dsh-tokenlens 1.7.5 → 1.7.6

package/CHANGELOG.md

@@ -4,6 +4,18 @@
 
 **版本记录约定**：同一发布日只保留一个章节，标题取当日**最后一个**已发布版本号；当日多次发布的改动合并到该章节，并在标题下注明当日实际发布过的版本号（npm 上仍可安装这些独立版本）。若当日已有发布，后续小改动不再单独增加版本号，而是并入下一篇日志。这样日志按天规整、不出现同日多个小版本。
 
+## 1.7.6 — 2026-11-06
+
+（当日已发布版本：v1.6.5 / v1.6.6 / v1.6.7 / v1.7.0 / v1.7.1 / v1.7.2 / v1.7.3 / v1.7.4 / v1.7.5 / v1.7.6）
+
+- **feat(ui): 版本号左侧新增仓库链接**——弹窗底部一行现在是 `[GitHub] 插件 v1.7.6 · 已是最新版本`，链接指向 `github.com/senyingxidi/dsh-tokenlens`，`target=_blank` + `rel=noopener`，带 `title` 说明；图标沿用同一套 stroke 图标（`aria-hidden`，对读屏透明）
+- **fix(ui): TPS 悬停索引不再跨视图残留**——`speedHover` 原先只存在默认状态里，切换「当前供应商 / 全部 Harness」、切换「数据 / 图表」、切换供应商时都不会重置，于是同一个索引会指向新序列里的另一天（或越界）。现在它与 `chartHover` 一起被清理
+- **fix(ui): 单点 TPS 曲线的命中区溢出**——只有一天样本时命中列从负 x 开始，左半区域无法触发悬停；现在把它夹进 viewBox
+- **fix(stats): 启动期统计结果可能带着未锚定的窗口被缓存**——重启后「套餐起点」要等第一次额度读取才知道，而在此之前算出的统计结果会按 TTL 一直复用。现在 Host 在锚点集合变化后强制重算一次
+- **perf: 模型切换不再无谓重取**——只有额度真正随模型变化的供应商（当前仅 `commandcode`）才会在切换模型时丢弃快照；在 DeepSeek / OpenCode 等模型无关的供应商上切模型不再触发上游请求。客户端也只在 `commandcode` 下补发一次快照
+- 文档：`PROTOCOL.md` 补上 `credits.poolUsd / allowanceUsd / allowanceModel / allowanceBasis`、`speed` 序列，以及 v1.7.2 的窗口锚定与 v1.7.3 的分母结论
+- 测试：370 → 373（仓库链接的位置/属性/图标、悬停状态清理、按供应商差异化的缓存失效）
+
 ## 1.7.5 — 2026-11-06
 
 （当日已发布版本：v1.6.5 / v1.6.6 / v1.6.7 / v1.7.0 / v1.7.1 / v1.7.2 / v1.7.3 / v1.7.4 / v1.7.5）

package/PROTOCOL.md

@@ -84,7 +84,15 @@ from `dsh-app://app/` and forwards non-asset paths to the Host).
     "monthly": { … } | null
   } | null,                                  // null for balance-only providers
   "balance": { "amount": 2.26, "currency": "CNY", "granted": 0, "toppedUp": 2.26 } | null,
-  "credits": { "remaining": 4.5, "granted": 10, "purchased": 0 } | null,
+  "credits": {
+    "remaining": 4.5, "granted": 10, "purchased": 0,
+    // v1.7.2: the SELECTED model's advertised allowance. Advertising only: the
+    // monthly percentage is always measured against the credit pool (v1.7.3),
+    // because that is what the provider's own console does.
+    "poolUsd": 10, "allowanceUsd": 14,
+    "allowanceModel": "deepseek/deepseek-v4.1-flash" | null,
+    "allowanceBasis": "model" | "pool" | "unknown"
+  } | null,
   "account": { "name": "…", "userName": "…" } | null,   // non-secret display info
   "links": { "console": "https://opencode.ai/console" } | null,  // provider-owned web page, follows the selection
   "fetchedAt": "2026-10-25T09:03:00.000Z",
@@ -129,9 +137,13 @@ plugin never sizes a month from a plan catalog.
 
