The task registry has run every background bash, pwsh, pty-send, and
one-shot subagent since it landed, but only the model could read it: a
human at the Web client could not see that a build was running, tell a
finished task from a stuck one, or find its outcome anywhere but the
`run_in_background` tool card that printed an id and never updated.
Task state now reaches the browser as one whole-snapshot `session/tasks`
mux frame per session, pushed at every registry commit that changes what
that session can see. `TaskService` gains `onTasksChanged`, which is
owner-granular because owner-disposal removal is a change no per-task
record can express. The carrier reads the exact owner the listener hands
it, so a push stays correct while that scope tears down, and reads the
baseline through the non-resuming `ctx.agents.get` so listing never
revives a cold session. The client keeps a last-wins mirror on
`SessionListState`, and a new `dsh-client-ui-task` package renders it
beside the subagent catalog — rendering nothing at all until the session
has a task, so an ordinary conversation grows no new chrome.
Streamed per-task output and human-initiated cancellation are separate
phases; the note records why neither has to undo this channel, and why
no Web path may call the consuming `ctx.tasks.read()`.
Assistant footers and the stats line gain TTFT/tok-per-second readings
folded from step timings; context occupancy moves off the stats line onto
a composer ring whose panel shows a heuristic system/tools/messages
breakdown from the new token-meter contextBreakdown session projection.
Public snapshot state stays in the store engine's plain-data vocabulary
(immer drafts reject Sets without the MapSet plugin, which stays off):
manager/service/contract carry readonly SessionId[] in Host order, and
the tree derivations build their own transient Set — the
expandedProjects pattern. Membership-unchanged installs still keep the
array reference for Object.is short-circuits.
WorkspaceListState gains archivedSessionIds (ReadonlySet, replaced only
on membership change), installed as full snapshots from the list
baseline, the unary echo, and the changed frame. Archiving the current
session clears the selection into the New Session view state. Test
doubles (test-runtime, fake APIs, fixture client) follow the widened
IWorkspaces/IApiClient faces.
The chat stats line took its token totals from the loaded conversation nodes,
so paging changed them and compaction erased the billing behind replaced
content. It also had no way to show context occupancy: the numerator and
capacity never reached the browser.
Both now come from token-meter session projections read through the standard
useProjection seat. Window nodes keep supplying turn and step counts plus LLM
and tool wall times, which are correctly window-scoped facts about what is on
screen; accounting no longer comes from there.
`tokenUsage` supplies billing and cache hit. `contextPressure` supplies
occupancy, pairing the newest provider-reported prompt size with the newest
capacity recorded by `request/context`. Deployments without token-meter drop
the token groups; a route whose adapter advertises no capacity drops the
occupancy group rather than rendering a placeholder.
Occupancy is deliberately approximate: the numerator and capacity are
independent last-wins fields, not one atomic request observation, so switching
models pairs a fresh capacity with the prior route's pressure until the next
request reports usage. It is a user-facing reference figure that nothing in the
harness makes decisions from, and it matches how the TUI status line has always
computed occupancy. The Agent Note and token-meter README state this as a
decision, including why the atomic alternative was implemented and rejected, so
it is not re-litigated as a defect.
Snapshot delta is one added `Context N% of 128K` segment across eight web
goldens; the preceding commit absorbed master's pre-existing golden drift.
- gate model selection on steering-placement image carriers from enqueue
until their steering/message event publishes; release the gate when an
admission ends idle without publication (both behaviorally asserted)
- reject session.updateQueue edits carrying non-text blocks at the RPC
boundary (queue edits cannot bypass image admission)
- extend the durable-directory walk past a first-created DSH_HOME to the
deepest pre-existing ancestor
- strip Windows-style separators from attachment display names on POSIX
- verify attachment reads with a header-only probe (digest already proves
the bytes decoded fully at admission); document the read path
- make SessionInputShell.addImages refusal observable and keep workspace
transfers/composer intake from leaking refused drafts
- own ONE recursive image walk (dsh-llm contentHasImage) across apiproxy,
pi-ai, compact-basic, and the DeepSeek text-only assertion
- drop the redundant canonical-base64 regex and the no-op role read
- move AttachmentId/AttachmentError out of types.ts (brand.ts/error.ts);
document why AttachmentError does not extend HarnessError
- document the hard attachments inject in both consumer READMEs
The seat's height publisher needs a ResizeObserver stub in every spec that
renders ConversationRoot (jsdom has none), and the two branch-tail StatsLine
assertions move to the grouped detail-row output.
Replace the surface-ordered fold with a log-ordered human transcript:
append-origin surface events at their own log positions plus one marker per
landed compaction checkpoint. Command folding, the tool-call index, and the
rev-keyed memo carry over unchanged.
Removes foldDegraded, the padding sentinels, baseSeq, and degradedSeqs() --
they existed only to satisfy the core fold's seq === index assertion. That
also closes the pagination hole A1 exposed: a page can carry a checkpoint
whose shadowed range fell outside the window, and nothing resolves
surfaceOp.start anymore.
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.
React half of the session-projection client base: the renderer contract gains
an open-key projections face on SessionMaybeProvideInfo (cellOf(key), distinct
from the static hooks roster), web-react mints projectionHook (per-bundle cache;
per-cell uSES binding via the shared observableHook cache; unresolved keys read
undefined through the absent source so hook order stays constant), standardKit
delivers kit.useProjection, and the runtime merges UseProjection into
SessionStandardProps/SessionMaybeStandardProps (overloads mirror useSession).
3 jsdom specs (kit delivery + live re-render, selector over undefined, faceless
bundle = all absent); existing direct-prop-feed specs gain the one-line stub the
new required seat mandates.
