DeepSeek Harness 插件

dsh-strata

滚动条槽位里的会话缩略图:按事件类型着色的等比色块、用户与失败锚点、滚到顶自动补载历史、悬停线索墙铺开全部发言,附近/中/全三档地图尺度切换。

跳到安装方式

来源信息

GitHub 仓库
jsdvjx/dsh-strata
最近更新
2026年8月19日
分类
界面增强
GitHub stars
0

安装

默认先复制一段 Prompt,让 Agent 读页面和仓库;需要自己装时再切到命令。

复制这段 Prompt,发给 DSH、Codex 或其他 Agent,让它先读页面和仓库。

请先不要安装。阅读这个 DeepSeek Harness 插件,说明它解决什么问题、会访问哪些文件、网络或密钥,以及如何安装和卸载。

插件页面:https://deepseekplugins.org/zh/plugins/jsdvjx/dsh-strata
GitHub:https://github.com/jsdvjx/dsh-strata
插件名:dsh-strata
作者:jsdvjx
安装命令:dsh plugin --profile web add github:jsdvjx/dsh-strata

确认前不要执行安装命令。

检查来源文件

安装前先看这个插件目录里的 README 和其他文件。

文件资源管理器3 个文件
README.md来源说明 · 只读预览

dsh-strata

> Read the strata of a run.

English | [中文](README.zh.md)

A real minimap of the session trajectory for the DeepSeek Harness Web GUI — it takes the transcript's scrollbar seat and turns it into a scaled, colored picture of the whole loaded conversation, with your own messages emphasised and clickable anchors beside it.

Not a tick rail. Every other conversation navigator draws one evenly-spaced dot per prompt, which throws away the only thing a long agent session has too much of: volume. Here a band's height is the row's real rendered height, so the map is a proportional compression of the scroll extent — a 300-line answer looks like 300 lines, forty tool calls look like forty tool calls, and the viewport lens maps 1:1 onto the scrollbar. What you get is the shape of the run at a glance: where you spoke, how much work each prompt cost, and where it went wrong.

<p align="center"> <img src="https://raw.githubusercontent.com/jsdvjx/dsh-strata/main/docs/demo.gif" alt="Live demo: hovering expands the rail with a preview card, clicking a band or an anchor dot jumps, dragging the lens scrubs the whole session" width="840"> </p>

What the map shows

BandMeaning
Full-width blueyour message (and steering) — always the widest, brightest, never thinner than 5px
Grey blocka model reply; block height is how much it wrote
Thin grey tickone tool call
Greena slash command
Ambera model retry, or a turn cut short by the output cap
Reda failed turn — or any tool/command row that reported an error
Horizontal rulea compaction checkpoint: where the model stopped seeing the history above
Rounded outlinethe viewport lens (drag it)

The ragged left edge is the index: every blue bar is a turn you started, and the block of agent work under it is what that turn cost.

It replaces the scrollbar

The rail sits in the transcript's own scrollbar gutter and the native thumb is suppressed while it is up, so there is one scroll control, not two — no layout shifts, because the gutter stays reserved either way. It is a takeover, not a theft: the moment the map stands down (Trajectory tab, no session, a transcript that does not scroll) the native scrollbar comes straight back, and uninstalling restores it permanently.

Anchors

Beside the rail is a column of clickable anchor dots — blue for every message you sent, red for every failed tool call or command. Click one to jump there. The dot for wherever you are reading stays enlarged, so the anchors double as a position indicator. Anchors that would collide collapse to keep the column readable; failures never collapse, since they are usually the reason you reached for the map.

The clue wall

Hovering a user anchor dot floats the clue wall over the right half of the screen: every user message in the session — including the ones above the loaded window (read from the session's own export log) — packed like an evidence board: every card takes exactly the size its content needs (a one-word prompt is a small slip, a long one a tall note), flowing into masonry columns so the whole session fits on one board whenever it can. Each card is tied to its anchor dot by a bezier that tracks live as the transcript scrolls; unloaded prompts get a dashed border, tag and dashed string, and the message you came from is spotlighted. Only when the board truly cannot hold everything do ↑/↓ pager buttons appear. Click a card to jump — an unloaded one chain-loads the missing history first. The wall retires when the pointer leaves it (grace period), or instantly on Esc; clicking a dot jumps straight to that message.

Map scale

A small 近 / 中 / 全 switcher sits under the rail while it is awake:

  • (initial) restores the view from when the session opened — however

much history has been loaded since, the map keeps that scale and slides with your reading position;

  • (medium) spans twice the initial view, loading the difference on

demand;

  • (full) maps everything — pulling in any unloaded history first,

with the same rail progress bar the unloaded jump uses.

The choice persists per browser. Zoomed scales are pinned: loading more history shifts the window, never the scale.

Use

  • Click an anchor dot to jump to that message or failure.
  • Hover the rail — it widens and shows a preview card for the row under the cursor

(kind, n/total for your own messages, and the row's text).

  • Click a band to scroll it into reading position; the row flashes when it lands.
  • Click empty track or drag to scrub proportionally, like a scrollbar.
  • Wheel over the rail or the dots to scroll the transcript.
  • Keyboard: the rail is focusable — arrows nudge, PageUp/PageDown page,

Home/End jump to the ends (role=scrollbar with a live aria-valuenow).

  • Double-click to pin the rail open (persisted per browser).
  • Older history loads by itself: scroll (or drag the lens) to the top and the transcript's

own load older fires, the map rescaling as history arrives. The faint above the rail just says there is more above; it retires once everything is loaded.

The map hides itself when there is nothing to navigate: no session, a transcript that does not scroll, or a non-chat view such as Trajectory.

Install

dsh plugin --profile web add dsh-strata

Or straight from the repo:

dsh plugin --profile web add "github:jsdvjx/dsh-strata#main"

Then restart dsh web. To remove:

dsh plugin --profile web remove dsh-strata

How it works

Pure browser half; the Node half is empty and no session data crosses the wire for it. Geometry and semantics both come from anchors the conversation view already publishes — [data-conversation-scroll] for the scrollport, [data-chat-anchor-key] per flow row, data-chat-flow-kind for that row's registered Chat Node kind, data-state="error" for a failed tool or command, [data-composer-seat] to stay clear of the sticky composer. It contributes one entry to the frame-wide shell.overlay list slot, so it adds a surface instead of replacing one, and uninstalling leaves the native UI untouched. The scrollbar takeover uses the theme's documented seam — rebinding --dsh-scrollbar-thumb to transparent on the scrollport, the same mechanism ui-sidebar uses — so both the WebKit and Firefox rendering paths are covered and no stylesheet is overridden.

Rendering is a canvas repainted on a rAF, re-measuring rows only when the transcript mutates or its scroll height changes — a plain scroll moves the lens and nothing else, so a streaming turn does not drag layout through the map. Colors are read from the theme's own --dsw-alias-* tokens, so light and dark both come out right, and prefers-reduced-motion disables the transitions.

Limits

  • The map covers the loaded window. DSH pages older history in on demand; until it is

loaded it has no layout to map. The indicator is the honest signal that more exists; reaching the top pulls it in automatically.

  • Chat view only. The Trajectory tab renders its own event ledger with different

anchors; the map stands down there rather than guessing.

  • The rail occupies a ~14px strip of the transcript's right padding, so clicks in that

strip go to the map.

License

MIT