 **The denominator is deliberately model-independent.** GOAT advertises "up to $100" and lists
 per-model allowances ($100 on DeepSeek V4.1 Flash, $60 on several others) because the same
-credits stretch further on a discounted model; the pool itself is one number. Letting the
-picker change the denominator would move the meter while the account did not change, so
-`parseCommandCode` takes no model input.
+credits stretch further on a discounted model; the pool itself is one number. Measured against
+the provider's own console: with $2.94 spent it prints 每月限额 **4%**, i.e. `2.94 / 70 = 4.2%`;
+`2.94 / 100` would print 2.9%. So the percentage is pool-based, and the per-model allowance is
+reported as `credits.allowanceUsd` for context only (v1.7.3). The model *is* forwarded with the
+request (`session.modelSelection.next.model`) so that the allowance can be resolved, and the Host
+drops the cached snapshot on a model switch **only** for providers whose allowance actually
+follows the model (`MODEL_SENSITIVE_ALLOWANCE`, currently `commandcode`).
 
 `windowLimits.limited` is a **capability flag** ("this plan paces you with rolling windows"),
 not a state: the live payload carries `limited: true` together with `fiveHour.exceeded: false`.
@@ -311,7 +323,11 @@ provider effective-date fallback (`effectiveDay`, `effectiveDays`,
 folder basename, used by the client to match a sidebar hover card; the full
 `cwd` is Host-only and never sent on the wire), the rolling 5-hour local window
 (`fiveHours`, DSH-only because external harnesses are day-granular), and the
-bounded `daily[]` series used by the chart view. Absolute local paths are
+bounded `daily[]` series used by the chart view, the `speed` series behind the
+TPS trend (v1.7.0), and the subscription window (`effectiveDay` is the plan's
+period start when the Host learned it, and `sinceEffective` is summed from the
+per-day buckets inside `[window start, today]` rather than borrowed from the
+30-day or all-time bucket — v1.7.2). Absolute local paths are
 stripped from the wire; free-form error messages have path substrings replaced
 with `[path]`.
 
@@ -357,7 +373,7 @@ with `[path]`.
         "last30Days": { "total": …, "messages": … },
         "allTime": { "total": …, "messages": … } }
     ],
-    "daily": [                                  // <=120 days, ascending; chart series
+    "daily": [                                  // <=30 days (v1.7.0), ascending; chart series
       { "day": "2026-10-25",
         "total": { "total": 123456, "messages": 42 },
         "hours": [0, 0, 0, 1234, 5678, …],      // 24 integers: local-hour token totals
@@ -366,6 +382,20 @@ with `[path]`.
                            "hours": [0, 0, 0, 500, 734, …] }
         } }
     ],
+    // v1.7.0: output-speed (TPS) series. `tps` is output tokens over the summed
+    // model time of the steps that produced them (`step/start` ->
+    // `assistant/message`), so a step's tool calls — which are emitted after the
+    // message — do not dilute it. A day with no positive duration is omitted
+    // instead of being given an infinite rate. `total.tps` is null when the
+    // window holds no samples. Each provider row carries its own `speed` object
+    // with the same shape, following the same window as its tokens (v1.7.2).
+    "speed": {
+      "byDay": [
+        { "day": "2026-11-05", "outputTokens": 516951, "seconds": 8493.1,
+          "tps": 60.9, "steps": 859 }
+      ],
+      "total": { "outputTokens": 939643, "seconds": 14123.9, "tps": 66.5, "steps": 1424 }
+    },
     "zcode": {                                  // extra-harness detail
       "dbPath": "C:\\Users\\…\\.zcode\\cli\\db\\db.sqlite",
       "available": true, "error": null, "rows": 65, "cached": false,

package/README.en.md

@@ -144,7 +144,7 @@ Host bridge route: `/api/tokenlens`; agent tool: `tokenlens_usage`.
 
