive/installWindow (projections block read structurally,
TODO(gui) switch to the interface package), SessionsService.registerProjectionCell
roster (live scopes now + future scopes at mint; disposer sweeps every session),
and the provideInfo projections face (key-addressed bare cell sources).
15 object-layer specs: watermark no-rollback, late-baseline seq rule,
capability absence, schema-failure degrade, duplicate-key throw, resync e2e.
- unused SessionId import and an unnecessary assertion in ui-model
- two over-length fake-api lines split
- trigger chrome calibrated against figma 313:14108's ToggleButton
(13/20 medium secondary at 28px; chevron in label-caption = the mock's
#ADB2B8; the mock's leading icon is invisible there, so the seat has
none, and its 'High' effort text has no wire concept to render)
An omitted `todos` on a tail request was treated as "no projection carried"
and preserved the prior value. Every installWindow caller is a tail request
(doOpen, its gap re-pull, repairGap; loadOlder prepends without it), which the
host answers with the full-log projection or omits only when the log holds no
todo/write — so the field's absence is the authoritative empty list. A live
write whose host crashed before persisting therefore left the rolled-back plan
on screen indefinitely; the assignment now clears it on the next open or
resync. Widened the parameter to an explicit `| undefined` so the two meanings
cannot be conflated again, and updated the JSDoc at both declaring seams plus
the bilingual README/note pair.
- host trio kept on the merged api-proxy: session.models (provider-grouped
advisory directory), session.selectModel (validated provider, advisory
model), installAgentLlmTarget threaded through create/resume/ensureSession;
the gateway declares the llm inject it reads
- history no longer piggybacks modelTarget: the current target travels on
session.models alone (the /model popup is the sole consumer)
- new @deepseek-ai/dsh-client-ui-model plugin: /model popupSelect over the
wire — options load the directory (group label in the detail column,
provider-local failures listed inline), onSelect routes selectModel;
failures ride the popup shell's error/retry surface
- ModelSelector package, conversation.composer.control slot, and the
Session-side modelSelection state machine are removed: model selection
belongs to the /model popup; the named conversation.input.model seat
stays empty until a control-seat entry is designed for it
The session/queued frame no longer carries steering — AgentMessage no longer
has the field. The client derives it from the same ordered turn boundaries the
host saw (a frame arriving while a turn is open joined the steering FIFO).
- Mount WorkspacesService.startInitialSelection in the runtime apply (the
one-shot baseline follower shipped in 98633b5aa without a caller): a
restored current session wins, an explicit clear stays cleared, a failed
connect retries on the next baseline projection.
- Cover the policy in client-apply and the assembled workspace-flow
snapshot; startup now lands in the recent Workspace's blank session, so
the draft-carry scenario starts from the hero directly.
- Bring docs along: startup-selection paragraphs in the session-scope RFC
note (both languages), bilingual README pairs for the four new client
packages, doc-graph regeneration with client-declared events exempt from
the dispatcher requirement (client dispatch sites are structurally
invisible to the host-side ts.Program), and pairing re-records.
eslint --fix autofixes plus manual repairs: max-len line splits
(fake-api handlers, notifier/slots JSDoc, spec signatures), charAt over
non-null-asserted indexing in slash detect/menu cores, Array.from for
code-point capping, typeof assertions for unbound-method in specs,
generic getByRole for the send-button cast, effect disposer void-wrap in
command register, and dropped unused type imports.
Conflict resolution: session.ts keeps this branch's paired-start callTime
(the start event exists here; a settle-only window stays null — the PR2
fix's semantics compose). The README's codeDispatches section is restored
to the running→settled lifecycle this branch implements, keeping PR2's
null-callTime nuance for starts outside the window.
Responding to ds-review-bot round 2 on #653: the tool/code-dispatch event
is appended at settlement, so using its time as callTime fabricated a
zero-duration call for duration-aware consumers — it is now null (start
unknown) per the ToolResultNode contract, pinned in the session spec. The
README's codeDispatches section described the PR3 running→settled
lifecycle a stack ahead of this tree; it now documents the settled-only
index this PR ships (the running shape lands with the start event in
#658, which already merges cleanly over this).
Responding to ds-review-bot on #664 (root cause lives here): when a history
window carries a tool/code-dispatch settle without its paired start, the
runtime fabricated callTime = settle time, so downstream duration views
presented a measured 0 ms. Match the native tool-result contract instead —
callTime: null = unknown — and pin it; the trajectory cell already renders
null as the em dash, and the waterfall gains explicit unknown handling in
its own PR.
