dsh-balance-widget 0.5.0 → 0.5.2

package/README.en.md

@@ -24,8 +24,8 @@ A balance & cost widget for the [DeepSeek Harness](https://github.com/deepseek-a
 | **Zero external dependencies** | ✅ Imports no `@deepseek-ai/*` packages, no native modules | ❌ Most depend on dsh SDK packages |
 | **Node 24 ready** | ✅ Works out of the box on any profile layout | ⚠️ Many community plugins still error on Node 24 |
 | **Boot stability** | ✅ Registers routes via official `ctx.webServer`; never conflicts with apiproxy | ⚠️ Some self-host HTTP servers that crash `dsh web` on boot |
-| **On-demand queries** | ✅ Fetches balance only on click; no polling, zero background requests | Some always-on badges refresh on a timer |
-| **Peak/off-peak pricing** | ✅ Built-in official 2026-08-17 rate table, auto-switches by window | Partial support |
+| **Always-fresh auto refresh** | ✅ Sidebar card refreshes balance & costs every 60s; opening the popover triggers an immediate refresh | Some always-on badges refresh on a timer |
+| **Peak/off-peak pricing** | ✅ Built-in official 2026-09-10 rate table, re-synced from the official page at startup and every 12h | Partial support |
 | **Security** | ✅ API key stays in the host process; loopback-only guard | Varies |
 
 **In one line**: *The zero-dependency, Node 24-ready balance/cost widget that never breaks `dsh web` boot.*
@@ -40,11 +40,10 @@ A balance & cost widget for the [DeepSeek Harness](https://github.com/deepseek-a
 - **Peak/off-peak status** — The card and popover borders are tinted by the current window (orange at peak / green at off-peak), a "Peak/Off-peak" tag sits next to the popover title, and hovering it shows the current price tier (input/output per 1M tokens).
 - **Token usage** — Also shows the session's input (incl. cache hits) / output tokens.
 - **One-click top-up** — a "Top up" link in the popover footer jumps to the official DeepSeek top-up page (platform.deepseek.com/top_up) in a new tab.
-- **Sidebar card** — a persistent card at the sidebar footer shows balance and today's cost; globally visible, refreshes every 60s.
+- **Sidebar card** — a persistent card at the sidebar footer shows balance and today's cost; globally visible, auto-refreshes every 60s (balance via the official API, costs parsed locally), and opening the popover triggers an immediate refresh.
 - **Balance color warning** — the balance number is tinted in three tiers: default (healthy) → amber (below `lowThreshold`) → red (below `criticalThreshold`).
 - **Official price auto-sync** — fetches the DeepSeek official pricing page on startup and every 12h; falls back to built-in rates on failure.
 - **Agent tool** — a `deepseek_billing` tool lets the model answer "how much balance do I have / how much did today cost".
-- **On-demand refresh** — No polling, no background requests; endpoints are only hit when you click the icon. Costs zero tokens to use.
 
