dsh-token-counter

A global DSH plugin that shows a realtime token counter right after the blue "深度思考 / 深度求索" (deep-thinking) indicator, plus per-operation token usage after each completed step. It is global: once enabled, every conversation gets it.

Features

1. Live in / out token counter — after the deep-thinking indicator, a "华光划过" (shimmer sweep) badge shows the streaming token count, painted with the platform's own shimmer so it reads as part of the same sentence.
   • in N token = input tokens, out N token = output tokens.
   • Unit scaling (auto by default) — numbers render as K (1e3), M (1e6), G (1e9) or T (1e12) by magnitude (1866 → 1.9K, 2 500 000 → 2.5M); below 1000 the raw number shows. A rounding that would print 1000K promotes to 1M instead. Each site (in / out / use) can instead LOCK one unit in its gear menu and then always scales by it.
   • The input count is the number the harness itself prices the prompt with (its contextPressure projection), so it is live during the input phase and does not wait for the provider's usage. When the provider does report inputTokens, that authoritative figure takes precedence; when neither is available yet, the last known input is reused.
   • The streaming output count is a client-side estimate (CJK glyph ≈ 1 token, latin ≈ 4 chars/token), then snaps to the authoritative value when the stream settles.
2. Display mode (configurable) — merged (default) or split.
   • merged: the label swaps rather than stacking — only in … during the input phase, only out … while generating (the phase flips as soon as the first output token streams).
   • split: always shows in … · out … together, for the whole turn.
3. In / out visibility (configurable) — independent switches to show/hide the input and output counters (both on by default). Each of the three switches carries a gear button: clicking it drops open a per-site editor with font size (0/empty = follow the interface text), color (a native picker plus a "follow theme" reset) and unit (auto / K / M / G / T). Settings are staged in the same form and saved together with everything else.
4. Per-operation "use N token" (configurable) — after each completed operation ("已完成思考", "已完成命令", …) a use N token label appears, styled to match that row, showing the tokens that operation's own step consumed (input + output, or the provider's totalTokens when it reports one) — never the whole reply's total. One switch toggles it; its own gear menu styles it.
5. All options are configurable from four places: 设置 → Token 计数器 (Settings → Token Counter), the same tab under 设置 → 内置插件, the configuration section on this bundle's page in the 插件 / Plugins panel, and the card in that panel's 官方 / Official group.
6. Visual settings UI — a light white/grey, adaptive two-pane layout: merged and split each get a preview card that recreates the real transcript scene (a "Command finished · use N token" row and the deep-thinking badge with the shimmer sweep, all with fake data). Hovering a card animates the numbers (in climbs to its cap, pauses, then out climbs); clicking it switches the display mode; and the in / out / use toggles update both cards live — turn in off and both cards drop the in segment immediately. The gear menus' unit/font/color choices are reflected in the preview cards before saving.

Install

# 1. pack the plugin
cd packages/dsh-token-counter
pnpm pack

# 2. install into the desktop profile
dsh plugin --profile desktop add ./dsh-token-counter-<version>.tgz
# (or manually: add the tarball to the profile's package.json dependencies and
#  append "dsh-token-counter" to dsh.profile.bundles, then `pnpm install`)

# 3. restart DSH

Usage

1. Open any conversation and send a message.
2. While the model is thinking/streaming, the counter appears after the "深度思考" indicator and counts up in real time.
3. When the answer completes, each "已完成…" operation row shows use N token for that operation's own step.
4. Open 设置 → Token 计数器: hover a preview card to watch the numbers animate, click it to switch merged/split, and use the in / output / per-op toggles — the cards update in lockstep. Save when you are done. The same form is also reachable as a tab under 设置 → 内置插件, as the configuration section on this bundle's page in the 插件 / Plugins panel ("dsh-token-counter" → 配置), and as the "Token 计数器" card in that panel's "官方 / Official" group. If the settings form reports "This plugin is not loaded, so it cannot be configured right now", the host half is not mounted — run dsh --profile desktop --dump-config (or check the profile's cordis.patch.yml) before filing an issue.

How it works

• The browser half subscribes to the DSH client sessions service and the active (main-view) session's binding.eventSource.
• It counts text-delta / reasoning-delta stream chunks as they arrive, reads the live input count from the harness's contextPressure projection, and records authoritative usage per step from the usage chunk / assistant/message event (keyed by turn + step).
• It injects a DOM badge into the running indicator (there is no slot around it). Each completed operation row is one step, so rows walked in document order claim the next settled step of their turn and show that step's own tokens.

Package layout

• lib/index.js — host half: schemastery Config (the configurable options) + apply.
• lib/client.js — browser half via window.__ModuleLoader__.load.
• cordis.patch.yml — registers the plugin profile entry.
• package.json — dsh.client = { platform: "web", inject: [...] }.

Develop / test

node test/run-tests.mjs   # mocked-browser unit tests (all features)