Review-round hardening of canonical admission:
- passthrough now requires a single-frame source free of EXIF/XMP/IPTC
metadata, so location/device metadata never enters durable storage and
stored dimensions always describe the perceived pixels; animated WebP joins
GIF on the always-re-encode path (first frame only)
- SourceImageInfo records orientation-applied dimensions, keeping source and
stored raster on shared axes for coordinate mapping
- validateImage runs a canonical-encoding dry run, so a validated batch can no
longer be refused mid-write by the byte target (no partial writes)
- read_image names per-axis multipliers when rounding splits the two ratios
and maps IMAGE_TOO_LARGE to actionable downscale guidance
READMEs (both languages) describe the wide source envelope, the canonical
encoding and its fixed encoder parameters, and read_image's downscale
envelope; tool/config catalogs regenerate for the new schema and Config
fields. The read-image-dimension scenario now pins the acceptance the old
2000px admission cap refused: the 2001x1 source is admitted and stored
byte-identically, so the fixture stays platform-independent.
The canonical ref serializes mediaType, width, height, bytes in the order the
pre-canonicalization store used, so existing session-log fixtures and logged
histories keep byte-identical reference JSON.
When the attachment store's canonical encoding shrinks the file on disk, the
read_image envelope names the original dimensions and the multiplier that
maps coordinates measured on the attached image back onto the file, and the
output schema carries sourceWidth/sourceHeight for programmatic callers.
Admission now validates a wide source envelope (32MiB, 100MP, 16384px per
side) and persists a canonical encoding instead of refusing large sources:
EXIF orientation baked in, metadata stripped, long edge downscaled to the
configured canonical target (default 2048px), PNG palette for alpha/PNG/GIF
sources and a fixed JPEG quality ladder (85/75/60/45) until the canonical
byte target holds (default 1MiB). In-budget PNG/JPEG/WebP passes through
byte-identically so equal sources keep deduplicating to the same content
address; GIF always re-encodes to the PNG of its first frame, pinning the
first-frame meaning providers apply. Encoder parameters are fixed by design;
only the canonical budget is deployment configuration.
AttachmentStore.saveImage now resolves SavedImageAttachment: the durable
reference paired with the submitted raster's intrinsic facts, so a store may
persist a canonical re-encoding while callers keep the source dimensions for
coordinate mapping. saveImages keeps returning refs; every fake store and the
cordis API catalog follow the new signature.
Every image in history is base64-inlined into every request, so a long
session's request body grows until a gateway request-size cap rejects it
with 413 and every retry resends the same oversized body, permanently
wedging the session. Each provider route now carries maxRequestImageBytes
(default 24MiB): when the accumulated base64 image payload exceeds it,
the oldest images are replaced by a fixed model-facing placeholder until
the request fits, so the newest images survive and the session keeps
completing requests. 413 and request-body-cap wording now classify as
INVALID_REQUEST instead of the generic PI_AI_ERROR.
Fixes#2644
An image with a side above the deployed routes' 2000px many-image bound
could be durably committed by read_image, ride every later request, and
permanently fail the session with provider 400s. Admission now enforces a
configurable maxImageDimension (default 2000) during the full decode, so
read_image surfaces a recoverable tool error naming the limit instead of
poisoning durable history; the Web composer gets dedicated copy for the
new IMAGE_DIMENSION_TOO_LARGE reason.
Fixes#2626
master introduced AttachmentStore.saveImages as the batch admission
(count/aggregate-byte/media-type limits, validate-all-before-save,
ordered commit). admitEncodedImages narrows to the shared wire entry:
canonical-base64 enforcement plus delegation to saveImages, keeping one
home for batch policy while both wire endpoints (prompt RPC and the
command executor) still call one function. Test doubles gain saveImages;
batch-limit error texts follow saveImages' wording.
- Honor a cancellation that lands during image admission before the
handler runs, settling command/done with the abort reason (executor
re-check after admitEncodedImages; the committed objects stay
unreferenced, deferred-GC territory, now recorded in the Agent Note).
- Never let a pending draft-image serialization reach claim.submit after
the attempt died (dispose/session teardown race).
- Refuse image removal while a command submit is in flight so the rail
cannot diverge from the serialized snapshot mid-transaction.
- Declare dsh-llm as a runtime peer dependency of command-goal.
- Mirror the host executor's ordering and the producer grammar
rejections in the fixture command plane: image checks run after
command resolution (unknown names stay lifecycle-free), bare /goal and
/plan//plan off with images answer the producers' error texts.
- Explain the deliberate serialize/release asymmetry in the hub's
commandImages plumbing.
A claimed slash command consumed only the text half of the composer
submission: /goal with reference images executed, cleared the draft, and
silently stranded the images in the rail. Model-visible attachment intent
had no route through the command plane.