 ## Why this plugin
 
@@ -137,23 +136,68 @@ If you mainly use DeepSeek-V4-Pro, point the pricing model at it for a more accu
     modelId: deepseek-v4-pro
 ```
 
+**Note**: the provider has announced that V4-Pro is being retired — from 2026-09-14 12:00 CST, V4-Pro requests are routed to V4.1-Flash and billed at Flash rates. Keeping `deepseek-v4-pro` past that date will over-estimate costs, so the default Flash tier is the better choice.
+
 **Note**: `cordis.patch.yml` may already contain lines for other plugins — append new lines without touching existing ones.
 
 ## Pricing
 
-Built-in DeepSeek official peak/off-peak pricing (CNY per 1M tokens), effective 2026-08-17. Peak windows are Beijing time 09:00–12:00 and 14:00–18:00; prices are double the off-peak rates:
+Built-in DeepSeek official peak/off-peak pricing (CNY per 1M tokens), effective 2026-09-10. Peak windows are Beijing time 09:00–12:00 and 14:00–18:00; prices are double the off-peak rates:
 
 | Model | Window | Cache hit (input) | Cache miss (input) | Output |
 | --- | --- | --- | --- | --- |
-| V4-Flash | Off-peak | 0.05 | 1.5 | 4.5 |
-| V4-Flash | Peak | 0.10 | 3.0 | 9.0 |
+| V4.1-Flash | Off-peak | 0.02 | 1.0 | 4.0 |
+| V4.1-Flash | Peak | 0.04 | 2.0 | 8.0 |
 | V4-Pro | Off-peak | 0.15 | 4.5 | 13.5 |
 | V4-Pro | Peak | 0.30 | 9.0 | 27.0 |
 
-`deepseek-chat` / `deepseek-reasoner` aliases map to Flash / Pro pricing respectively. Costs are **estimates**; actual billing from the provider is authoritative.
+The official page currently lists only `deepseek-flash` and `deepseek-v4-pro`; older names (`deepseek-v4-flash`, `deepseek-v4-flash-vision-exp`, `deepseek-chat`) still work and are billed at the Flash rate. The plugin re-parses the page at startup and every 12h, and falls back to the table above only when parsing fails. Costs are **estimates**; the provider's bill is authoritative.
+
+## Security & permission boundaries
+
+This section is for the DSH Store / plugin audit: dependencies, runtime permissions, external services, and failure bounds.
+
+**Dependencies & compatibility**
+- Zero runtime dependencies: imports no `@deepseek-ai/*` packages; no third-party host deps
+- `peerDependencies["@deepseek-ai/dsh"]`: `>=0.1.2-rc.1 <0.2.0` (DSH compatibility range)
+- `engines.node`: `^22.19.0 || >=24.0.0`
+- `peerDependencies["react"]`: `^18.2.0` (browser rendering only)
+
+**Runtime permissions**
+- `files`: reads only `~/.dsh/sessions/` session JSONL (cost stats); never writes or mutates any session file
+- `network`: only the DeepSeek official endpoints — `api.deepseek.com` (`GET /user/balance`) and `api-docs.deepseek.com` pricing page (fetched every 12h); no third-party proxy
+- `commands`: spawns `zstd -d -c` to decompress session files (macOS needs `brew install zstd`); no other commands
+- `credentials`: reads `DEEPSEEK_API_KEY` (resolved via the host credentials service), used only in the host process behind a loopback-only route guard; the browser never sees the key
+- All host routes are bound to the loopback address and unreachable externally
+
+**External services**
+- DeepSeek official balance endpoint `GET /user/balance` (on click / 60s refresh)
+- DeepSeek official pricing page (on startup + every 12h, for peak/off-peak rates)
+
+**Failure bounds**
+- Balance fetch failure: the panel shows the error and keeps the last successful snapshot (no interruption)
+- Pricing fetch failure: falls back to the built-in 2026-09-10 rate table, `pricingSource` marked `default` (`synced` once parsing succeeds)
+- Missing `zstd`: returns an actionable error (points to the install command) instead of failing silently
+- Missing/corrupt session files: that session is skipped; other sessions are unaffected
+- All costs are estimates; the provider's bill is authoritative
 
 ## Changelog
 
+### v0.5.2 — DSH 0.1.5: new session format & new pricing page
+- 🐛 **Fixed**: DSH 0.1.5 writes session logs under a generation-tagged name (`session.v3.jsonl.zstd`), but the plugin only matched `session.jsonl.zstd`, so every session created after the upgrade was invisible — viewing one produced red error text in the popover, and today's cost was understated (measured ~36% low). Logs are now found by taking the highest generation per session directory; a migrated session's older file is a subset of the new one, so reading only the newest avoids double-counting the session
+- 🐛 **Fixed**: the official pricing page renamed the Flash column to `deepseek-flash` and cut its rates. The parser's model-id anchor landed on a page footnote, so the sync reported success while silently keeping the stale, higher rates — overstating costs by ~1.7x. The parser now anchors on the table's row labels and applies that column to every Flash-family name
+- 💰 **Rates**: the built-in fallback table now carries V4.1-Flash pricing (off-peak 0.02 / 1.0 / 4.0, peak 0.04 / 2.0 / 8.0 CNY per 1M tokens); the V4-Pro tier is unchanged
+- 📄 **Docs**: corrected the `pricingSource` values to `default` / `synced` (previously misdocumented as `builtin`)
+
+### v0.5.1 — Cold-start speedup & cleanup
+- 🚀 **Performance**: today-cost cold start dropped from ~5.4s to ~0.01s — only session files modified today are decompressed (mtime filter), parsed results are cached by (path, mtime), and parsing runs in parallel
+- ⏰ **Refresh wording**: README now matches the code — the persistent card auto-refreshes every 60s (the old "no polling" claims contradicted the code and have been corrected)
+- 🌏 **Timezone fix**: peak/off-peak windows are now computed in Beijing time (UTC+8) explicitly instead of the host's local timezone
+- 🧹 **Cleanup**: removed a dead client-side pricing stack (PRICING/priceSession) that was never called; all pricing now goes through the host
+- 📐 **Pricing parsing hardened**: peak rates are parsed explicitly from the official page (no more hard-coded "off-peak × 2"); if a cache-hit rate cannot be parsed the whole table falls back to built-in rates instead of silently pricing cache hits at 0
+- 🏷️ **Card label**: the card footer now reads "Today · all" to make clear it is the global (all-workspaces) figure
+- 🎨 **Visual refresh**: SVG wallet icon replaces 💰, press/pop micro-animations, muted secondary tiers in the popover
+
 ### v0.5.0 — Five-tier costs & peak/off-peak status
 - ✨ **Added**: the cost breakdown is now five tiers — balance / last prompt / today-this-session / today-this-workspace / today-all-workspaces
   - The last-prompt row labels the **session name** underneath (based on the most recent session)
@@ -176,7 +220,9 @@ Built-in DeepSeek official peak/off-peak pricing (CNY per 1M tokens), effective
 
 - Config tree: `dsh --profile web --dump-config` should show a `balance-widget` entry.
 - Balance route: after restarting dsh web, `curl -s http://127.0.0.1:3080/api/dsh-balance/balance` should return `{ ok, balance_infos, modelId }`.
-- Cost routes: `curl -s "http://127.0.0.1:3080/api/dsh-balance/last-cost?session=SESSION_ID"` and `curl -s http://127.0.0.1:3080/api/dsh-balance/today-cost` should return `{ cost, inputTokens, outputTokens, modelId }`.
+- Session costs: `curl -s http://127.0.0.1:3080/api/dsh-balance/active-cost` should return `{ lastPrompt, todaySession, title, sessionId, peak, workspaceName, ... }`; append `?session=SESSION_ID` to target a specific session.
+- Today costs: `curl -s http://127.0.0.1:3080/api/dsh-balance/today-cost` should return `{ workspace: { cost, ..., cwd }, all: { cost, ... }, modelId }`.
+- Legacy route: `curl -s "http://127.0.0.1:3080/api/dsh-balance/last-cost?session=SESSION_ID"` still works and returns `{ cost, inputTokens, outputTokens, modelId }`.
 
 ## License
 

package/README.md

@@ -24,8 +24,8 @@ DeepSeek Harness (DSH) Web GUI 的余额与成本小部件：侧边栏底部常
 | **零外部依赖** | ✅ 不 import 任何 `@deepseek-ai/*` 包，无原生模块 | ❌ 多数依赖 dsh SDK 包 |
 | **Node 24 兼容** | ✅ 天然兼容（零依赖设计），任何 profile 布局可加载 | ⚠️ 不少社区插件在 Node 24 下报错 |
 | **启动稳定性** | ✅ 用官方 `ctx.webServer` 注册路由，不与 apiproxy 冲突 | ⚠️ 有的自建 HTTP 服务导致 dsh web 启动崩溃 |
-| **按需查询** | ✅ 点击才查余额，无轮询、零后台请求 | 有的常驻徽章定时刷新 |
-| **峰谷定价** | ✅ 内置官方 2026-08-17 峰谷价表，自动按时段切换 | 部分支持 |
+| **常驻自动刷新** | ✅ 侧边栏卡片每 60s 自动刷新余额与成本，弹框点击即强刷 | 有的常驻徽章定时刷新 |
+| **峰谷定价** | ✅ 内置官方 2026-09-10 峰谷价表，启动时与每 12h 自动从官方页刷新 | 部分支持 |
 | **安全性** | ✅ API key 仅在宿主进程，loopback-only 守卫 | 参差不齐 |
 
 **一句话**：*零依赖、Node 24 就绪、永不拖垮 dsh web 启动的余额/成本小部件。*
@@ -40,11 +40,10 @@ DeepSeek Harness (DSH) Web GUI 的余额与成本小部件：侧边栏底部常
 - **峰谷状态标签** — 卡片与弹框边框按当前时段着色（峰时橙色 / 谷时绿色），弹框标题旁显示「峰时/谷时」标签，悬停可查看当前价格档位（输入/输出单价）。
 - **Token 用量** — 同时展示输入（含缓存命中）/ 输出 token 数。
 - **一键充值** — 弹层底部「去充值」链接直达 DeepSeek 官方充值页（platform.deepseek.com/top_up），新窗口打开。
-- **侧边栏常驻卡片** — 侧边栏底部（设置上方）显示余额 + 今日花费，全局可见，60 秒自动刷新。
+- **侧边栏常驻卡片** — 侧边栏底部（设置上方）显示余额 + 今日花费，全局可见；每 60 秒自动刷新（余额走官方接口、成本为本地会话解析），打开弹框时也会立即刷新一次，无需手动操作。
 - **余额数字变色预警** — 余额数字按三档着色：充足（默认色）→ 琥珀（低于 `lowThreshold`）→ 红色（低于 `criticalThreshold`），一眼判断余额健康度。
 - **官方价格自动同步** — 启动时 + 每 12 小时抓取 DeepSeek 官方定价页，改价自动跟进；失败回退内置价目表。
 - **模型工具查询** — 新增 `deepseek_billing` 工具，可直接问模型"余额多少/今天花了多少"。
-- **按需刷新** — 无轮询、无后台请求；只有点击图标时才发起查询，不消耗任何 token。
 
 ## 架构
 
@@ -134,23 +133,68 @@ DSH 的插件配置统一放在这个文件里：
     modelId: deepseek-v4-pro
 ```
 
+**注意**：官方已公告 V4-Pro 将于北京时间 2026-09-14 12:00 下线——此后 V4-Pro 请求会被路由到 V4.1-Flash 并按 Flash 计费。若在这之后仍配置 `deepseek-v4-pro`，估算会偏高，建议使用默认的 Flash 档。
+
 **提示**：`cordis.patch.yml` 可能有其他插件的配置行，追加时注意**不要改动已有的行**，只加新内容。
 
 ## 定价说明
 
-内置 DeepSeek 官方 2026-08-17 峰谷定价（元 / 百万 tokens），高峰时段为北京时间 09:00–12:00、14:00–18:00，价格为空闲时段两倍：
+内置 DeepSeek 官方 2026-09-10 峰谷定价（元 / 百万 tokens），高峰时段为北京时间 09:00–12:00、14:00–18:00，价格为空闲时段两倍：
 
 | 模型 | 时段 | 缓存命中(输入) | 缓存未命中(输入) | 输出 |
 | --- | --- | --- | --- | --- |
-| V4-Flash | 空闲 | 0.05 | 1.5 | 4.5 |
-| V4-Flash | 高峰 | 0.10 | 3.0 | 9.0 |
+| V4.1-Flash | 空闲 | 0.02 | 1.0 | 4.0 |
+| V4.1-Flash | 高峰 | 0.04 | 2.0 | 8.0 |
 | V4-Pro | 空闲 | 0.15 | 4.5 | 13.5 |
 | V4-Pro | 高峰 | 0.30 | 9.0 | 27.0 |
 