Mechanical --fix output over the newly linted .tsx files (indent,
arrow-parens, comma-dangle, member-delimiter-style, unnecessary type
assertions), plus the three generic-arrow test hooks converted to
function declarations up front: the comma-dangle fixer strips the
<T,> disambiguation comma and turns them into parse errors otherwise.
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.
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.
Four rounds of structural rework on the conversation surface, converging
on one registration model for the whole client:
- Review fixes: open() leaves the inject factory (SessionsService owns
the semantic); ConversationService mounts via ctx.plugin(); the
bespoke view registry retires into the 'conversation.view' list slot.
- Ring alignment: createChatView factory retired (components get
everything through checkable shares at the register call site); the
hand-rolled t/i18n threading is deleted wholesale — a future
framework-level i18n will supply t as a standard prop keyed by slot
name, so no interim manual channel.
- Toolview dissolution: ToolViewRegistry / ToolViewResolver /
ToolViewOutlet / ctx.toolviews retire. Tool rows are entries of the
'conversation.chat.toolview' keyed slot (scope: session) declared by
the chat entry; ToolRowOwnerProps is the unified owner payload;
GenericToolCard becomes the call-site fallback; registrants are plain
plugins (inject ['slots','conversation'] as the load-order seam);
session-dimension dispatch moves into components (useSessions reads
parentId); trajectory/waterfall gain same-shape slots the day they
render tool rows (RendersCheck rejects empty declarations). Slot
names mirror the composition path (<domain>.<entry>.<hole>).
- Staging follows current: cell()/binding() are pure resolution
(render-safe); the constructor subscribes to the list store and
followCurrent opens the event window when the current session
changes — staging IS the open signal, business verbs are the timing,
React render/commit is decoupled from window lifecycle. A masked
current (projection gap) keeps the stage untouched so deferred
teardown semantics survive reconnects.
Agent Note: .agents/notes/implemented/architecture/
2026-07-23-toolview-dissolution.md (bilingual pair) records the
decision, the four rejected alternatives, and the accepted semantic
changes; the web client architecture note and packages/client/AGENTS.md
carry the current-state narrative.
Verified: typecheck 0, duplication 0 clones (478 files), full coverage
run 6190 passed with zero threshold errors, knip 0, doc-sync 24/24,
client aggregate tsc 0, render-count checks (one commit per chunk, zero
row re-renders under streaming) green.
The data layer no longer depends on the React glue package, and business
plugins no longer depend on web-react at all:
- The store engine (zustand vanilla + immer + persist + dev freeze),
defineStore, and shallowEqual move to @deepseek-ai/dsh-client-runtime,
exported from the ./client main entry — no ./store subpath survives on
either package (the web-react one is deleted, none is opened on runtime).
- Store products are bare snapshot sources: useSelector leaves
SnapshotStore/StoreInstance and Session; every hook is composed at the
binding site in web-react's renderer (per-source cached uSES binding).
The SlotRendererHost sessions face carries bare observables only.
- SessionProvider becomes a standard-kit seat: an entry whose children
declare a session-scope slot receives the framework component as a prop,
retiring the last value import of web-react from plugin packages.
UseSession and the session-area types now live in ui-slots.
- web-react shrinks to the shell-only React glue (renderer, providers,
uSES bridge); zustand/immer belong to runtime alone; the module-table
seed and tsdown externals drop the web-react/store seat.
- NODE_ENV replacement is defined once in the shared tsdown client preset
(browser bundles inline the engine and lost vite's define); the 3-line
process.env typecheck shim moves to runtime with the engine.
- Stray tsc artifacts (.js/.d.ts/.d.ts.map beside sources under src/)
swept repo-wide; they shadow real sources under vitest resolution.
Verified: both aggregate typecheck programs at zero; 604 client tests
green; repo-wide grep for web-react/store at zero; real-host playwright
run 7/7 including persist round-trip.
ci: fix test/docs
The definitive slot model for the web client, replacing the first-generation
define/register two-step, ScopedSlots whitelist faces, and binding handles:
- 'root' is the only a-priori slot (SlotsService built-in); the shell renders
exactly ctx.slots.renderSlot('root', {}).
- register is the single API: children = slot declaration + render
authorization + runtime spec in one options object; misconfiguration fails
loud at load (duplicate declaration, undeclared contribution, one store
handle under two scopes).
- Component props arrive in four auto-derived shares: PropsRuntime<K>
(owner params + session/global standard kits via declare-merge),
PropsRenderSlots<S>, PropsStore<H>, and the inject business face.
sessionId is framework-supplied; hooks are framework-made only.
- Framework store seat: defineStore factories declare schema/actions/persist;
read = useStore, write = baked actions only; store scope derives from the
mounting entry; per-session persist keys and clearPersisted lifecycle.
- inject factories read the apply closure's own ctx (binding handles retired;
root-ctx back door closed); SessionProvider is self-wired render-prop.
- Rendering sits behind the SlotRenderer install seam; runtime stays
React-free; ownership ledger keyed to the single entry axis closes the
stale-authority window (StaleAuthorizationError probes).
Docs: the slot type-chain note is refreshed in place as the slot system
standard RFC (bilingual pair re-recorded); the web client architecture RFC
defers its slot sections there; packages/client/AGENTS.md gains the slot and
props discipline; gui-testing/web-styling notes drop missions/ references.
Tests: suites rewritten to the standard (props fed directly, real store
engines via createXXXStore().create(), no render machinery); load-time
negative samples for declaration/authorization/store conflicts; verified by
real-host playwright run (three columns, empty state, collapse, keyed session
remount, cross-slot selection sharing).
docs(ui-sidebar): point contract reference at the committed slot standard RFC
missions/ is workspace-local and never committed; the README must not cite it.
