release: v0.6.0 — /status calibration, status tray, logging, accuracy fixes

Features:
- /status calibration for Claude usage: enter the session % from Claude
  Code's /status and TokenWatch back-solves your real effective limit so
  the displayed percentage tracks it (Settings -> Claude plan).
- Status-colored tray icon (green/amber/red) on Windows/Linux, with a
  make-tray-icons script; falls back to the plain icon when absent.
- Persistent main-process log at ~/.tokenwatch/logs/main.log (1 MB
  rotation) wired to global + IPC error paths; "Open logs" button in
  Settings.
- Projected 30-day cost tile in Analytics.

Fixes:
- Stop showing fabricated demo data when there is no session history;
  render an honest zero state instead.
- Ctrl/Cmd+2 no longer opens a blank screen; number shortcuts now map to
  the four visible tabs (Dashboard/Analytics/Terminal/Settings).
- Surface in-app refresh/save toasts (were pushed into an unrendered
  queue and dropped).
- Codex context-window % used cumulative session tokens (showed ~2890%);
  use the latest turn, clamped to 100%.
- Analytics summary tiles now follow the 7d/30d toggle; keep the final
  X-axis label on 30d.
- sessionTracker sorts sessions newest-first.

Stability / cleanup:
- Async Codex session file I/O (off the main thread) so a large archive
  doesn't block the UI on its 15s refresh.
- Deterministic teardown on quit (clear timers + terminate ccusage worker).
- Remove dead code: mock-data generators, unused LiveMonitoring and
  NotificationSystem components.

Author: sooua

CHANGELOG.md

@@ -9,6 +9,49 @@ Pre-v0.1.0 history is the CCSeva lineage ([Iamshankhadeep/ccseva](https://github
 
 ## [Unreleased]
 
+## [0.6.0] — 2026-06-18
+### Added
+- **Persistent main-process log** at `~/.tokenwatch/logs/main.log`
+  (1 MB rotation). The global `uncaughtException` / `unhandledRejection`
+  handlers, startup, single-instance denials, and update/IPC errors now
+  write here — packaged GUI builds previously discarded all console
+  output. A new **Open logs** button in Settings opens the folder.
+- **`/status` calibration** for Claude usage. Plan token limits aren't
+  published, so the percentage is an estimate; entering the session %
+  from Claude Code's `/status` back-solves your real effective limit so
+  future percentages track it. Settings → Claude plan → *Calibrate
+  against /status*; clearable anytime.
+- **Status-colored tray icon** on Windows/Linux — the mark's accent
+  turns green / amber / red at the safe / warning / critical
+  thresholds (run `npm run make-tray-icons` to generate the assets;
+  falls back to the plain icon when absent).
+- **Projected 30-day cost** tile in Analytics.
+
+### Fixed
+- **No more fabricated numbers.** With no session history the dashboard
+  showed demo data ($2.45 / 850 tokens / random charts); it now renders
+  an honest zero state.
+- **Ctrl/Cmd+2 no longer opens a blank screen** — the unused `live`
+  route is gone and the number shortcuts now map to the four visible
+  tabs (1 Dashboard · 2 Analytics · 3 Terminal · 4 Settings).
+- **In-app feedback toasts are shown again.** Refresh / settings-save
+  notifications were pushed into a queue that nothing rendered; they
+  now route through the toaster.
+- **Codex context-window % was nonsensical** (cumulative session tokens
+  ÷ context window → e.g. 2890%); it now uses the latest turn's tokens,
+  clamped to 100%.
+- **Analytics summary tiles now follow the 7d/30d toggle** (were hardcoded
+  to 7 days), and the final day's X-axis label is no longer dropped on 30d.
+
+### Changed
+- **Codex session parsing moved off the main thread** (async file I/O) so
+  a large `~/.codex/sessions` archive no longer blocks the tray/UI on its
+  15-second refresh.
+- **Deterministic teardown on quit** — all timers are cleared and the
+  ccusage worker thread is terminated instead of relying on process exit.
+- Removed dead code: mock-data generators and the unused `LiveMonitoring`
+  / `NotificationSystem` components.
+
 ## [0.5.6] — 2026-04-19
 ### Added
 - **Installed version readout** next to the "Check for updates now"