-`deepseek-chat` / `deepseek-reasoner` 别名分别映射到 Flash / Pro 价格。成本为**估算值**，实际以官方账单为准。
+官方页当前只列 `deepseek-flash` 与 `deepseek-v4-pro` 两列；`deepseek-v4-flash`、`deepseek-v4-flash-vision-exp`、`deepseek-chat` 等旧名仍可调用，按 Flash 价计费。插件在启动时与每 12h 从官方页重新解析，只有解析失败才回退上表。成本为**估算值**，实际以官方账单为准。
+
+## 安全与权限边界
+
+本节面向 DSH Store / 插件审计，列出依赖、运行时权限、外部服务与失败边界。
+
+**依赖与兼容**
+- 零运行时依赖：不 import 任何 `@deepseek-ai/*` 包，宿主端无第三方依赖
+- `peerDependencies["@deepseek-ai/dsh"]`: `>=0.1.2-rc.1 <0.2.0`（DSH 兼容范围）
+- `engines.node`: `^22.19.0 || >=24.0.0`
+- `peerDependencies["react"]`: `^18.2.0`（仅浏览器端渲染）
+
+**运行时权限**
+- `files`：只读 `~/.dsh/sessions/` 下的会话 JSONL（成本统计）；不写入、不修改任何会话文件
+- `network`：仅请求 DeepSeek 官方端点——`api.deepseek.com`（`GET /user/balance` 余额）与 `api-docs.deepseek.com` 定价页（每 12h 抓取）；不走任何第三方代理
+- `commands`：spawn `zstd -d -c` 解压会话文件（macOS 需 `brew install zstd`）；不执行其他命令
+- `credentials`：读取 `DEEPSEEK_API_KEY`（经宿主凭据服务解析），仅宿主进程使用、loopback-only 路由守卫；浏览器不接触密钥
+- 所有 host 路由均绑定 load 回环地址，外部不可达
+
+**外部服务**
+- DeepSeek 官方余额接口 `GET /user/balance`（点击/60s 刷新时调用）
+- DeepSeek 官方定价页（启动时 + 每 12h 抓取，用于峰谷单价）
+
+**失败边界**
+- 余额接口失败：面板提示失败信息，保留上次成功快照（不中断）
+- 定价页抓取失败：回退内置 2026-09-10 价目表，`pricingSource` 标记为 `default`（解析成功则为 `synced`）
+- `zstd` 缺失：返回可读错误提示（指引安装），而非静默失败
+- 会话文件缺失/损坏：跳过该会话，不影响其他会话统计
+- 所有成本为估算值，实际以官方账单为准
 
 ## 版本历史
 
+### v0.5.2 — 适配 DSH 0.1.5：新会话格式与新定价页
+- 🐛 **修复**：DSH 0.1.5 将会话日志改为代数命名 `session.v3.jsonl.zstd`，插件此前只认 `session.jsonl.zstd`，导致升级后新建的会话全部不可见——查看这类会话时弹层出现红色报错，今日成本也漏算（实测少约 36%）。现按「每个会话目录取最高代」发现日志；迁移会话保留的旧文件是新代的子集，因此只读最高代，避免整个会话被重复计算
+- 🐛 **修复**：官方定价页将 Flash 列改名为 `deepseek-flash` 并下调价格，插件按模型名定位价格块的锚点因此落到页面脚注，同步"成功"却静默沿用旧高价，成本被高估约 1.7 倍。现改为按价格表行标签锚定，并把该列价格套用到全部 Flash 系名称
+- 💰 **价目**：内置兜底价目表更新为 V4.1-Flash 价（空闲 0.02 / 1.0 / 4.0，高峰 0.04 / 2.0 / 8.0，元/百万 tokens）；V4-Pro 档位保留
+- 📄 **文档**：更正 `pricingSource` 取值为 `default` / `synced`（此前误写为 `builtin`）
+
+### v0.5.1 — 冷启动提速与代码清理
+- 🚀 **性能**：今日成本冷启动从 ~5.4s 降至 ~0.01s——只解压当天活跃的会话文件（按 mtime 过滤）+ 解压结果按 (path, mtime) 长期缓存 + 并行解析
+- ⏰ **轮询口径**：README 与代码统一——常驻卡片每 60s 自动刷新（此前文档"无轮询"表述自相矛盾，已更正）
+- 🌏 **峰谷时区修正**：峰/谷判断改为显式按北京时间（UTC+8）计算，不再依赖宿主机时区
+- 🧹 **清理**：移除客户端一套从未调用的死计价代码（PRICING/priceSession），计价统一走宿主
+- 📐 **定价解析加固**：峰价改为从官方页面显式解析（不再硬编码"谷价 ×2"）；解析不到缓存命中价时整表回退内置价目表，避免静默按 0 计价
+- 🏷️ **卡片语义**：卡片底部"今日"改为"今日·全部"，明确是全局（所有工作区）口径
+- 🎨 **外观**：💰 换 SVG 钱包图标、弹框/卡片微交互动画、层级弱化（详情见 v0.5.1 diff）
+
 ### v0.5.0 — 五层级成本与峰谷状态
 - ✨ **新增**：成本明细改为五层级——余额 / 最近一次提问 / 今日·本会话 / 今日·本工作区 / 今日·所有工作区
   - 最近一次提问下方标注**会话名**（基于最近活跃会话）
@@ -173,7 +217,9 @@ DSH 的插件配置统一放在这个文件里：
 
 - 配置树：`dsh --profile web --dump-config` 应出现 `balance-widget` 条目
 - 余额路由：重启 dsh web 后 `curl -s http://127.0.0.1:3080/api/dsh-balance/balance` 应返回 `{ ok, balance_infos, modelId }`
-- 成本路由：`curl -s "http://127.0.0.1:3080/api/dsh-balance/last-cost?session=SESSION_ID"` 与 `curl -s http://127.0.0.1:3080/api/dsh-balance/today-cost` 应返回 `{ cost, inputTokens, outputTokens, modelId }`
+- 会话成本：`curl -s http://127.0.0.1:3080/api/dsh-balance/active-cost` 应返回 `{ lastPrompt, todaySession, title, sessionId, peak, workspaceName, ... }`；加 `?session=SESSION_ID` 可指定会话
+- 今日成本：`curl -s http://127.0.0.1:3080/api/dsh-balance/today-cost` 应返回 `{ workspace: { cost, ..., cwd }, all: { cost, ... }, modelId }`
+- 兼容旧路由：`curl -s "http://127.0.0.1:3080/api/dsh-balance/last-cost?session=SESSION_ID"` 仍可用，返回 `{ cost, inputTokens, outputTokens, modelId }`
 
