deepseek-harness/packages/client/ui-chat/README.md

---

description: "Browser Chat target that renders Session conversation nodes, historical images, actions, localization, and scroll state."

kind: "package-reference"

@deepseek-ai/dsh-client-ui-chat

English | 中文

Summary

Use this package to render a browser chat from recorded Session conversations, including historical images, localized actions, and restored scroll position. Compact display folds completed-turn process rows while keeping the final answer and independently useful context visible; packed historical Assistant runs remain collapsed. Local transcript and steering submissions appear immediately, remain in their original surface, and disappear atomically when authoritative Session records arrive, while queued submissions stay outside Chat. The package does not assemble or modify model requests.

File-mention providers receive the viewed Session ID with the closing-turn owner, so links into inherited history can address the fork itself.

Table of Contents

• Reference previews
• System prompt row
• Turn token usage
• Completed-turn footer
• Turn Process Folding
• Scroll ownership
• Model Experience
• Known Limitations and Deferred Work
• Dev Note

---

Reference previews

Sent file references and skills confirmed by the message’s logged invocation open in the right Sidebar. File paths use the viewed Session; skill names resolve through its current input-trigger source. Both use the prose file-link dotted underline on hover or focus. Sessions, directories, and command labels remain non-navigating references.

System prompt row

Each nonempty appended system/message owns a collapsed prompt row, including a complete prompt at the start of a headerless window; the same-step header does not duplicate it. Chat also shows a collapsed System prompt row for a non-empty initial request, explicit message-series start, or system/message surface node replacement whose text differs, reading the last nonempty surviving system node in surface order at the request/header; a non-initial request whose preceding header is outside the loaded history window also shows one. A resume repeats the row even when its system text is unchanged, including after pagination supplies the preceding header and system node; same-series config-only or tool-only changes, tool steps, and retries create no repetition, and a system/message event is never rendered as a transcript message. The row appears before that request's user messages, matching the provider envelope, and expands to the exact model-visible text with its original line breaks. A request whose system node is empty or outside the loaded window creates no row until the page holding the node arrives.

Turn token usage

A completed Turn shows an expandable usage row only when the loaded window includes turn/start and every started model attempt reports safe, exact usage. The row omits unavailable optional buckets. Incomplete or contradictory accounting hides the complete disclosure instead of presenting a partial total.

Completed-turn footer

The completed-turn action footer starts 20px below the preceding prose or extension content.

---

Turn Process Folding

Settings → General exposes a persisted, localized Normal / Compact conversation-display preference in the ui-chat namespace; Compact is the default. Normal leaves process rows visible and renders no Turn-process control. In Compact mode, the System prompt remains independently visible before the opening User throughout the Turn. Context injection, reasoning, Assistant material, Tool rows, and Retry rows remain expanded while a Turn is open. At turn/end, its latest Step becomes the final-answer boundary only when it contains non-blank text, an image, or an unknown visible block—and no Tool-call block; preceding Context injection, reasoning, earlier Assistant material, Tool rows, and Retry rows then collapse by default. The control reports Turn-wide durable counts for non-subagent Tool calls, reply-bearing Assistant messages before the final answer, and subagent delegation calls; zero-valued segments are omitted, the Tool and subagent figures are mutually exclusive, and neither System prompt nor Context injection contributes a count. When all three counts are zero, the process still folds and the control reads Thought for a while. A full-width divider below the summary separates it from the answer or expanded process rows. User and steering messages, System prompt, error, max-token, and turn-tail rows stay outside, and a closed Turn with no final answer keeps all process evidence visible. A newly available process control is inserted without changing the relative order of existing rows: opening human input precedes the control and process rows from their first projection, while System prompt remains above that input. While older history remains available through Load earlier, process controls stay absent and no members are hidden; once history is complete, every eligible closed Turn uses the collapsed default immediately. Stable Chat Node Seats keep every renderer mounted, hidden members add no flow spacing, and a closed control sits 8px above its answer only when no independent input intervenes. Completion collapse does not depend on tail-follow position, so a reader above the tail may see the transcript reflow. An automatic collapse that would hide keyboard focus keeps the group open and leaves focus in place; a manual close focuses the process control before hiding its members. The session-scoped store records only manually expanded Turn-and-answer-Step generations; a different answer generation starts collapsed.

---

Scroll ownership

Chat restores semantic anchors across history prepend and renderer remounts. Pinned scroll deliveries without reader movement update follow ownership immediately, before subsequent layout changes can invalidate their floor. Reader movement remains pending until the sampling interval or scrollend, even inside the follow threshold, so layout growth cannot erase small scroll gestures. While the reader is pinned to the floor, ResizeObserver follows the new floor and selects the latest loaded Turn without reading row geometry. Once the reader moves away, flow-height changes preserve the top position and the reading-line geometry selects the active Turn. Turn-rail previews paint above sticky Markdown code-block banners, while the rail frame remains inside the transcript band above the composer.

---

Model Experience

None, as this package renders logged conversation state in the browser and registers nothing model-facing.

KV Cache effect

None; Chat presentation does not assemble or mutate provider requests.

Known Limitations and Deferred Work

• The transcript reflects the loaded Session window — older transcript nodes become available only after Session Controller loads the preceding event page. Turn navigation is wider than the window: the rail merges the loaded Turns with the host turnOutline projection, so every started Turn gets a fixed-pitch mark (10px apart; a ladder taller than the frame scrolls inside it with gradient fades), and activating an unloaded mark pages history through the Turn's turn/start seq before landing on its row. Without the projection (assemblies not mounting dsh-session-turn-outline) the rail falls back to loaded Turns only.
• Rail previews are card-sized — one prompt line (50 characters) and up to three response lines (120), on loaded and unloaded Turns alike; an unloaded Turn's response arrives from the outline only once the Turn settled, so an open Turn previews its prompt (or just the Turn number) until then.

Dev Note

Working context for maintainers — click to expand

None.

Runtime invariant: No companion is published. Conversation and Slot registration enforce Chat target consistency.