 ## Development
 
-Node.js ≥ 24, zero runtime dependencies: `node --test` runs the 370 tests.
+Node.js ≥ 24, zero runtime dependencies: `node --test` runs the 373 tests.
 
 Release: `node scripts/bump-version.mjs <version> "<note>"` updates `package.json`, `PLUGIN_VERSION` and `CHANGELOG.md`; see [CONTRIBUTING.md](CONTRIBUTING.md) and [SECURITY.md](SECURITY.md).
 

package/README.md

@@ -445,7 +445,7 @@ Host 半注册 `tokenlens_usage`，参数 `action`（默认 `snapshot`）与可
 
 ## 验证
 
-- `node --test`：**370 个自动化用例**，覆盖 Host 数据层、供应商适配器（含小米 MiMo 控制台 Cookie 适配器、Command Code 月度额度反推）、统计折叠、各 harness 读取、Design Plan 周期账本与额度计价、TPS 速度序列、窗口锚点切片、工作区自动识别、插件集成与浏览器端渲染 / 交互。
+- `node --test`：**373 个自动化用例**，覆盖 Host 数据层、供应商适配器（含小米 MiMo 控制台 Cookie 适配器、Command Code 月度额度反推）、统计折叠、各 harness 读取、Design Plan 周期账本与额度计价、TPS 速度序列、窗口锚点切片、工作区自动识别、插件集成与浏览器端渲染 / 交互。
 - 在真实环境中验证过 **5 个 harness** 的数据源；本机统计与独立的 SQLite 查询、独立的日志解析逐项对账（差额为 0），跨 harness 合并与供应商映射算术自洽。
 - 供应商接口按各官方文档与参考客户端行为验证；未配置凭据时 chip 静默隐藏，接口异常时按可重试错误降级。
 

package/client.js

@@ -165,6 +165,8 @@ window.__ModuleLoader__.load({
     }
     /** The npm package page the 「打开 npm」 update links point at. */
     const NPM_PACKAGE_URL = 'https://www.npmjs.com/package/dsh-tokenlens'
+    /** The source repository, linked to the left of the version (v1.7.6). */
+    const REPO_URL = 'https://github.com/senyingxidi/dsh-tokenlens'
 