 ## License
 

package/lib/client.js

@@ -8,7 +8,7 @@ window.__ModuleLoader__.load({
 		let react = require("react");
 		let _deepseek_ai_dsh_client_ui_primitives = require("@deepseek-ai/dsh-client-ui-primitives");
 		//#region lib/types/client/styles.js
-		const css = ".dshbw_wrap{position:relative;display:inline-flex;align-items:center}.dshbw_btn{background:none;border:none;cursor:pointer;color:var(--dsw-alias-label-secondary);display:inline-flex;align-items:center;gap:4px;padding:2px 6px;border-radius:6px;font-size:12px;line-height:18px;font-family:inherit}.dshbw_btn:hover{color:var(--dsw-alias-label-primary);background:var(--dsw-alias-interactive-bg-hover)}.dshbw_btn[data-level=\"1\"]{color:var(--dsw-alias-state-warn-primary)}.dshbw_btn[data-level=\"2\"]{color:var(--dsw-alias-state-error-primary)}.dshbw_pop{position:absolute;top:calc(100% + 6px);right:0;z-index:60;min-width:230px;background:var(--dsw-alias-bg-layer-3);border:1px solid var(--dsw-alias-border-l3);border-radius:10px;box-shadow:0 8px 24px rgba(0,0,0,.18);padding:12px 14px;font-size:12px;line-height:20px;color:var(--dsw-alias-label-primary)}.dshbw_head{display:flex;align-items:center;justify-content:space-between;margin-bottom:4px}.dshbw_title{font-weight:600;color:var(--dsw-alias-label-primary)}.dshbw_refresh{background:none;border:1px solid var(--dsw-alias-border-l2);cursor:pointer;color:var(--dsw-alias-label-secondary);display:inline-flex;align-items:center;justify-content:center;width:24px;height:24px;border-radius:6px;font-size:13px;line-height:1;padding:0;font-family:inherit}.dshbw_refresh:hover{color:var(--dsw-alias-label-primary);border-color:var(--dsw-alias-border-l3);background:var(--dsw-alias-interactive-bg-hover)}.dshbw_refresh[data-spinning=true]{animation:dshbw-spin .8s linear infinite}.dshbw_row{display:flex;justify-content:space-between;align-items:center;gap:12px;padding:2px 0}.dshbw_label{display:inline-flex;align-items:center;min-width:0}.dshbw_val{font-weight:600;color:var(--dsw-alias-label-primary)}.dshbw_val[data-level=\"1\"]{color:var(--dsw-alias-state-warn-primary)}.dshbw_val[data-level=\"2\"]{color:var(--dsw-alias-state-error-primary)}.dshbw_err{color:var(--dsw-alias-state-error-primary);font-size:11px;line-height:16px;margin-top:6px}.dshbw_hint{color:var(--dsw-alias-label-tertiary);font-size:11px;line-height:16px;margin-top:6px;border-top:1px solid var(--dsw-alias-border-l1);padding-top:6px}.dshbw_foot{display:flex;align-items:center;justify-content:space-between;gap:8px;margin-top:8px}.dshbw_link{background:none;border:none;cursor:pointer;color:var(--dsw-alias-state-business-primary);display:inline-flex;align-items:center;gap:4px;padding:2px 6px;border-radius:6px;font-size:12px;line-height:18px;font-family:inherit;text-decoration:none}.dshbw_link:hover{color:var(--dsw-alias-state-business-primary);text-decoration:underline}@keyframes dshbw-spin{from{transform:rotate(0deg)}to{transform:rotate(360deg)}}.dshbw_side{display:flex;flex-direction:column;gap:4px;padding:6px 8px;margin:0 8px 4px;border-radius:8px;background:var(--dsw-alias-bg-layer-2);color:var(--dsw-alias-label-primary)}.dshbw_side:hover{background:var(--dsw-alias-interactive-bg-hover)}.dshbw_sideTop{display:flex;align-items:center;justify-content:space-between;gap:8px;font-size:12px;line-height:18px}.dshbw_sideAmount{font-weight:700}.dshbw_sideAmount[data-level=\"1\"]{color:var(--dsw-alias-state-warn-primary);font-weight:700}.dshbw_sideAmount[data-level=\"2\"]{color:var(--dsw-alias-state-error-primary)}.dshbw_sideFoot{display:flex;justify-content:space-between;font-size:10px;line-height:14px;color:var(--dsw-alias-label-tertiary)}.dshbw_sideWrap{position:relative}.dshbw_side{display:flex;flex-direction:column;gap:4px;width:100%;padding:6px 8px;margin:0 8px 4px;border:none;border-radius:8px;background:var(--dsw-alias-bg-layer-2);color:var(--dsw-alias-label-primary);cursor:pointer;font-family:inherit;text-align:left}.dshbw_side:hover{background:var(--dsw-alias-interactive-bg-hover)}.dshbw_sidePop{position:absolute;bottom:calc(100% + 8px);left:8px;z-index:70;min-width:240px;background:var(--dsw-alias-bg-layer-3);border:1px solid var(--dsw-alias-border-l3);border-radius:10px;box-shadow:0 8px 24px rgba(0,0,0,.18);padding:12px 14px;font-size:12px;line-height:20px;color:var(--dsw-alias-label-primary)}.dshbw_sidePopHead{display:flex;align-items:center;justify-content:space-between;margin-bottom:6px}.dshbw_sidePopTitle{font-weight:600}.dshbw_sidePopClose{background:none;border:none;cursor:pointer;color:var(--dsw-alias-label-tertiary);font-size:16px;line-height:1;padding:0 2px;font-family:inherit}.dshbw_sidePopClose:hover{color:var(--dsw-alias-label-primary)}.dshbw_sidePopFoot{display:flex;align-items:center;justify-content:space-between;gap:8px;margin-top:10px}.dshbw_sidePopRefresh{background:none;border:1px solid var(--dsw-alias-border-l2);cursor:pointer;color:var(--dsw-alias-label-secondary);display:inline-flex;align-items:center;justify-content:center;width:22px;height:22px;border-radius:6px;padding:0;font-family:inherit}.dshbw_sidePopRefresh:hover{color:var(--dsw-alias-label-primary);border-color:var(--dsw-alias-border-l3);background:var(--dsw-alias-interactive-bg-hover)}.dshbw_side[data-peak=\"true\"]{border:1px solid var(--dsw-alias-state-warn-primary);background:color-mix(in srgb, var(--dsw-alias-state-warn-primary) 8%, var(--dsw-alias-bg-layer-2))}.dshbw_side[data-peak=\"false\"]{border:1px solid var(--dsw-alias-state-success-primary);background:color-mix(in srgb, var(--dsw-alias-state-success-primary) 8%, var(--dsw-alias-bg-layer-2))}.dshbw_sidePop[data-peak=\"true\"]{border-color:var(--dsw-alias-state-warn-primary)}.dshbw_sidePop[data-peak=\"false\"]{border-color:var(--dsw-alias-state-success-primary)}.dshbw_peakTag{display:inline-flex;align-items:center;gap:4px;font-size:11px;line-height:16px;padding:1px 6px;border-radius:999px}.dshbw_peakTag[data-peak=\"true\"]{color:var(--dsw-alias-state-warn-primary);background:var(--dsw-alias-state-warn-tertiary)}.dshbw_peakTag[data-peak=\"false\"]{color:var(--dsw-alias-state-success-primary);background:var(--dsw-alias-state-success-tertiary)}.dshbw_sessionNote{display:flex;align-items:center;gap:4px;margin-top:6px;padding-top:6px;border-top:1px solid var(--dsw-alias-border-l1);font-size:11px;line-height:16px;color:var(--dsw-alias-label-tertiary);min-width:0}.dshbw_sessionName{max-width:140px;overflow:hidden;text-overflow:ellipsis;white-space:nowrap}";
+		const css = ".dshbw_row{display:flex;justify-content:space-between;align-items:center;gap:12px;padding:2px 0}.dshbw_label{display:inline-flex;align-items:center;min-width:0}.dshbw_val{font-weight:600;color:var(--dsw-alias-label-primary)}.dshbw_val[data-tier=\"sec\"]{font-weight:400;color:var(--dsw-alias-label-secondary)}.dshbw_val[data-tier=\"ter\"]{font-weight:400;color:var(--dsw-alias-label-tertiary)}.dshbw_val[data-level=\"1\"]{color:var(--dsw-alias-state-warn-primary)}.dshbw_val[data-level=\"2\"]{color:var(--dsw-alias-state-error-primary)}.dshbw_err{color:var(--dsw-alias-state-error-primary);font-size:12px;line-height:16px;margin-top:6px}.dshbw_link{background:none;border:none;cursor:pointer;color:var(--dsw-alias-state-business-primary);display:inline-flex;align-items:center;gap:4px;padding:2px 6px;border-radius:6px;font-size:12px;line-height:18px;font-family:inherit;text-decoration:none}.dshbw_link:hover{color:var(--dsw-alias-state-business-primary);text-decoration:underline}.dshbw_sideTop{display:flex;align-items:center;justify-content:space-between;gap:8px;font-size:12px;line-height:18px}.dshbw_sideAmount{font-weight:600}.dshbw_sideAmount[data-level=\"1\"]{color:var(--dsw-alias-state-warn-primary);font-weight:600}.dshbw_sideAmount[data-level=\"2\"]{color:var(--dsw-alias-state-error-primary)}.dshbw_sideFoot{display:flex;justify-content:space-between;font-size:12px;line-height:16px;color:var(--dsw-alias-label-tertiary)}.dshbw_sideWrap{position:relative}.dshbw_side{display:flex;flex-direction:column;gap:4px;width:100%;padding:6px 8px;margin:0 8px 4px;border:none;border-radius:6px;background:var(--dsw-alias-bg-layer-2);color:var(--dsw-alias-label-primary);cursor:pointer;font-family:inherit;text-align:left;transition:transform 120ms cubic-bezier(0.23,1,0.32,1),background 120ms cubic-bezier(0.23,1,0.32,1)}.dshbw_side:active{transform:scale(.97)}.dshbw_side:hover{background:var(--dsw-alias-interactive-bg-hover)}.dshbw_sidePop{position:absolute;bottom:calc(100% + 8px);left:8px;z-index:70;min-width:240px;background:var(--dsw-alias-bg-layer-3);border:none;border-radius:12px;box-shadow:0 0 1px rgba(0,0,0,.3),0 4px 14px rgba(0,0,0,.1);padding:12px 12px;font-size:12px;line-height:20px;color:var(--dsw-alias-label-primary);animation:dshbw-pop-in 160ms cubic-bezier(0.23,1,0.32,1) backwards}.dshbw_sidePopHead{display:flex;align-items:center;justify-content:space-between;margin-bottom:6px}.dshbw_sidePopTitle{font-weight:600}.dshbw_sidePopClose{background:none;border:none;cursor:pointer;color:var(--dsw-alias-label-tertiary);font-size:16px;line-height:1;padding:0 