The submission envelope is now modeled end to end. CommandDefinition
input.images declares acceptance; the declaration rides the descriptor to
every client, onto the minted CommandClaim, and into the input machine's
claim snapshot. commands.execute carries the submission's base64 images
and enforces the declaration in the executor: non-declaring commands, a
missing attachment store, and exceeded batch limits settle as logged
error results before the handler runs. Admission reuses the attachment
package's new admitEncodedImages, extracted from api-proxy's prompt path
so both wire endpoints share one limits/validation/commit sequence.
Producers own model visibility: /goal submits one user followup (image
blocks + a fixed reference line) after a successful create/edit so goal
rounds read the images from session history; /plan folds them into its
steered message. Grammar misfits (/goal pause, bare /plan, /plan off)
return direct errors and the composer keeps the images.
On the client, enter adjudication carries a SubmitEnvelope and every
command route that cannot consume images throws a localized refusal that
renders as one composer notice with draft and images retained; the
claimed pre-gate applies the same copy. An accepting claim serializes the
draft images, forwards them to commands.execute, and clears plus releases
them only on a success outcome.
The assembled web test roster gains the ui-input-trigger and ui-commands
plugins, mirroring the shipped composition, so slash submissions exercise
the command plane; a new keyless snapshot pins the refusal banner and the
accepting /goal flow over the built client graph.
- wire the directory pick's continue flag through InsertTextRequest to the
input shell, which re-tracks at the caret so descent reopens completion
(the flag was produced and forwarded but never consumed)
- guard the image-only send against a second Enter during the Host round-trip
- keep text appended after the sent snapshot when a submit settles ok; only
interleaved edits clear with the committed content
- drop the dead restoreImages left from the sink rewrite
- read recall labels through a shared defensive sessionRecallLabels helper
instead of unchecked casts over durable log data, and take the reference
summary separator from the locale dictionary
- align the composer specs merged from master with the transactional submit
contract (sinks resolve SubmitOutcome, settlement is awaited, call
assertions carry the AbortSignal), and restore master's forms where the
divergence served nothing (queue-mode sink case, single-line filter
expectation, component-identity slot lookup)
Replace the legacy reference.* API Proxy domain with @Remote methods on the
owning services, following the typert gateway design master adopted on
2026-08-02 (message-feedback and plugin-inventory precedents):
- FileReferenceService and SessionReferenceResolver extend TypertRemoteService;
fileReferences/list and sessionReferenceResolver/candidates are unary Remote
methods cancelled through the reserved trailing signal, and the candidates
face attaches each candidate's canonical mention under the configured limit
- move the wire types to type-only ./types subpaths (FileReferenceCandidate,
SessionReferenceMentionCandidate) and export ./typert plus ./remote artifacts
- mount both contributions in the api-remotes client assembly; ui-reference
consumes ctx.remote instead of connection.api.references and registers zh/en
locale dictionaries for its sections and labels
- delete the reference.* routes, schemas, map rows, client stubs, and fixtures;
the connection fixture serves the Remote endpoints instead
- release deliverPrompt admission listeners when the agent is disposed with the
prepared prompt still pending, and cover the reference-* RpcError codes in
the schema spec
- add the missing tsconfig paths for the /grammar and /types subpaths (clean-
tree vitest could not resolve @deepseek-ai/dsh-file-reference/grammar)
- regenerate the cordis catalog, capability seams, and event matrix; update the
owning bilingual READMEs, Agent Notes, and the reference-composer golden
Anchor the max-tokens notice between the closing assistant and the
turn-tail so the tail stays the turn's last chat node and its branch
action survives; forward the ChatNodeDataMap augmentation through the
client entry so built declarations publish the new kind; advance the
fixture's nextTurn past the appended history; drop the engine-unreachable
buildViewNode fallback; deduplicate the runtime README tail sentence.
A turn the provider ends at its per-request output cap previously produced
no visible sign in the Web chat flow: only error-kind turn/end events built
a conversation node, so the truncated answer read as a normal completion.
Add a turn-max-tokens conversation node Definition over the durable
turn/end event, a warning-styled localized notice row distinct from
turn-error, a fixture max-tokens sample turn, and an assembled keyless
snapshot pinning the dot state, title, and hint.
Closes#1522
Second alignment step for #2248: document-level drag intake behind the new
DropOverlay atom, lightbox close icon and shared dialog mask, DeepSeek Chat
thumbnail rules (single 240px long edge with ratio clamp, 64px tiles, merged
consecutive assistant images), image limits raised to 20/10MiB/100MiB and
published to clients as the imageLimits projection, whole-batch intake
pre-check with product-copy banners, and attachment-error reasons mapped to
localized copy.
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.
creatixchu