     const zh = {
       title: '词元透镜（dsh-TokenLens）',
@@ -272,6 +274,8 @@ window.__ModuleLoader__.load({
       capacityTokenNoEstimate: '预计可消耗 —',
       capacityTokenEstimateHint: '本机 Token 记录 ÷ 当前套餐百分比估算，仅供参考',
       pluginVersion: '插件 v{version}',
+      repoLink: 'GitHub',
+      repoLinkHint: '在 GitHub 上打开 dsh-TokenLens 仓库（源码、更新日志、问题反馈）',
       statsToday: '今日',
       statsAllTime: '总计',
       statsInput: '输入',
@@ -547,6 +551,8 @@ window.__ModuleLoader__.load({
       capacityTokenNoEstimate: 'Est. quota —',
       capacityTokenEstimateHint: 'Local token records ÷ current plan percentage; estimate only',
       pluginVersion: 'Plugin v{version}',
+      repoLink: 'GitHub',
+      repoLinkHint: 'Open the dsh-TokenLens repository on GitHub (source, changelog, issues)',
       statsToday: 'Today',
       statsAllTime: 'All time',
       statsInput: 'Input',
@@ -774,6 +780,10 @@ window.__ModuleLoader__.load({
       windows: ['M2.6 3.4h10.8v9.2H2.6z', 'M8 3.4v9.2', 'M2.6 7.9h10.8'],
       coin: ['M8 2.4a5.6 5.6 0 1 0 0 11.2A5.6 5.6 0 0 0 8 2.4Z', 'M8 5.2v5.6', 'M6.3 6.6h3.4', 'M6.3 9.4h3.4'],
       check: ['M3 8.4 6.2 11.4 13 4.4'],
+      // v1.7.6: the repository mark, drawn stroke-only like every other icon.
+      github: [
+        'M8 2.4a5.6 5.6 0 0 0-1.7 10.9c.3.1.4-.1.4-.3v-1c-1.5.3-1.9-.7-1.9-.7-.2-.6-.6-.7-.6-.7-.5-.4 0-.4 0-.4.5 0 .8.5.8.5.5.9 1.3.6 1.7.5 0-.4.2-.7.3-.8-1.3-.1-2.6-.6-2.6-2.7 0-.6.2-1.1.6-1.5-.1-.1-.2-.7 0-1.4 0 0 .5-.2 1.6.6a5.4 5.4 0 0 1 2.8 0c1.1-.8 1.6-.6 1.6-.6.3.7.1 1.3.1 1.4.3.4.5.9.5 1.5 0 2.1-1.3 2.6-2.6 2.7.2.2.4.5.4 1v1.5c0 .2.1.4.4.3A5.6 5.6 0 0 0 8 2.4Z',
+      ],
       alert: ['M8 2.5 14 13.2H2z', 'M8 6.2v3.2', 'M8 11.3h.01'],
       spinner: ['M8 2.5a5.5 5.5 0 1 1-5.5 5.5', 'M8 5.5v2.8'],
       ban: ['M3.2 3.2 12.8 12.8', 'M8 2.4a5.6 5.6 0 1 0 0 11.2A5.6 5.6 0 0 0 8 2.4Z'],
@@ -2206,6 +2216,9 @@ window.__ModuleLoader__.load({
         detailsWindow: settings.detailsWindow,
         chartHover: null,
         chartDay: null,
+        // v1.7.6: the TPS hover index is view state like the chart's, so it must
+        // not survive a provider switch (the series behind it changed).
+        speedHover: null,
       }
     }
 