2px;font-family:inherit}.dshbw_sidePopClose:hover{color:var(--dsw-alias-label-primary)}.dshbw_sidePopFoot{display:flex;align-items:center;justify-content:space-between;gap:8px;margin-top:10px}.dshbw_sidePopRefresh{background:none;border:1px solid var(--dsw-alias-border-l2);cursor:pointer;color:var(--dsw-alias-label-secondary);display:inline-flex;align-items:center;justify-content:center;width:22px;height:22px;border-radius:6px;padding:0;font-family:inherit}.dshbw_sidePopRefresh:hover{color:var(--dsw-alias-label-primary);border-color:var(--dsw-alias-border-l3);background:var(--dsw-alias-interactive-bg-hover)}.dshbw_side[data-peak=\"true\"]{border:1px solid var(--dsw-alias-state-warn-primary)}.dshbw_side[data-peak=\"false\"]{border:1px solid var(--dsw-alias-state-success-primary)}.dshbw_peakTag{display:inline-flex;align-items:center;gap:4px;font-size:12px;line-height:16px;padding:1px 7px;border-radius:6px}.dshbw_peakTag[data-peak=\"true\"]{color:var(--dsw-alias-state-warn-primary);background:var(--dsw-alias-state-warn-tertiary)}.dshbw_peakTag[data-peak=\"false\"]{color:var(--dsw-alias-state-success-primary);background:var(--dsw-alias-state-success-tertiary)}.dshbw_sessionNote{display:flex;align-items:center;gap:4px;margin-top:6px;padding-top:6px;border-top:1px solid var(--dsw-alias-border-l1);font-size:12px;line-height:16px;color:var(--dsw-alias-label-tertiary);min-width:0}.dshbw_sessionName{max-width:140px;overflow:hidden;text-overflow:ellipsis;white-space:nowrap}.dshbw_ico{color:var(--dsw-alias-state-business-primary);display:inline-flex;align-items:center}@keyframes dshbw-pop-in{from{opacity:0;transform:translateY(4px)}to{opacity:1;transform:translateY(0)}}.dshbw_sidePopRefresh,.dshbw_sidePopClose{transition:transform 120ms cubic-bezier(0.23,1,0.32,1),color 120ms cubic-bezier(0.23,1,0.32,1)}.dshbw_sidePopRefresh:active{transform:scale(.92)}.dshbw_sidePopClose:active{transform:scale(.92)}";
 		const tagId = "dsh-balance-widget/styles.css";
 		if (typeof document !== "undefined" && document.querySelector("style[data-plugin-css=" + JSON.stringify(tagId) + "]") === null) {
 			const tag = document.createElement("style");
@@ -18,18 +18,10 @@ window.__ModuleLoader__.load({
 			document.head.appendChild(tag);
 		}
 		var styles = {
-			"wrap": "dshbw_wrap",
-			"btn": "dshbw_btn",
-			"pop": "dshbw_pop",
-			"head": "dshbw_head",
-			"title": "dshbw_title",
-			"refresh": "dshbw_refresh",
 			"row": "dshbw_row",
 			"label": "dshbw_label",
 			"val": "dshbw_val",
 			"err": "dshbw_err",
-			"hint": "dshbw_hint",
-			"foot": "dshbw_foot",
 			"link": "dshbw_link",
 			"side": "dshbw_side",
 			"sideWrap": "dshbw_sideWrap",
@@ -47,85 +39,12 @@ window.__ModuleLoader__.load({
 			"sessionName": "dshbw_sessionName"
 		};
 		//#endregion
-		//#region lib/types/client/pricing.js
-		/**
-		* DeepSeek official pricing, in CNY per 1M tokens, effective 2026-08-17
-		* (peak/off-peak two-tier billing). Peak windows are Beijing time:
-		* 09:00-12:00 and 14:00-18:00; off-peak is half the peak price.
-		* Prices are hard-coded from the official announcement and are estimates
-		* for display — actual billing is authoritative from the provider.
-		*/
-		const PRICING = {
-			"deepseek-v4-flash": {
-				peak: { inputMiss: 3.0, inputHit: 0.10, output: 9.0 },
-				offPeak: { inputMiss: 1.5, inputHit: 0.05, output: 4.5 }
-			},
-			"deepseek-v4-pro": {
-				peak: { inputMiss: 9.0, inputHit: 0.30, output: 27.0 },
-				offPeak: { inputMiss: 4.5, inputHit: 0.15, output: 13.5 }
-			},
-			"deepseek-chat": {
-				peak: { inputMiss: 3.0, inputHit: 0.10, output: 9.0 },
-				offPeak: { inputMiss: 1.5, inputHit: 0.05, output: 4.5 }
-			},
-			"deepseek-reasoner": {
-				peak: { inputMiss: 9.0, inputHit: 0.30, output: 27.0 },
-				offPeak: { inputMiss: 4.5, inputHit: 0.15, output: 13.5 }
-			}
-		};
-		/** Fall back to Flash pricing for unknown model ids. */
-		function pricingFor(modelId) {
-			return PRICING[modelId] ?? PRICING["deepseek-v4-flash"];
-		}
-		/**
-		* Whether `date` falls inside a Beijing-time peak window.
-		* The host's local timezone is used; the widget is for a CN deployment.
-		*/
-		function isPeak(date) {
-			const hour = date.getHours();
-			return (hour >= 9 && hour < 12) || (hour >= 14 && hour < 18);
-		}
-		/**
-		* Price one session's token usage into a CNY estimate.
-		* @param usage - tokenUsage projection value.
-		* @param modelId - provider model id used for pricing.
-		* @returns { input, output, total } in CNY, or null when usage is absent.
-		*/
-		function priceSession(usage, modelId) {
-			if (usage === void 0 || usage === null) return null;
-			const input = usage.uncachedInputTokens ?? 0;
-			const hit = usage.cacheReadTokens ?? 0;
-			const write = usage.cacheWriteTokens ?? 0;
-			const output = usage.outputTokens ?? 0;
-			if (input + hit + write + output <= 0) return null;
-			const table = isPeak(new Date()) ? pricingFor(modelId).peak : pricingFor(modelId).offPeak;
-			const inputCost = (input * table.inputMiss + (hit + write) * table.inputHit) / 1e6;
-			const outputCost = (output * table.output) / 1e6;
-			return {
-				input: inputCost,
-				output: outputCost,
-				total: inputCost + outputCost
-			};
-		}
 		/** Format a CNY amount: ¥ + 3 significant decimals (drop trailing zeros). */
 		function formatCny(value) {
 			if (value === void 0 || value === null || !Number.isFinite(value)) return "—";
 			const rounded = Math.round(value * 1e4) / 1e4;
 			return `¥${rounded.toLocaleString("zh-CN", { minimumFractionDigits: 0, maximumFractionDigits: 4 })}`;
 		}
-		/** Format token counts compactly: 1234 -> 1.2K. */
-		function formatTokens(value) {
-			if (value === void 0 || value === null) return "—";
-			if (value >= 1e6) return `${(value / 1e6).toFixed(1)}M`;
-			if (value >= 1e3) return `${(value / 1e3).toFixed(1)}K`;
-			return String(value);
-		}
-		/** Format an epoch ms as HH:MM local time. */
-		function formatTime(ms) {
-			if (ms === void 0 || ms === null) return "—";
-			const d = new Date(ms);
-			return `${String(d.getHours()).padStart(2, "0")}:${String(d.getMinutes()).padStart(2, "0")}`;
-		}
 		//#endregion
 		//#region lib/types/client/balance-api.js
 		/** Same-origin balance route (host half registers it over ctx.webServer). */
@@ -354,8 +273,8 @@ window.__ModuleLoader__.load({
 							children: [(0, react_jsx_runtime.jsx)("div", {
 								className: styles.sideTop,
 								children: (0, react_jsx_runtime.jsxs)(react_jsx_runtime.Fragment, {
-									children: [(0, react_jsx_runtime.jsx)("span", {
-										children: "💰"
+									children: [(0, react_jsx_runtime.jsx)("span", { className: "dshbw_ico",
+										children: (0, react_jsx_runtime.jsx)("svg", { width: 16, height: 16, viewBox: "0 0 24 24", fill: "none", stroke: "currentColor", strokeWidth: 1.7, strokeLinecap: "round", strokeLinejoin: "round", children: [(0, react_jsx_runtime.jsx)("rect", { x: 3, y: 6, width: 18, height: 13, rx: 2 }), (0, react_jsx_runtime.jsx)("path", { d: "M3 10h18" })] })
 									}), (0, react_jsx_runtime.jsx)("span", {
 										className: styles.sideAmount,
 										"data-level": level > 0 ? level : undefined,
@@ -438,7 +357,7 @@ window.__ModuleLoader__.load({
 											})]
 										})
 									}), (0, react_jsx_runtime.jsx)("span", {
-										className: styles.val,
+										className: styles.val, "data-tier": "ter",
 										children: loading ? "…" : lastPrompt !== null ? formatCny(lastPrompt) : "—"
 									})]
 								})
@@ -481,7 +400,7 @@ window.__ModuleLoader__.load({
 											})]
 										})
 									}), (0, react_jsx_runtime.jsx)("span", {
-										className: styles.val,
+										className: styles.val, "data-tier": "sec",
 										children: loading ? "…" : todayWs !== null ? formatCny(todayWs) : "—"
 									})]
 								})
@@ -497,7 +416,7 @@ window.__ModuleLoader__.load({
 											})]
 										})
 									}), (0, react_jsx_runtime.jsx)("span", {
-										className: styles.val,
+										className: styles.val, "data-tier": "sec",
 										children: loading ? "…" : todayAll !== null ? formatCny(todayAll) : "—"
 									})]
 								})
@@ -554,7 +473,7 @@ window.__ModuleLoader__.load({
 			"warn.title.2": "余额低于 ¥1，即将耗尽！",
 			"warn.text.1": "⚠️ 余额不足 ¥5，建议及时充值。",
 			"warn.text.2": "⛔ 余额低于 ¥1，即将耗尽，请立即充值！",
-			"todayCostShort": "今日",
+			"todayCostShort": "今日·全部",
 			"peak.active": "峰时",
 			"peak.offpeak": "谷时",
 			"peak.tooltip.active": "当前为峰时（09:00-12:00、14:00-18:00，价格翻倍）：输入 ¥{input}/M · 输出 ¥{output}/M",
@@ -585,7 +504,7 @@ window.__ModuleLoader__.load({
 			"warn.title.2": "Balance below ¥1, almost exhausted!",
 			"warn.text.1": "⚠️ Balance below ¥5, consider topping up.",
 			"warn.text.2": "⛔ Balance below ¥1, almost exhausted — top up now!",
-			"todayCostShort": "Today",
+			"todayCostShort": "Today · all",
 			"peak.active": "Peak",
 			"peak.offpeak": "Off-peak",
 			"peak.tooltip.active": "Peak hours now (09:00-12:00, 14:00-18:00, 2× price): input ¥{input}/M · output ¥{output}/M",

