dsh-ssh-tui 0.3.2 → 0.3.4

package/README.en.md

@@ -1,12 +1,13 @@
 # dsh-ssh-tui
 
 A DeepSeek Harness terminal for jump hosts, headless servers, and high-latency
-SSH. Plain ANSI and incremental redraws. No browser, no Ink/React skin.
+SSH. Plain ANSI and incremental redraws. No browser required.
 
 中文部署指南：[README.md](README.md)
 
-Who it is for: a keyboard-only SSH session on a remote box. Who it is not for:
-a local desktop TUI with many themes — that lane already has a clear winner.
+If you mostly work over SSH — a jump host, a test box, a keyboard-only
+session — start here. A local desktop terminal with themes and layout
+you already like can stay as it is.
 
 ## Official headless vs this TUI
 
@@ -27,6 +28,19 @@ own card, the plan strip pinned above the input.
 
 Singles: [headless stdout](docs/screenshots/headless.png) · [dsh-ssh-tui](docs/screenshots/workspace.png)
 
+## Still visible on a slow SSH pipe
+
+Same task, replayed at **2 kB/s** through the real incremental painter
+(88×30, one `stdout.write` per frame). Official headless on that pipe
+would stay blank until the final markdown. Here reasoning, the edit
+diff, subagent cards, and the plan strip appear as the bytes arrive.
+
+![Same task replayed at 2 kB/s SSH](docs/screenshots/slow-link.gif)
+
+Reproducible, no model in the loop: `npm run screenshots:slow` writes
+`docs/screenshots/slow-link.json`. This capture is 14 paints, about
+**15.5 KB**, **7.6 s** at 2 kB/s. Byte ledger for this event sequence.
+
 ## Requirements
 
 - Node.js >= 22.19
@@ -175,7 +189,9 @@ The plugin runs on Linux, macOS, and Windows (Node ≥ 22.19):
 Type `/` to see slash-command suggestions — the panel merges the TUI's own
 commands (`/find`, `/model`, `/help`, ...) with every command the harness
 registers (`/goal`, `/plan`, `/compact`, `/permission`, `/feedback`, ...).
-`Tab` completes, `Enter` runs. `/help` lists everything.
+`/compact` shows a spinning 「压缩上下文」 card and footer until it
+finishes, then the tokens recovered. `Tab` completes, `Enter` runs.
+`/help` lists everything.
 
 `/model` lists models for the **current** provider first. On SuperGrok that
 is `grok-4.6` / `grok-4.5` plus reasoning effort (`xhigh` on 4.6). Switching
@@ -220,14 +236,17 @@ Interrupted streaming output keeps the already-generated prefix and is marked
 system messages. Harness slash commands that accept image attachments are
 labelled `(images ok)` in the command list and completion hints.
 
-`/usage` (alias `/quota`) works when the current provider is an OpenCode
-source and keeps the two billing models distinct:
+`/usage` (alias `/quota`) follows the **current** provider:
 
-- **OpenCode Go** queries the official quota endpoint and shows rolling
-  5-hour / weekly / monthly usage percentages, limit state, and reset times;
-- **OpenCode Zen** is metered per API bill and has no fixed quota, so the TUI
-  points to `https://opencode.ai/zen` for balance/billing and shows the
-  session token usage it has recorded instead of inventing a quota.
+- **SuperGrok** reads `GET cli-chat-proxy.grok.com/v1/billing` (weekly remaining %);
+- **OpenCode Go** reads the official `/v1/usage` windows (5-hour / week / month);
+- **OpenCode Zen** is metered — the TUI points at `https://opencode.ai/zen`.
+
+Quota is fetched at startup. Recheck cadence follows the tightest window:
+every 10 turns for a 5-hour cap (every 4 when near a threshold), every 50
+for weekly (every 10 when near), every 80 for monthly (every 20 when near).
+Crossing 50% / 25% / 10% / 5% remaining posts a ⚠ planning reminder. The
+footer shows the tightest window.
 
 The startup screen shows the official DeepSeek whale logo (rendered from the
 harness favicon) in the DeepSeek brand color, with the wordmark below it. The
@@ -323,12 +342,10 @@ switching while running.
 ## Jump-host / proxied SSH
 
 Each paint is one `stdout.write` of dirty rows only, so a jump host or
-corporate proxy does not see one SSH packet per line. Default cadence is
-about 160 ms, enough for most jump hosts. On a very slow path:
-
-```sh
-export DSH_TUI_PAINT_MS=250   # 40–1000; higher = fewer packets, choppier stream
-```
+corporate proxy does not see one SSH packet per line. Local ttys use 80 ms.
+Over SSH the TUI probes CSI 6n once and picks 80 / 160 / 250 / 400 ms from
+the round-trip. `DSH_TUI_PAINT_MS` always wins (40–1000). `/status` and the
+footer show the active tier.
 
