The toast sits 120px from the viewport top and centers over its anchor —
the composer card, so it centers on the chat column rather than the window;
a rejected model selection (e.g. picking a text-only model while the
session holds images) announces through the same banner while the in-menu
strip with Retry stays the catalog-load surface. The attachment rail
consumes every wheel tick with a vertical component: a diagonal pan keeps
its horizontal intent and nothing scrolls the conversation behind the
composer.
Body-portal the lightbox and toast so transformed ancestors cannot trap
their fixed positioning (a lightbox opened from a chat message covered only
the chat column); make the toast pointer-transparent; observe the rail
element's own size instead of window resizes; consume vertical wheel ticks
exclusively via a non-passive listener with LINE/PAGE delta normalization;
keep the start position when the rail mounts over an existing draft; honor
prefers-reduced-motion for the toast, remove-control, and paging; retry
loads through the guarded load effect; note the deliberate promptError
re-announce; pin the intake toast in the assembled snapshot; sync the
superseded multimodal note and package docs.
Single-click original preview in the composer rail and chat history; remove
control inside the thumbnail, revealed on hover/focus (always on touch);
hidden-scrollbar rail overflow paged by edge arrows with wheel panning and
end-reveal on add; image-intake rejections and prompt failures announce as a
transient top-center toast instead of inline strips.
The attachment atoms move to a new zero-cordis package
@deepseek-ai/dsh-client-ui-attachment (rail, message gallery, lightbox),
seeded as a platform module; the toast is a ui-primitives atom. Strings
arrive as label props bridged from the conversation dictionary.
Review round: the JSONL backend now refuses a foreign header version straight
from the raw header line, before validating today's header shape or decoding
any event row, so a structurally different future format reports the upgrade
direction instead of corruption (shared message builder
sessionFormatVersionRefusal). HMR live-prefix adoption runs the unknown-type
guard like the other read paths. The appendCore comment now states why the
unknown-type guard is read-side only, the loadStoredFrom JSDoc and README pin
the seek-vs-sequential refusal-scope divergence, and the generated catalog
preamble lists the ignorable envelope field.
Old runtimes meeting a newer session format now fail loud instead of
misreading: version refusal names the direction (newer: upgrade the
harness; older: no upgrade path) and points at the raw JSONL log, and an
event type outside the generated known vocabulary refuses resume unless
its envelope carries the new ignorable: true marker (default: required,
so a forgotten marker over-refuses instead of silently resuming a gutted
session). gen-persistence-catalog now also emits
KNOWN_SESSION_EVENT_TYPES; SQLite stores the marker in a dedicated
column (SCHEMA_VERSION 15). The versioning design (monotonic integer,
n->n+1 upgrader chain, migrate-on-continue) is recorded in the
session-log-version-mechanism Agent Note.
Release the Web Stream reader after either completion or cancellation. Strengthen the oversized-file test so a regression that downloads content before applying the stat bound fails directly.
- E2BFileSystem.readBytes now short-circuits on the stat size before any
content transfer and streams the remote object, cancelling at the first
chunk past the cap, honoring the seam's bounded-buffering contract; the
fs-e2b README pair documents the new primitive.
- The tool-fs HMR test now proves the attachments-scoped registration:
disposing the store withdraws read_image while read/write/edit stay,
remounting restores it, and disposing the plugin withdraws everything.
- read_image caps reads at the smaller of maxImageBytes and
maxMessageImageBytes, records an absent observation for a missing
target like its sibling read, widens the mismatch remedy to cover
out-of-family formats, and renames the gate's parameter to
requestedPath; module/apply/registration JSDoc now match the shipped
composition. zh terminology aligned; the examples manifest keeps its
literal arrow.
The model reads a PNG/JPEG/WebP/GIF file, the bytes commit through the
durable attachment lifecycle, and the tool result carries the real
ImageBlock so the image enters context from the next request onward.
FileSystem gains a bounded readBytes primitive (local + E2B providers);
registration is conditional on the attachment store, and a strict
execution gate refuses routes that do not declare image input, so a
text route's durable history stays free of image blocks. llm-replay
models may declare inputModalities, letting keyless ACP snapshots pin
both the sha256-referenced success and the verbatim refusal.
Supersedes the withdrawn route-scoped design of PR #598; the decision
record is .agents/notes/implemented/feature/2026-08-10-minimal-read-image-tool.md.