package/lib/index.js

@@ -220,16 +220,31 @@ async function fetchBalance(ctx, config) {
 }
 
 /**
- * Pricing table (CNY per 1M tokens, official 2026-08-17 peak/off-peak).
- * Mirrors the client half; kept here so host-computed costs use the same
- * numbers. Mutable: syncOfficialPricing() may replace it with freshly parsed
- * official rates (falling back to these defaults on failure).
+ * Built-in pricing table (CNY per 1M tokens). Used before the first successful
+ * sync and as the fallback for any model the official page cannot be parsed
+ * for, so it must be kept current by hand: V4.1-Flash rates as of 2026-09-10.
+ * Mutable: syncOfficialPricing() replaces entries with freshly parsed rates.
+ */
+const FLASH_RATES = {
+	peak: { inputMiss: 2.0, inputHit: 0.04, output: 8.0 },
+	offPeak: { inputMiss: 1.0, inputHit: 0.02, output: 4.0 }
+};
+const PRO_RATES = {
+	peak: { inputMiss: 9.0, inputHit: 0.30, output: 27.0 },
+	offPeak: { inputMiss: 4.5, inputHit: 0.15, output: 13.5 }
+};
+/**
+ * "deepseek-flash" is the current model name; the older names below are
+ * retired but still accepted by the API and billed at the Flash rate
+ * (V4-Pro requests are routed to V4.1-Flash from 2026-09-14 12:00 CST).
  */
 const DEFAULT_PRICING = {
-	"deepseek-v4-flash": { peak: { inputMiss: 3.0, inputHit: 0.10, output: 9.0 }, offPeak: { inputMiss: 1.5, inputHit: 0.05, output: 4.5 } },
-	"deepseek-v4-pro": { peak: { inputMiss: 9.0, inputHit: 0.30, output: 27.0 }, offPeak: { inputMiss: 4.5, inputHit: 0.15, output: 13.5 } },
-	"deepseek-chat": { peak: { inputMiss: 3.0, inputHit: 0.10, output: 9.0 }, offPeak: { inputMiss: 1.5, inputHit: 0.05, output: 4.5 } },
-	"deepseek-reasoner": { peak: { inputMiss: 9.0, inputHit: 0.30, output: 27.0 }, offPeak: { inputMiss: 4.5, inputHit: 0.15, output: 13.5 } }
+	"deepseek-flash": FLASH_RATES,
+	"deepseek-v4-flash": FLASH_RATES,
+	"deepseek-v4-flash-vision-exp": FLASH_RATES,
+	"deepseek-v4-pro": PRO_RATES,
+	"deepseek-chat": FLASH_RATES,
+	"deepseek-reasoner": PRO_RATES
 };
 let pricingTable = { ...DEFAULT_PRICING };
 /** Source of the current pricing: "default" | "synced". */
@@ -246,9 +261,15 @@ const PRICING_ALIASES = {
 	"deepseek-official": "deepseek-v4-flash"
 };
 
