XiaoLuoMin/dsh-resource-monitor

A DeepSeek Harness (dsh) sidebar plugin that samples your Windows machine once a minute, keeps a 30-day history, and tells you when sustained CPU load, memory pressure, or frequency throttling actually happened — instead of showing a live gauge you have to stare at.

DSH 侧栏插件：每分钟采集一次本机资源，保留 30 天历史，直接用「连续多久超阈值」判定长期高占用、内存吃紧与降频，而不是只给一个需要一直盯着看的实时仪表盘。

What it does / 功能

• Sidebar tab 资源监控; a matching entry button also sits in the session title bar. / 侧栏 tab「资源监控」，会话标题栏也有入口按钮。
• Panel order: KPI row, conclusion box, CPU chart (85% alert line), memory chart (90% alert line), frequency chart (70% throttle line), incident list. / 面板顺序：KPI、结论框、CPU 曲线（85% 告警线）、内存曲线（90% 告警线）、频率曲线（70% 降频线）、异常列表。
• KPI row: current CPU, current memory (used / total GB), current frequency. / KPI：当前 CPU、当前内存（已用/总量 GB）、当前频率。
• Summary: range average/peak, minutes spent at ≥85% CPU, minutes at ≥90% memory, throttled minutes. / 结论：区间平均/峰值、超阈值累计分钟、降频分钟。
• Incident list (last 7 days): CPU sustained load, memory near capacity, frequency held down under load. / 异常列表（近 7 天）：CPU 长时间高占用、内存接近上限、负载下降频。
• 24h / 7d toggle, auto-refresh every 30 s. / 24 小时 / 7 天切换，30 秒自动刷新。

Detection rules / 判定规则

Item / 项目	Rule / 规则
Sample interval / 采样间隔	60 s
Retention / 历史保留	30 days (samples.jsonl, append-only)
CPU alert	≥ 85% for ≥ 10 consecutive minutes
Memory alert	≥ 90% for ≥ 10 consecutive minutes
Throttle alert	CPU ≥ 50% while frequency ≤ 70% of maximum, for ≥ 5 consecutive minutes

Data sources / 数据来源

• CPU / memory: Node os.cpus() time deltas, os.totalmem() / os.freemem().
• Frequency: Windows performance counter \Processor Information(_Total)\% of Maximum Frequency.
• Disk: \PhysicalDisk(_Total)\Disk Read Bytes/sec, Disk Write Bytes/sec (still sampled into the history, not charted). / 磁盘读写仍写入历史，但不再绘图。
• Network: Get-NetAdapterStatistics cumulative bytes, converted to a rate between two samples (history only, not charted). / 网络同样只入库不绘图。
• Disk and network figures come from one powershell.exe child process per sample (~2–4 s, async, never blocks the UI).

No temperature readings / 不采集温度

Reading CPU core temperature on Windows (MSR / SMBus) needs a kernel driver. This plugin does not ship or load one; it deliberately infers throttling from the frequency / maximum-frequency ratio, which reflects the throttling event itself and needs no driver or elevation.

Windows 上读取 CPU 核心温度需要内核驱动。本插件不附带、也不加载任何驱动，改用「当前频率 / 最高频率」判定降频 —— 这个信号直接反映降频事实，且不需要驱动或提权。

Install / 安装

dsh plugin --profile desktop add github:XiaoLuoMin/dsh-resource-monitor

Then restart the DSH app so both the host and the client half load. / 装完重启 DSH 应用（host 与 client 两半都要重新加载）。

Uninstall / 卸载:

dsh plugin --profile desktop remove dsh-resource-monitor

Data location / 数据位置

%USERPROFILE%\.dsh\data\resource-monitor\samples.jsonl

One JSON object per line, one line per minute. Delete the file to reset the history. / 每行一个 JSON 样本，每分钟一行；删除该文件即清空历史。

Layout / 结构

lib/index.js        host: sampling loop, 30-day history, POST /api/resource-monitor.snapshot
client/client.js    client: sidebar tab, entry button, separate CPU / memory / frequency canvas charts, summary and incident list
cordis.patch.yml    bundle insert declaration

License

MIT — see LICENSE.

Changelog / 版本记录

• 1.1.2 — 结论框移到 CPU 曲线之上；面板顺序：KPI → 结论 → CPU → 内存 → 频率 → 事件。
• 1.1.1 — CPU 与内存拆成两张独立曲线（各带 85% / 90% 告警线），移除磁盘与网络曲线。
• 1.1.0 — 插件更名为 dsh-resource-monitor（原 dsh-resource-alerts）。
• 1.0.2 — 修复结论行的重复百分号；频率卡片增加 CPU 叠加线与 70% 降频基准线。
• 1.0.1 — 修复 exports 缺少根入口导致 DSH 无法导入。