Steering bubbles render as plain user bubbles; a mid-turn steer is
recognizable by its position in the flow. The runtime SteeringMessageNode
projection and pending-steering lifecycle are unchanged.
Partially supersedes the 2026-08-04 context-source and steer marks note;
the new simplification note owns the removal rationale.
The root-scripts workspace ignore existed only for
scripts/hero-composer-dom-continuity.mjs; apps/web declares its own
playwright dependency for the e2e lane.
Rewrite the three sibling Agent Note pairs that still described create-by-name
as current (workspace-ui-product-flow, session-list-browsing-and-manual-order,
same-basename-workspace-adoption) and the one-route note's own alternative and
section title; delete scripts/hero-composer-dom-continuity.mjs, which drove the
name dialog removed by the one-route change; mark WorkspaceRegistry.create's
now test-only title parameter with a deletion TODO; pin the retired { name }
spelling as a schema rejection; align the workspace spec on stageDir and the
fixture spec title on path creates.
The Web picker collapsed onto the directory flow (see the one-route-to-add-a-workspace
Agent Note), leaving workspace.create({ name }) with no product consumer. Delete the
whole feed line: the wire schema's name member and WorkspaceApi spelling, the gateway's
workspaceRoot config/default and the mkdir branch, the client seam that carried the name
(WorkspaceCreateInput, WorkspacesService.create, intentName), the dsh web
--workspace-root flag, and the fixture's name handling. workspace-name-conflict stays as
workspace.rename's duplicate-title error.
The branch control on user and consumed-steering bubbles could enable
only when a completed turn ended with no node after the message at all,
so readers saw a permanently unavailable control whose tooltip promised
a state it could not reach. An enabled one would mislead anyway: a fork
at a message seq cuts at the containing turn/end and takes the answer
along, the opposite of the branch-to-re-ask reading a control on one's
own bubble suggests.
MessageItem loses its fork props, PendingSteeringBubble loses the
showBranch special case, and messageBranchSeqs narrows to
assistantBranchSeqs: only a completed turn's transcript tail that is
the turn's own content-text assistant may fork. A steered turn keeps
its fork point under the settled answer, because fork is a log-prefix
cut and the steer is model-visible history the child inherits.
Web aria goldens drop the user-bubble disabled-branch row and its
hidden explanation text; the nested-subagent golden also loses the one
enabled user-tail fork handle, a loss the decision note accepts.
Correct the Agent Note's consequence: a running turn withholds the footer
below its own trigger bubble, while every earlier completed turn keeps its
seat — which the package test already asserts.
Give the running-phase barrier an explicit budget: it is armed before the
park and awaited after the stop click, so the 30s replay default left no
headroom for the marker poll, the UI polls, and two aria captures.
Number the running-turn test's boundary seqs like the log does, with each
turn/end strictly after its own nodes.
`assistantActionsSeqs` picked the last content-text assistant of each turn
from the finalized transcript alone. That quantity is stable only once the
turn closes: while a turn is still producing steps, the narration written
before a tool call is the last content assistant so far, so copy, branch,
and the clock appeared under an intermediate sentence for as long as the
tool ran and then moved down to the next step's text.
Pass `ConversationSnapshot.turnEnds` into the derivation and grant the row
only inside a turn that has a durable `turn/end`. This is the same
completion fact the branch control and the `Ran for` label already read, so
the three parts of one row now agree; mid-turn narration owns nothing, and
the seat appears once under the settled answer.
`hasContentText` moves to chat-flow.ts so the ownership gate and
AssistantMarkdown's mount gate cannot drift apart.
apps/web/tests/turn-tail-actions.e2e.ts pins both states through the
assembled application: a hang sidecar on the second model call parks a turn
whose first step narrated before calling bash, and the two goldens hold the
parked flow and the flow after stopping.
Review follow-ups on the column-overflow scenario:
- Memoize the sweep so the golden and the assertions consume the same
readings, which is what its contract already claimed; two runs could
disagree if a resize settled differently between them.
- Settle the column width before the mutation control measures. The test
arrives from 1680 alone and from the sweep's 600 in a full run, and the
frame eases its column tracks, so an immediate read can report the
previous viewport's bleed.
- Name the wheel delta, assert the bleed stays inside it, and compare the
travelled distance rounded: a clamp or a sub-pixel would otherwise read
as a broken fix.