-/** Whether `date` falls inside a Beijing-time peak window. */
+/**
+ * Whether `date` falls inside a Beijing-time (UTC+8) peak window.
+ * DeepSeek's peak pricing applies to Beijing time 09:00-12:00 and 14:00-18:00
+ * regardless of the host's timezone, so the window is computed in UTC+8
+ * explicitly instead of the machine's local time.
+ */
 function isPeak(date) {
-	const hour = date.getHours();
+	const beijing = new Date(date.getTime() + 8 * 60 * 60 * 1000);
+	const hour = beijing.getUTCHours();
 	return (hour >= 9 && hour < 12) || (hour >= 14 && hour < 18);
 }
 
@@ -303,57 +324,74 @@ async function syncOfficialPricing() {
 }
 
 /**
- * Best-effort parse of the official pricing page HTML. Looks for the standard
- * "deepseek-v4-flash" / "deepseek-v4-pro" model blocks and per-million-token
- * rates. This is heuristic: it tolerates missing fields and returns a partial
- * table rather than failing the whole sync.
+ * Pricing keys served by the page's leftmost (Flash) model column. Retired
+ * names are still billed at the Flash rate, so they share its entries.
+ */
+const FLASH_PRICING_KEYS = ["deepseek-flash", "deepseek-v4-flash", "deepseek-v4-flash-vision-exp", "deepseek-chat"];
+
+/**
+ * Best-effort parse of the official pricing page HTML. Only the leftmost model
+ * column is readable by {@link parseRates}, i.e. the Flash-tier column, so its
+ * rates are applied to every flash-family pricing key. Heuristic: returns a
+ * partial table rather than failing the whole sync.
  */
 function parsePricingPage(html) {
+	const block = extractPricingTable(html);
+	if (block === null) return {};
+	const rates = parseRates(block);
+	if (rates === null) return {};
 	const out = {};
-	const modelIds = ["deepseek-v4-flash", "deepseek-v4-pro"];
-	for (const model of modelIds) {
-		const block = extractModelBlock(html, model);
-		if (block === null) continue;
-		const rates = parseRates(block);
-		if (rates !== null) out[model] = rates;
-	}
+	for (const key of FLASH_PRICING_KEYS) out[key] = rates;
 	return out;
 }
 
-/** Extract the HTML section mentioning a model id (crude but robust). */
-function extractModelBlock(html, model) {
-	const idx = html.indexOf(model);
+/**
+ * Extract the pricing-table region: a window around the first per-metric row
+ * label. Anchoring on the row label rather than on a model id keeps the sync
+ * alive across DeepSeek's model renames — anchoring on "deepseek-v4-flash"
+ * silently broke once that id survived only in a footnote *after* the table.
+ */
+function extractPricingTable(html) {
+	const idx = html.indexOf("缓存命中");
 	if (idx < 0) return null;
-	const start = Math.max(0, idx - 400);
-	const end = Math.min(html.length, idx + 3000);
-	return html.slice(start, end);
+	return html.slice(Math.max(0, idx - 200), Math.min(html.length, idx + 1600));
 }
 
 /**
- * Parse per-1M-token rates from a block of HTML text. Looks for numbers near
- * "输入"/"缓存命中"/"输出" (CN page) or "input"/"cache"/"output" (EN page),
- * preferring the off-peak row. Returns { offPeak, peak } or null.
+ * Parse per-1M-token rates from a block of HTML text. The pricing table lists
+ * rows per metric with "空闲时段 <off-peak> … 高峰时段 <peak>" columns (CN) or
+ * "off-peak … peak" (EN); the first model column is the one we price. Each
+ * metric is captured as an (offPeak, peak) pair from its labelled row.
+ * Returns { offPeak, peak } or null when any off-peak figure is missing
+ * (the caller then falls back to the built-in rate table rather than silently
+ * pricing cache hits at 0).
  */
 function parseRates(block) {
 	// Strip tags to plain text for simpler matching.
 	const text = block.replace(/<[^>]+>/g, " ").replace(/\s+/g, " ").trim();
-	const num = (pattern) => {
-		const m = text.match(pattern);
-		if (!m) return void 0;
-		const v = parseFloat(m[1].replace(/[¥$,\s]/g, ""));
-		return Number.isFinite(v) ? v : void 0;
+	const pair = (label) => {
+		// label + short gap + "空闲时段 <n>" + gap + "高峰时段 <n>" (first model
+		// column). Also tolerate EN "off-peak … peak".
+		const re = new RegExp(
+			label + "[\\s\\S]{0,120}?(?:空闲时段|off[- ]?peak)\\s*([\\d.]+)[\\s\\S]{0,80}?(?:高峰时段|peak)\\s*([\\d.]+)",
+			"i"
+		);
+		const m = text.match(re);
+		if (!m) return null;
+		const a = parseFloat(m[1]);
+		const b = parseFloat(m[2]);
+		return Number.isFinite(a) && Number.isFinite(b) ? [a, b] : null;
 	};
-	// Try patterns for off-peak (空闲/off-peak) first, then any.
-	const offPeakInputMiss = num(/缓存未命中[^\d]*([\d.]+)/) ?? num(/cache[\s-]*miss[^\d]*([\d.]+)/);
-	const offPeakInputHit = num(/缓存命中[^\d]*([\d.]+)/) ?? num(/cache[\s-]*hit[^\d]*([\d.]+)/);
-	const offPeakOutput = num(/输出[^\d]*([\d.]+)/) ?? num(/output[^\d]*([\d.]+)/);
-	if (offPeakInputMiss === void 0 || offPeakOutput === void 0) return null;
-	const peakInputMiss = (offPeakInputMiss ?? 0) * 2;
-	const peakInputHit = (offPeakInputHit ?? 0) * 2;
-	const peakOutput = (offPeakOutput ?? 0) * 2;
+	const hit = pair("缓存命中");
+	const miss = pair("缓存未命中");
+	const out = pair("百万tokens输出") ?? pair("输出");
+	if (miss === null || out === null) return null;
+	// Cache-hit row missing is acceptable only if it cannot be parsed at all —
+	// reject rather than guess zero.
+	if (hit === null) return null;
 	return {
-		peak: { inputMiss: peakInputMiss, inputHit: peakInputHit, output: peakOutput },
-		offPeak: { inputMiss: offPeakInputMiss, inputHit: offPeakInputHit ?? 0, output: offPeakOutput }
+		offPeak: { inputMiss: miss[0], inputHit: hit[0], output: out[0] },
+		peak: { inputMiss: miss[1], inputHit: hit[1], output: out[1] }
 	};
 }
 
@@ -381,19 +419,61 @@ function normalizeSessionId(sessionId) {
 	return sessionId.startsWith("session-") ? sessionId.slice("session-".length) : sessionId;
 }
 