@@ -3312,7 +3325,12 @@ window.__ModuleLoader__.load({
       if (modelChanged) store.set({ modelId: nextModel })
       syncProvider(providerId)
       const after = store.get()
-      if (modelChanged && after.providerId !== null && after.inFlight !== true) {
+      // v1.7.6: only Command Code's allowance follows the model, so only its
+      // snapshots are worth refetching on a switch (see index.js
+      // MODEL_SENSITIVE_ALLOWANCE). Everyone else keeps the cached snapshot.
+      const modelSensitive =
+        after.providerId !== null && canonicalProviderId(after.providerId) === 'commandcode'
+      if (modelChanged && modelSensitive && after.inFlight !== true) {
         void requestUsage('snapshot')
       }
     }
@@ -3614,7 +3632,11 @@ window.__ModuleLoader__.load({
 .dsh-tokenlens-chart-drill-title{color:var(--dsw-alias-label-secondary);font-size:var(--dsw-alias-font-size-sm,12px);line-height:1.4;min-width:0;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;font-variant-numeric:tabular-nums}
 /* 明细 donut window switch (v1.2.0): one pill row under the details title. */
 .dsh-tokenlens-details-window{display:flex;align-items:center;gap:6px;flex-wrap:wrap;min-width:0}
-.dsh-tokenlens-version{color:var(--dsw-alias-label-tertiary);font-size:11px;line-height:1.5;text-align:right;opacity:.75;overflow-wrap:anywhere}
+.dsh-tokenlens-version{color:var(--dsw-alias-label-tertiary);font-size:11px;line-height:1.5;text-align:right;opacity:.75;overflow-wrap:anywhere;display:flex;align-items:center;justify-content:flex-end;gap:6px;flex-wrap:wrap}
+/* v1.7.6: the repository link sits left of the version text. */
+.dsh-tokenlens-repo{display:inline-flex;align-items:center;gap:4px;color:inherit;text-decoration:none;flex:0 0 auto;border-radius:var(--dsw-radius-sm);transition:color .12s ease,opacity .12s ease}
+.dsh-tokenlens-repo:hover,.dsh-tokenlens-repo:focus-visible{color:var(--dsw-alias-label-primary,currentColor);opacity:1;text-decoration:underline}
+.dsh-tokenlens-version-text{flex:0 0 auto}
 /* Update status on the plugin-version line (dialog body): wraps on narrow screens. */
 .dsh-tokenlens-version-update{white-space:normal;overflow-wrap:anywhere}
 .dsh-tokenlens-version .dsh-tokenlens-update-link{color:inherit;font-size:inherit;vertical-align:baseline}
@@ -4495,7 +4517,7 @@ window.__ModuleLoader__.load({
             'data-scope': scope,
             'data-active': effective === scope ? 'true' : 'false',
             disabled: stats === null,
-            onClick: () => store.set({ statsScope: scope, chartHover: null, chartDay: null }),
+            onClick: () => store.set({ statsScope: scope, chartHover: null, chartDay: null, speedHover: null }),
           },
           label,
         )
@@ -4521,7 +4543,7 @@ window.__ModuleLoader__.load({
             'data-active': active === view ? 'true' : 'false',
             'aria-pressed': active === view ? 'true' : 'false',
             onClick: () => {
-              if (active !== view) store.set({ statsView: view, chartHover: null, chartDay: null })
+              if (active !== view) store.set({ statsView: view, chartHover: null, chartDay: null, speedHover: null })
             },
           },
           Icon(iconName, 12),
@@ -6173,9 +6195,11 @@ window.__ModuleLoader__.load({
                   'rect',
                   {
                     key: `hit-${entry.day}-${index}`,
-                    x: (points[index].x - columnWidth / 2).toFixed(1),
+                    // v1.7.6: clamp the hit column into the viewBox — a single-day
+                    // series used to start at a negative x and lose half its area.
+                    x: Math.max(0, Math.min(width - 1, points[index].x - columnWidth / 2)).toFixed(1),
                     y: 0,
-                    width: Math.max(1, columnWidth).toFixed(1),
+                    width: Math.max(1, Math.min(columnWidth, width)).toFixed(1),
                     height: String(height),
                     fill: 'transparent',
                     'data-speed-day': entry.day,
@@ -7804,11 +7828,29 @@ window.__ModuleLoader__.load({
       const body = [buildManageButton(state, t), statsArea]
       const displayVersion = currentPluginVersionOf(state)
       if (displayVersion !== '') {
+        // v1.7.6: the repository link sits to the LEFT of the version, so the
+        // version stays the right-most element of the row.
+        const repo = h(
+          'a',
+          {
+            key: 'repo',
+            className: 'dsh-tokenlens-repo',
+            href: REPO_URL,
+            target: '_blank',
+            rel: 'noreferrer noopener',
+            'data-repo-link': 'true',
+            title: t('repoLinkHint'),
+            'aria-label': t('repoLinkHint'),
+          },
+          Icon('github', 13),
+          t('repoLink'),
+        )
         body.push(
           h(
             'div',
             { key: 'version', className: 'dsh-tokenlens-version' },
-            t('pluginVersion', { version: displayVersion }),
+            repo,
+            h('span', { key: 'text', className: 'dsh-tokenlens-version-text' }, t('pluginVersion', { version: displayVersion })),
             // Update status sits right next to the plugin version: 有新版本 vX
             // (link to the npm page) or 已是最新版本, only for a successful check.
             buildVersionUpdateStatus(state, t),

package/index.js

@@ -84,7 +84,16 @@ export const inject = []
 export const PLUGIN_NAME = 'dsh-tokenlens'
 