The 0 the shipped column reports cannot be reached by polling for a settled
value, so the read is a fixed wait; make that wait cover a smooth-scroll
animation on any engine the lane runs on. Identical on both sides of the
mutation control, which is what keeps the 0 evidence rather than a race won.
The hero's backdrop ellipse is sized 1051/776 of the hero box so its blur
scales with the input card, which means it reaches past the column whenever
the column is narrower than the glow. `[data-conversation-scroll]` declared
only `overflow-y: auto`, and a box that scrolls in one axis computes the
other axis's initial `visible` to `auto` — so that bleed came back as a real
horizontal scrollbar, 24–95px of range across ordinary laptop widths.
Declare `overflow-x: hidden` on the column instead of leaving the second axis
to be derived. Clipping is unchanged (the box already clipped both axes); the
declaration withdraws only the bar and the user gesture.
- `relay` resolves its sender in `contextBody` like every other form. It was
the one shape whose marker could claim a form the body did not render: an
unreadable sender fell back inside the body while the row still said relay,
contradicting the contract this PR's own note states.
- `recall` requires the retained, omitted, and truncated fields. Completeness
is what the card exists to report, so a reference that cannot state it is
not a readable recall — showing the label alone presents a confident card
over unknown loss.
- The snapshot body states the supersession its producer framing line carries.
That line is the one part of the model-facing text no section contains, and
unlike an instruction context's `<system-reminder>` it states the form's own
semantics rather than wrapping content.
- `GoalMessageSource` is a discriminated pair, so `{ form: 'notice' }` without
its account no longer compiles. The guarantee this PR claims now holds at
that seam too, not only through `ContextFormed` on plugin sources.
- Goal and tool-goal summaries are bounded by a shared `boundContextSummary`,
which tool-tasks now uses as well. A goal objective is unbounded caller text
in exactly the way a task label is.
- The runtime snapshot interpolates once per request: agent-loop renders the
sections and joins them through `joinContextSections`.
- Every form's fallback branch is pinned, not only the notice one.
The rebase carried the simple-interface block from the base branch, while
this branch turned ContextFormed into a per-form discriminated union;
ContextSnapshotSection gets its own block so the page stays self-contained.
The row header already names the producer, so `goalChangeSummary` dropped its
redundant `goal ` prefix; the ACP and headless goal scenarios still pinned the
old string.
The e2e lane pinned the pre-form source shape, so it failed on CI while the
unit lane passed; the reading is now a `snapshot`-form context carrying one
named contribution.
Four values complete the vocabulary, so the opaque body is reached only by
producers that genuinely promise no shape.
`snapshot` — current state a later snapshot supersedes. system-prompt now
exposes `renderContextSections()`, the named contributions
`renderContextSnapshot()` already joins for the model, so the body attributes
each part to the subsystem that produced it instead of re-splitting joined
prose. The runtime snapshot, time-context, and tmux-context declare it.
`notice` — a one-off account of what just happened, declared by tool-tasks,
goal state changes, tool-goal wrap-up, plan-mode switches, and
repeat-tool-guard. Its `summary` rides the COLLAPSED row: these five are the
majority of shipped producers and none of them needs expanding to be read.
The task summary bounds itself because its inputs are unbounded caller text.
`relay` — a message another agent addressed to this one; both subagent
sources declare it and the body names the sender above what it said.
`recall` — material lifted from another session's log. session-reference
needed no new field: its references already record retained and omitted
counts and the truncation flag, which the body shows first, because recalled
context is bounded on the way in.
`ContextFormed` is now discriminated by `form`, so a producer cannot declare
a shape without the facts that shape is presented from — a notice without its
summary, or a snapshot without its sections, fails to compile.
Only the two hook bridges stay opaque, by design: their content is whatever
an external program printed, so no shape can be promised for it. Unknown
kinds and unreadable records land there too.
- Content blocks render in the order the model received them. Partitioning
hoisted every unknown block past the text around it, so an interleaved log
read back in an order the model never saw.
- A delta distinguishes a newly reconciled file from a rewritten one; `set`
and `replace` already separate them at the producer, and collapsing both to
"updated" misread a new file.
- The superseded note states current fact in its consequences and testing
rather than keeping claims the implementation now contradicts, per
implemented/AGENTS.md; the decision itself stays as the record of that
change, with the supersession pointer above it.
- An empty replacement catalog is a real catalog: `renderCatalogUpdate()`
publishes zero entries when the last skill disappears, and falling back
would hide that every earlier name was retired.