+/**
+ * Canonical session log basenames (DSH's @deepseek-ai/dsh-session-format):
+ * generation 0 is "session.jsonl.zstd"; later generations carry a lowercase
+ * ".vN" tag, e.g. "session.v3.jsonl.zstd" (Session format v3, DSH 0.1.5+).
+ */
+const SESSION_LOG_RE = /^session(?:\.v([1-9][0-9]*))?\.jsonl\.zstd$/;
+
+/** Session format generation of a canonical log name, or undefined if not one. */
+function sessionLogVersion(filename) {
+	const match = SESSION_LOG_RE.exec(filename);
+	if (match === null) return void 0;
+	return match[1] === void 0 ? 0 : Number(match[1]);
+}
+
+/**
+ * Collect the newest-generation log of every session directory, walking the
+ * sessions tree. A session that migrated formats keeps its older artifact on
+ * disk, and the newer generation is a full rewrite that already contains those
+ * rows — so only the highest generation per directory may be read. Reading
+ * both would double-count the whole session.
+ */
+function collectNewestSessionLogs(fs) {
+	const out = [];
+	const walk = (dir) => {
+		let entries;
+		try {
+			entries = fs.readdirSync(dir, { withFileTypes: true });
+		} catch {
+			return;
+		}
+		let newest;
+		let newestVersion = -1;
+		for (const entry of entries) {
+			if (entry.isDirectory()) {
+				walk(`${dir}/${entry.name}`);
+				continue;
+			}
+			const version = sessionLogVersion(entry.name);
+			if (version === void 0 || version <= newestVersion) continue;
+			newestVersion = version;
+			newest = `${dir}/${entry.name}`;
+		}
+		if (newest !== void 0) out.push(newest);
+	};
+	walk(sessionsRoot());
+	return out;
+}
+
 /** Resolve the current session's JSONL path by walking the sessions tree. */
 async function resolveSessionFile(sessionId) {
 	if (sessionId === void 0 || sessionId === "") return void 0;
-	const bareId = normalizeSessionId(sessionId);
-	const root = sessionsRoot();
+	const marker = `/session-${normalizeSessionId(sessionId)}/`;
 	try {
 		const fs = await awaitImportFs();
-		const dirs = fs.readdirSync(root, { withFileTypes: true });
-		for (const dir of dirs) {
-			if (!dir.isDirectory()) continue;
-			const candidate = `${root}/${dir.name}/session-${bareId}/session.jsonl.zstd`;
-			if (fs.existsSync(candidate)) return candidate;
-		}
+		return collectNewestSessionLogs(fs).find((path) => path.includes(marker));
 	} catch {
 		/* ignore filesystem errors */
 	}
@@ -531,31 +611,22 @@ function todaySessionCost(samples) {
  * Walks the sessions tree and picks by mtime; returns the path or undefined.
  */
 async function findMostRecentSessionFile() {
-	const root = sessionsRoot();
 	try {
 		const fs = await awaitImportFs();
 		const { statSync } = fs;
 		let best = void 0;
 		let bestMtime = -1;
-		const walk = (dir) => {
-			for (const entry of fs.readdirSync(dir, { withFileTypes: true })) {
-				const full = `${dir}/${entry.name}`;
-				if (entry.isDirectory()) {
-					walk(full);
-				} else if (entry.name === "session.jsonl.zstd") {
-					try {
-						const mtime = statSync(full).mtimeMs;
-						if (mtime > bestMtime) {
-							bestMtime = mtime;
-							best = full;
-						}
-					} catch {
-						/* ignore unreadable files */
-					}
+		for (const file of collectNewestSessionLogs(fs)) {
+			try {
+				const mtime = statSync(file).mtimeMs;
+				if (mtime > bestMtime) {
+					bestMtime = mtime;
+					best = file;
 				}
+			} catch {
+				/* ignore unreadable files */
 			}
-		};
-		walk(root);
+		}
 		return best;
 	} catch {
 		return void 0;
@@ -587,31 +658,48 @@ function todayCost(sessionList, cwdFilter) {
 	return { cost, inputTokens: input, outputTokens: output };
 }
 
-/** Walk every session file under the sessions root, decompress, and parse. */
+/**
+ * Decompressed session parse cache, keyed by path and validated against the
+ * file's mtime. Long-lived: an untouched file is never re-decompressed, so a
+ * cold /today-cost after the outer 30s TTL expires only re-reads files that
+ * actually changed. Cleared wholesale when it outgrows a bound.
+ */
+const sessionParseCache = new Map();
+const SESSION_PARSE_CACHE_MAX = 600;
+
+/** Decompress+parse one file, honoring the (path,mtime) cache. */
+async function parseSessionFileCached(path, mtimeMs) {
+	const hit = sessionParseCache.get(path);
+	if (hit !== void 0 && hit.mtimeMs === mtimeMs) return hit.parsed;
+	const result = await decompressSession(path);
+	if (!result.ok) return null;
+	const parsed = parseSessionUsages(result.text);
+	if (sessionParseCache.size >= SESSION_PARSE_CACHE_MAX) sessionParseCache.clear();
+	sessionParseCache.set(path, { mtimeMs, parsed });
+	return parsed;
+}
+
+/**
+ * Walk every session file under the sessions root and parse the ones that can
+ * contribute to "today" numbers. Only files whose mtime falls on the current
+ * local calendar day can contain today's samples, so older files are skipped
+ * entirely (avoids decompressing the whole ~/.dsh/sessions tree on every cold
+ * call). Parsing runs in parallel; results are cached by (path, mtime).
+ */
 async function collectAllSessionSamples() {
-	const root = sessionsRoot();
 	const fs = (await awaitImportFs());
-	const files = [];
-	try {
-		const walk = (dir) => {
-			for (const entry of fs.readdirSync(dir, { withFileTypes: true })) {
-				const full = `${dir}/${entry.name}`;
-				if (entry.isDirectory()) walk(full);
-				else if (entry.name === "session.jsonl.zstd") files.push(full);
-			}
-		};
-		walk(root);
-	} catch {
-		return [];
-	}
-	const perSession = [];
+	const files = collectNewestSessionLogs(fs);
+	const now = new Date();
+	const dayStart = new Date(now.getFullYear(), now.getMonth(), now.getDate()).getTime();
+	const candidates = [];
 	for (const file of files) {
-		const result = await decompressSession(file);
-		if (!result.ok) continue;
-		const parsed = parseSessionUsages(result.text);
-		perSession.push(parsed);
+		try {
+			const mtimeMs = fs.statSync(file).mtimeMs;
+			if (mtimeMs >= dayStart) candidates.push({ path: file, mtimeMs });
+		} catch { /* file vanished mid-walk */ }
 	}
-	return perSession;
+	const parsed = await Promise.all(candidates.map((c) => parseSessionFileCached(c.path, c.mtimeMs)));
+	return parsed.filter((p) => p !== null);
 }
 
 /** Simple in-memory TTL cache for cost computations (30s). */
@@ -782,7 +870,7 @@ function makeRoutes(ctx, config) {
 					const segments = parsed.cwd.split("/").filter((s) => s !== "");
 					workspaceName = segments.length > 0 ? segments[segments.length - 1] : parsed.cwd;
 				}
-				// Session id from the file path (…/session-<id>/session.jsonl.zstd).
+				// Session id from the file path (…/session-<id>/session[.vN].jsonl.zstd).
 				let fileSessionId = void 0;
 				const sidMatch = file.match(/session-([0-9a-f-]+)/);
 				if (sidMatch !== null) fileSessionId = sidMatch[1];

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "dsh-balance-widget",
   "description": "Balance & cost widget for the dsh web GUI: a sidebar footer card shows the DeepSeek account balance and today's costs; clicking opens a popover with a five-tier cost breakdown (last prompt with session name / today-this-session / today-this-workspace / today-all-workspaces), peak/off-peak status tags, term explanations, and a one-click top-up link. Zero external dependencies — works on Node 24.",
-  "version": "0.5.0",
+  "version": "0.5.2",
   "type": "module",
   "main": "lib/index.js",
   "files": [
@@ -33,7 +33,8 @@
     }
   },
   "peerDependencies": {
-    "react": "^18.2.0"
+    "react": "^18.2.0",
+    "@deepseek-ai/dsh": ">=0.1.2-rc.1 <0.2.0"
   },
   "keywords": [
     "deepseek",
@@ -49,5 +50,9 @@
   "repository": {
     "type": "git",
     "url": "git+https://github.com/LL-cmyk-so/dsh-balance-widget.git"
-  }
+  },
+  "engines": {
+    "node": "^22.19.0 || >=24.0.0"
+  },
+  "packageManager": "pnpm@11.7.0"
 }