 /** Plugin version; mirrors package.json (kept here so diagnostics need no fs read). */
-export const PLUGIN_VERSION = '1.7.5'
+export const PLUGIN_VERSION = '1.7.6'
+
+/**
+ * Providers whose monthly allowance depends on the SELECTED model (v1.7.6).
+ *
+ * Only these need their cached snapshot dropped when the picker changes: for
+ * everyone else a model switch does not alter a single number, so refetching
+ * would be pure waste.
+ */
+const MODEL_SENSITIVE_ALLOWANCE = new Set(['commandcode'])
 
 /** Documented defaults; every field of the raw row config is optional. */
 const DEFAULTS = {
@@ -1460,8 +1469,7 @@ export function apply(ctx, rawConfig) {
     return catalog
   }
 
-  const statsService = config.stats.enabled
-    ? createUsageStats({
+  const statsService = config.stats.enabled    ? createUsageStats({
         dshHome: config.stats.dshHome !== '' ? config.stats.dshHome : dshHomeOf(process.env),
         config,
         providerCatalog: providerModelCatalog,
@@ -1614,6 +1622,14 @@ export function apply(ctx, rawConfig) {
     return scrubMessagePaths(out)
   }
 
+  /**
+   * The subscription-anchor set the last stats result was computed with (v1.7.6).
+   *
+   * Empty until a plan period start is known; a change invalidates the cached
+   * stats result so the 最近X天 window is anchored immediately.
+   */
+  let statsAnchorSignature = '[]'
+
   async function getStats(options = {}) {
     if (statsService === null) {
       return {
@@ -1622,11 +1638,18 @@ export function apply(ctx, rawConfig) {
       }
     }
     try {
+      const starts = Object.fromEntries(state.subscriptionStarts)
+      // v1.7.6: a cached stats result computed BEFORE a plan period start was
+      // known would keep the un-anchored window until the TTL expired. Recompute
+      // once whenever the anchor set changes, so the window is right immediately.
+      const signature = JSON.stringify(Object.entries(starts).sort())
+      const anchorChanged = signature !== statsAnchorSignature
       const stats = await statsService.get({
-        fresh: options.fresh === true,
+        fresh: options.fresh === true || (signature !== '[]' && anchorChanged),
         // v1.7.2: anchor each subscription provider's window to its plan start.
-        subscriptionStarts: Object.fromEntries(state.subscriptionStarts),
+        subscriptionStarts: starts,
       })
+      if (signature !== '[]') statsAnchorSignature = signature
       // v0.22.1: only subscription-backed providers use the 最近X天 window;
       // pay-as-you-go/balance providers keep the fixed 最近30天 column.
       const decorated =
@@ -1835,6 +1858,10 @@ export function apply(ctx, rawConfig) {
    * V4.1 Flash $100 while its credit pool is $70), so a model switch must also
    * drop the cached snapshot — otherwise the chip would keep showing the old
    * allowance until the next poll.
+   *
+   * v1.7.6: only for providers whose allowance is actually model-sensitive.
+   * Invalidating on every provider made switching models on DeepSeek/OpenCode
+   * throw away a perfectly good snapshot and refetch it.
    */
   function rememberRequestedModel(body, providerId) {
     const modelId = typeof body?.modelId === 'string' ? body.modelId.trim() : ''
@@ -1842,8 +1869,10 @@ export function apply(ctx, rawConfig) {
     const id = canonicalProviderId(providerId)
     if (state.selectedModels.get(id) === modelId) return
     state.selectedModels.set(id, modelId)
-    state.caches.delete(id)
-    state.flights.delete(id)
+    if (MODEL_SENSITIVE_ALLOWANCE.has(id)) {
+      state.caches.delete(id)
+      state.flights.delete(id)
+    }
     noteStatus('client-model', { providerId: id, modelId })
   }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-tokenlens",
-  "version": "1.7.5",
+  "version": "1.7.6",
   "type": "module",
   "description": "dsh-TokenLens (词元透镜) — provider-aware plan capacity plus local token statistics (today/7d/30d/all-time, by harness/provider/model/workspace) for DSH.",
   "license": "MIT",