- The opaque fallback keeps a `form` declaration this version cannot present.
It is the one place a newer or foreign log's declared shape would otherwise
vanish from the UI entirely, since the row marker is also absent there.
- An instruction change with an unrecognized `action` disqualifies the record.
The action decides the word the row shows, so an unknown one would be
presented as loaded or updated.
- The catalog list bounds itself and reports the withheld count. Entry count
is unbounded and the scrollport bounds height, not node count.
- A catalog message keeps content blocks this version does not know, instead
of dropping model-visible content the extensible union may carry.
- `core.md` defines `ContextFormed`, the interface actually carrying the
optional field, beside `ContextForm`.
- The superseded-in-part bullet states the affected clauses as one rule rather
than enumerating them; two rounds of enumeration each missed some, which is
the shape being fragile rather than the list being wrong.
- The note records the one migration case that does not self-heal: an
old-format catalog as the only one, with an empty current view, leaves a
stale catalog nothing replaces.
The opaque and instructions bodies rendered the same text-plus-unknown-block
pair, which the duplication gate flagged as a clone. Both now use one
ModelFacingContent component; the CI lint gate (lint + duplication) passes.
- `catalogHistory` validated its durable read. `agent.session.events` is a
JSONL/SQLite seed on resume or fork, and seed validation guarantees only a
source object with a non-empty `kind`; a `skill-catalog` record with missing
or wrongly shaped `entries` threw inside the step listener, failing every
later turn of that session. It is now skipped as an unrecognizable record,
the posture the replaced content digest had, with a regression test over six
malformed shapes.
- The headless keyless smoke still filtered catalogs by the old plugin source,
so the `built-bin-smoke` gate would not have found the catalog message.
- Entries record the published description unescaped. The pseudo-XML escaping
belongs to the `<available_skills>` frame and is applied when rendering it,
so a description containing `<` no longer reaches the card as `<`.
`escapeText` is injective, so republish semantics and the model-facing text
are unchanged.
- Adjacent text blocks join with no separator, matching how provider adapters
flatten them; the body no longer shows a line break the model never saw.
- Provenance fields are bounded like the text: an unknown producer may record
an arbitrarily large value.
- Both readers are all-or-nothing, and the row's form marker reports what
rendered rather than what was declared, so a partly unreadable record cannot
present a confident but incomplete account.
- The catalog body consumes `update` as a replacement notice; the digest
canonicalizes per entry as JSON, since every separator character is itself
legal in a description.
- `core.md` documents the form axis with a `ContextForm` type-equiv block, and
both projections assert the wiring they duplicate.
Injected context reached the transcript as one anonymous shape whose
expanded body was the whole message serialized as inline JSON, so the
escaping collapsed the only readable part — the model-facing prose —
into a single line.
`MessageSource` gains an optional producer-declared `form`, a small
semantic vocabulary of information shapes independent of `kind`: kind
says who produced the context, form says what shape it is, so several
producers may share one presentation. Two values ship.
`instructions` (workspace-context) lists the reconciled files above
their text and keeps the `<system-reminder>` framing verbatim, because
the framing is part of what the model read. `catalog` moves dsh-tool-skill
off the shared plugin kind onto a `skill-catalog` source carrying the
published name/description entries, and the body lists those instead of
re-parsing `<available_skills>` out of the prose. Catalog identity moves
with it: the republish digest now covers the durable entries, deleting
the text-slicing that recovered them from a logged message.
Everything else renders the opaque body — the model-facing text with its
real line breaks, then the remaining provenance as fields. That is the
documented default, not a leftover: a resumed, forked, or foreign log
must render whether or not its producer is mounted here, which is why
the classification lives in the durable source rather than a client-side
table keyed by producer.
Review follow-ups on the context provenance row:
- the producer name now follows ToolRow's aria-hidden separator dot instead
of a literal middot inside the text, so the two disclosure rows expose one
accessible-name shape;
- the English steering caption becomes the noun `Interjection`, parallel to
the Chinese 插话 and to the label this bubble carried before;
- the history fold gets its own provenance assertion, so the mapping it
declares separately from TranscriptAdapter is pinned on both sides;
- the superseded-in-part bullet on the no-interjection-chrome note now
enumerates the third Decision bullet too, whose no-badge and no-IconActions
clauses had both lapsed.