README.md

@@ -152,6 +152,30 @@ The app reads your Claude Code usage history from `~/.claude/projects/**/*.jsonl
 | Window anchor | Top-right, near the menu bar | Near the tray icon, clamped to the active display |
 | Notifications | Native Notification Center | Toast center (requires `AppUserModelId`, set automatically) |
 
+## FAQ
+
+### Why doesn't the percentage match Claude Code's `/status`?
+
+Because it **can't be derived exactly from local data.** Claude Code's JSONL logs only store raw per-message token counts (`input` / `output` / `cache_read` / `cache_creation`). The percentages, remaining quota, and reset times that `/status` shows are computed **server-side by Anthropic** (using an undisclosed, model-weighted formula and unpublished limits) and are only returned in API response headers — they are **never written to disk**.
+
+So TokenWatch estimates the percentage as `tokens used ÷ plan limit`, where:
+
+- **tokens used** excludes cache-read tokens. On a cache-heavy session these can be 95%+ of all tokens — including them would inflate the number ~10–100×, so they're left out (they're also weighted far below 1× by Anthropic's rate limiter).
+- **plan limit** is a community-estimated value (Pro / Max5 / Max20), not an official figure.
+
+Both differ from Anthropic's own method, so the raw estimate will not line up with `/status`.
+
+### How do I make the percentage track `/status`?
+
+Use **calibration** (Settings → Claude plan → *Calibrate against /status*):
+
+1. Run `/status` in Claude Code and note the **current session percentage** it shows.
+2. Enter that number in the calibration box and click **Calibrate**.
+
+TokenWatch back-solves your real effective limit (`limit = tokens used ÷ percentage`) and remembers it, so future percentages track `/status`. It's a linear fit — most accurate near your current usage level, and you can **Clear** it anytime to return to the estimate.
+
+> **Codex is different:** Codex CLI writes its authoritative rate-limit percentages (5-hour and weekly windows) into its session logs, so the Codex card's percentages already match `codex` `/status` with no calibration needed.
+
 ## Requirements
 
 - macOS 10.15+ **or** Windows 10/11 (x64)

README.zh-CN.md

@@ -152,6 +152,30 @@ Remove-Item -Recurse -Force "$env:LOCALAPPDATA\electron-builder\Cache\winCodeSig
 | 窗口锚定 | 右上角，靠近菜单栏 | 托盘图标附近，自动限制在当前显示器范围内 |
 | 通知 | 原生通知中心 | Toast 通知（需 `AppUserModelId`，已自动设置） |
 
+## 常见问题
+
+### 为什么百分比和 Claude Code 的 `/status` 对不上？
+
+因为**它无法从本地数据精确复现**。Claude Code 的 JSONL 日志只记录每条消息的原始 token 数（`input` / `output` / `cache_read` / `cache_creation`）。`/status` 显示的百分比、剩余额度和重置时间是 Anthropic **在服务器端**算出来的（用的是未公开、按模型加权的公式和未公布的额度），只在 API 响应头里返回 —— **从不落盘**。
+
+所以 TokenWatch 用 `已用 token ÷ 套餐额度` 来估算百分比，其中：
+
+- **已用 token** 不含缓存读取（cache-read）token。在缓存密集的会话里它能占到全部 token 的 95% 以上，算进去会让数字膨胀约 10–100 倍，因此排除（它在 Anthropic 的限流里权重也远低于 1×）。
+- **套餐额度** 是社区估算值（Pro / Max5 / Max20），并非官方数字。
+
+这两点都和 Anthropic 自己的算法不同，所以原始估算值不会和 `/status` 对齐。
+
+### 怎样让百分比贴近 `/status`？
+
+使用**校准**功能（设置 → Claude 套餐 → *按 /status 校准*）：
+
+1. 在 Claude Code 里运行 `/status`，记下它显示的**当前会话百分比**。
+2. 把这个数字填进校准框，点击**校准**。
+
+TokenWatch 会反推出你真实的有效额度（`额度 = 已用 token ÷ 百分比`）并记住它，之后的百分比就会贴着 `/status` 走。这是线性拟合 —— 在你当前用量水平附近最准；随时可点**清除**恢复为估算值。
+
+> **Codex 不一样：** Codex CLI 会把官方的限流百分比（5 小时窗口与每周窗口）写进自己的会话日志，所以 Codex 卡片的百分比无需校准就已经和 `codex` 的 `/status` 一致。
+
 ## 环境要求
 
 - macOS 10.15+ **或** Windows 10/11 (x64)