 ## Development
 

package/README.md

@@ -1,12 +1,12 @@
 # dsh-ssh-tui
 
 给跳板机、无桌面服务器、高延迟 SSH 用的 DeepSeek Harness 终端。纯 ANSI、增量重绘，
-不需要浏览器，也不需要 Ink / React 那套皮肤。
+不需要浏览器。
 
 English: [README.en.md](README.en.md)
 
-适合谁：SSH 直连的远程机器、慢链路、只有键盘的会话。不适合谁：本机想要多主题、
-官网级交互的桌面终端——那条赛道已经有明确赢家。
+如果你主要在 SSH 里写代码——公司跳板、测试机、只有键盘的会话——可以从这里开始。
+本机桌面终端若更在意主题和布局，也可以继续用你已经习惯的界面。
 
 SuperGrok / X Premium 订阅走配套插件 [dsh-llm-xai-oauth](https://github.com/cyjyyd/dsh-llm-xai-oauth)，复用本机 grok-bridge token，不需要 xAI API Key。
 
@@ -23,6 +23,14 @@ SuperGrok / X Premium 订阅走配套插件 [dsh-llm-xai-oauth](https://github.c
 
 单独看：[headless stdout](docs/screenshots/headless.png) · [dsh-ssh-tui](docs/screenshots/workspace.png)
 
+## 弱网 SSH 上过程还在
+
+同一条任务，按 **2 kB/s** 限速回放真实增量绘制（88×30，一帧一次 `stdout.write`）。官方 headless 这条链路上只会在全部结束后突然打出最终 Markdown；这里思考、`edit` diff、子代理卡和计划条是随着字节到达逐步出现的。
+
+![2 kB/s SSH 上回放同一任务](docs/screenshots/slow-link.gif)
+
+协议（可复现，不靠模型估）：`npm run screenshots:slow` → `docs/screenshots/slow-link.json`。这次回放 14 次绘制、约 **15.5 KB**，在 2 kB/s 上大约 **7.6 s** 画完。数字是这条固定事件序的 stdout 字节账。
+
 ## 功能一览
 
 - 纯终端渲染，无需浏览器/鼠标/重量级终端框架，适合慢速或远程 SSH；
@@ -157,7 +165,9 @@ dsh --profile tui --no-color
 
 斜杠命令：`/help`、`/find`、`/model`、`/submodel`、`/subeffort`、`/mode`、`/resume`、
 `/status`、`/subagents`、`/usage`（`/quota` 同义）、`/setup`、`/clear`，
-以及 harness 自带命令（`/goal`、`/plan`、`/compact` 等）。harness 命令若声明
+以及 harness 自带命令（`/goal`、`/plan`、`/compact` 等）。`/compact` 进行中会显示
+「压缩上下文」卡片和底栏转圈，结束时写出回收的 token 数。模型请求失败会显示重试
+进度；会话标题由模型生成后写到窗口标题。harness 命令若声明
 支持图片附件，会在命令列表和补全提示中标注“可附图”。
 
 `/model` 默认列出**当前提供商**的模型。已经在 SuperGrok 时，直接选
@@ -192,12 +202,13 @@ SuperGrok / X Premium 走本机 OAuth，不需要填 Key。`/status` 和底栏
 `提问用户` 卡片。`/goal` 是折叠的 `目标` 卡片。`/find 思考 padAnsi` 或 `Alt+1..4`
 可跳到对应类别的最新卡片。
 
-`/usage` 在当前提供商为 OpenCode 源时可用，并区分两种计费方式：
+`/usage`（`/quota` 同义）按**当前提供商**查额度：
+
+- **SuperGrok**：`GET cli-chat-proxy.grok.com/v1/billing`，显示本周剩余%；
+- **OpenCode Go**：官方 `/v1/usage`，滚动 5 小时 / 本周 / 本月剩余%；
+- **OpenCode Zen**：按量计费、没有固定额度，提示到 `https://opencode.ai/zen`。
 
-- **OpenCode Go**：调用官方额度接口，显示滚动 5 小时 / 本周 / 本月用量
-  百分比、限流状态与重置时间；
-- **OpenCode Zen**：按 API 账单计费、没有固定额度，TUI 不假装查询余额，
-  只提示到 `https://opencode.ai/zen` 查看，并附本会话已记录的 token 用量。
+启动时查一次。之后按最紧窗口调查询：5 小时额度每 10 轮（接近阈值改 4 轮），周额度每 50 轮（接近改 10 轮），月额度每 80 轮（接近改 20 轮）。剩余跨过 50% / 25% / 10% / 5% 时用 ⚠ 提示合理规划。底栏显示最紧窗口的剩余百分比。
 
 ## 配置
 
@@ -290,10 +301,7 @@ npm run build
 - **标题栏或铃声不生效**：确认终端支持 OSC 0 与 BEL；铃声可用
   `DSH_TUI_NO_BELL=1` 关闭。
 - **滚轮误触取消**：已加入转义序列缓冲，网络拆包也不会把 `ESC` 当取消。
-- **跳板机 / 多层代理 SSH 发画**：每一帧只发脏行，并且拼成一次 `stdout.write`，避免一行一个 SSH 包。默认约 160 ms 合一帧，多数跳板机不用改。特别慢时：
-  ```bash
-  export DSH_TUI_PAINT_MS=250   # 40–1000；越大越省链路，流式字越顿
-  ```
+- **跳板机 / 多层代理 SSH 发画**：每一帧只发脏行，并且拼成一次 `stdout.write`。本机 80 ms；SSH 启动时用 CSI 6n 测往返，按 RTT 选 80/160/250/400 ms。`DSH_TUI_PAINT_MS` 始终优先（40–1000）。`/status` 和底栏显示当前档。
 
 ## License