The bridge replaces its serialization queue with a pool that reuses the
native concurrency contract: submissions classify through
registry.executionMode (fail-closed isConcurrencySafe), start strictly in
submission order, overlap up to the validated maxParallelSubCalls config
(default 10; 1 restores serial), and exclusive calls drain the pool, run
alone, and bar later calls. Each started sub-call logs a
tool/code-dispatch-start event at pool entry; the existing
tool/code-dispatch settles the pair (started ⇔ settles exactly once;
abandoned queued calls log neither). SDK prompt guidance now states the
true Promise.all contract — re-recorded across every code/both-mode
snapshot (plus the stale cordis-dynamic-toolchain fixture gaining the
required description arg).
Client: CodeSubCall widens to RunningToolCall | ToolResultNode — starts
land the running shape (rows wear the native running ring), settles
replace in place preserving start order, callTime pairs to the start
time. Fixture emits start/settle pairs; jsdom pins the running sub-row;
runtime specs pin in-place settlement and out-of-order completion.
The client indexes tool/code-dispatch events into
ConversationSnapshot.codeDispatches (parent callId -> ToolResultNode-shaped
sub-calls; live mux and history replay build the identical index). ChatView
renders each run_code parent as the new code variant (description summary,
program as the expanded monospace body) with its sub-dispatches as
always-visible indented rows — every sub-row dispatches through the SAME
keyed conversation.chat.toolview hole with the same GenericToolCard
fallback, so custom registrations (bash sample) take over sub-rows exactly
as top-level rows. The details panel resolves sub-callIds to full logged
args and complete output through the native path.
Evidence: fixture turn 64 + built-bundle jsdom snapshot, real-machinery
jsdom suites (nesting, error state, details, running parent, reference
stability), and a recorded code-mode browser e2e round (keyless replay +
aria golden). Scaffold gains a toolsMode patch knob.
The hero composer renders the frontend Session Intent's retained prompt
from the sessions list snapshot, but updateIntent only reached that
snapshot through the intent watch's microtask-deferred markDirty. React
therefore rolled the controlled textarea back during the change tick,
which corrupted IME composition (Pinyin "nihao" committed fragments
like "nnini hni hani hao你好") and jumped the caret on plain typing.
SessionManager.updateIntent now calls notifyNow after
updatePendingPrompt, per the Notifier channel rule for direct echoes of
user gestures. The workspace-flow snapshot helper asserts the same-tick
echo instead of waiting for it, and a runtime unit test pins the
contract at the manager seam.
Sidebar session list grows the figma 239-10458 feature set and the
workspace/session browsing region moves wholesale into ui-workspace:
- Group-by menu (WorkSpace / In one list): flat mode lists every session
top-level, strictly newest-first; the choice persists across reloads.
- Session rows get a 500ms hover detail card (title / relative time /
status line) and a ... menu (Rename / Fork session / Delete session,
visual-only for now); workspace headers get ... with Rename (wired) and
Delete workspace (visual-only).
- workspace.rename RPC: trims, rejects duplicate titles on the create
chain (workspace-name-conflict), no-op on same title; modal dialog with
client-side duplicate pre-check.
- workspace.insertSessionBefore RPC (DOM-insertBefore semantics, omitted
anchor appends): HTML5 drag reorder of root sessions inside a workspace
group; order truth stays host-side, the view refreshes from the
response/changed frame.
- Activity pinning removed: the session/event touchSession chain is gone;
workspace accounts are manually owned (new sessions prepend, explicit
reordering only). Contracts and tests updated, api catalog regenerated.
- ui-sidebar reduced to the column shell (brand, fold state machine, New
Session, Settings) exposing one sidebar.workspaces hole with a two-fact
owner share {wide, expandSidebar}; ui-workspace owns the whole region
(header, search, grouped/flat lists, dialogs, drag) plus the picker via
a shared WorkspaceCreateFlow. The old sidebar.workspace picker slot and
its deferral indirection are gone.
- ui-primitives: Menu gains label entries, danger rows, and
closeOnPointerLeave; new HoverCard (portaled, open-delay, disabled
guard). Hover card and row menu never coexist.
repairGap installed the repulled window without the response projection;
a todo/write missed during the gap and already outside the new tail page
kept the stale list. The spec pins adoption through the repair path.
The client's todos projection derived only from the paged display window,
so reopening a session whose last todo/write preceded the tail page showed
an empty plan until the user paged back — session-level state cannot be
reconstructed from an arbitrary window. The host owns the full log, so the
tail history response now attaches todos (latest todo/write backscan, the
same posture as the view pairing); installWindow seeds it, window rebuilds
preserve it, and any in-window or live write keeps overwriting it. The
fixture mirrors the host; docs and both Agent Notes record the mechanism.
Session consumes the todo/write session event as a per-event side effect
(last write wins), rebuilds it on window replay/paging/resync, and exposes
snapshot.todos. TodoItem re-exported through the runtime surface.
