The thread reads the session's queued prompts from the chat_subscribe reply
and chat_queue frames: a reload draws them as Queued after the transcript,
a delivered one enters the transcript as Sent, a refused one becomes Not
sent, and Remove sends chat_unqueue. A prompt the server reports queued is
dropped from the device unsent store; the any-frame queued heuristic is
used only with a server that does not report its queue.
A sent message is stored on the device until the server accepts it, so a
reload mid-request no longer loses it: on reopening, a stored intent the
history records is cleared as sent, otherwise it shows as Not sent with
Retry (same operation id) and Edit. The operator's bubble shows Sending,
Sent (newest only), Queued or Not sent. A live refusal keeps the old
contract (exact draft and chips back in the composer, unchanged re-send
reuses the id) and marks the bubble Not sent with Retry. On web, leaving
or reloading the page while a prompt is in flight asks first.
The transcript listener read the lazily built _controller while _attach was
still building it (load() notifies synchronously), so it built another
controller, recursively, until the test process ran out of memory.
Extract the chat's segmented bar into lib/ui/adaptive (SegmentedBar,
GlassPill, SegmentedBarOverlay) and use it for the shell bar, DetailPane
and every page-level bar. Pieces are frosted glass over a fading scrim;
only pills take taps. Actions keep their natural width so the title gives
way before a model name. The chat transcript scrolls behind its header.
The open conversation draws its own bar of separate rounded pieces: the
platform Back in a circle, a title pill with the folder that opens the
chat information, and a pill with the model and mode controls. Detail
pages drop only their app bar for it, keeping the body mounted.
Operator-attached images render as rounded thumbnails above the user
bubble and open a zoomable Material/Cupertino viewer. Inline data: URIs
decode locally; file:// attachment-store URIs load through the new
chat_attachment_image websocket request, which serves only regular image
files directly inside the private attachment store (os.Root, <= 4 MiB,
sniffed raster types). Missing files show an 'Image no longer available'
placeholder; other files keep the file row.
Agent replies sit on the page without a bubble and get a copy control;
the operator's turn is a neutral rounded bubble; the composer is one
filled pill with +, a borderless field and a round Send, and its
placeholder names the chat's folder. Also fixes two picker tests left
stale by the native picker.
ChatScreen keeps a bounded in-memory cache of recently left transcripts
(16 chats / 4000 messages, LRU). A reopened chat is seeded from it with no
loading state, still re-subscribes and re-reads, and the newest page replaces
the cached copy while live frames received during the read are kept. /close
removes the entry; only transcripts backed by a successful read on the
current connection are written.
Adds tool/chat_reopen_probe.mjs and an optional history delay in the loopback
fixture. Measurement shows a second cost the cache does not remove: opening at
the end of the top-anchored transcript lays out every row (notes).
Transcripts pin to the newest end while following, release on the operator's upward scroll, and offer a native Jump to latest control carrying the count of messages that arrived meanwhile. The chat destination keeps each conversation's saved position so reopening restores it.
The conversation reads the session's recorded chat_shells snapshot and folds
pushed shell frames, so only tasks the server still reports as running appear.
The list is capped at three visible rows and scrolls beyond, and a task that
exits leaves it.
Selecting a task opens its recorded command, identity and running status plus
one bounded chat_shell_output page (cursor 0, 64 KiB). Empty, unavailable and
refused reads are reported with native Retry/Refresh controls, and a refused
snapshot is shown with its reason instead of claiming there are no tasks.
Both reads carry the session id, linking every task to its conversation.
Follow the web chat layout on both hosts without changing request shapes or
bubble/composer behaviour.
- Session rows (Material and Cupertino): title with trailing relative
recency, muted `directory · model` line with the `New`/`Closed` word at
the trailing edge. Shared row content keeps one arrangement; native tiles
keep their own padding, separators and subtitle size. The state word keeps
its intrinsic width and the metadata ellipsises first, so it is never
clipped at 320px/2x.
- Thread header: one row of compact icon-only Back, ellipsised title,
bounded model control and Chat information. Back keeps the destination as
its accessible name and both hosts share `Key('chat-back')`.
- Share the epoch-millis parser as `chatSessionMillis`, and match the web's
relative-time words, including `Sept`.
- Cover the row content, status word at narrow widths, header layout and
Back accessibility in widget tests; update existing tests for the
separate trailing status word.
Native model control and chooser in the chat thread header, matching the web
chat_models/chat_model exchange. Fixes the chat_models list parsing, keeps the
accepted server echo across Back/reopen, and resolves the session row for
/chat/:id deep links.
Show dismissible notices for refused history reads, refused live watches and a running compaction. A dismissal lasts only while the recorded condition is unchanged; a syncToken gap hides pre-gap notices until fresh state arrives.
The chat thread header gains an Info action that opens a platform sheet (Material bottom sheet, Cupertino popup) with the four harness-reported usage facts in the web order and wording: rate, tokens, context and reported cost. Unknown facts read Unavailable; reported zeros stay zero. A live usage frame replaces an open panel without adding messages or touching turn state.
The summary line now carries the host's own disclosure control and, when expanded, the recorded lifecycle, input, output, error, duration and edit diff. Each absent field keeps an exact sentence; no value is inferred from answer text.
Lift recorded activity/compaction parts out of an assistant turn into a
single activity row before the answer. Running kinds use the web's phase
words; terminal records keep the label the server recorded.