glm-quota-line 1.2.0 → 1.3.0

package/CHANGELOG.md

@@ -1,5 +1,22 @@
 # Changelog
 
+## 1.3.0
+
+- Added `commands` subcommand: list every command with a machine-readable schema via `--json` (name / summary / sideEffect / args / examples), for AI agents and scripting
+- Added per-subcommand `--help` (e.g. `model --help`, `config set --help`) — focused help derived from a single command registry
+- **User-customizable model mapping table** — model context window sizes are now managed by the user at runtime. The bundled `data/models.json` is just a frozen base; your local `modelMap` overlays it, so adapting to a new model (e.g. `glm-5.2` with 1M context) needs only `model set <id> <size>` — no package update or upgrade required
+- Added `model` subcommands: `list`, `get`, `set`, `remove` — manage model context window size mappings via CLI
+- Added `modelMap` persistence in tool config — custom model sizes survive across sessions
+- Added `config reset [--models] [--yes]` — restore user config to defaults. `--models` limits the reset to custom model mappings. Preserves install state; prompts unless `--yes` (required in non-interactive sessions)
+- Model ids reported with a bracket suffix (e.g. `glm-5.2[1M]`) are now normalized to the bare id for lookup and display
+- Updated default table: added `glm-5.2` (1M context), removed `glm-5` and `glm-5.1` (no longer in plan)
+- Added `reset-format` config (`time` / `countdown`) — show reset time as countdown duration instead of time point
+- Weekly quota now displays reset time (time point or countdown) alongside the progress bar
+- Context window resolution now relies solely on the local model map: stdin `context_window_size` is ignored, and stdin-provided window percentages are no longer used as a fallback (they are often inaccurate). If the current model id is not in the map, the context segment is simply not shown rather than guessing
+- ctx cache now keyed by session id + model id: token usage can be 0 or missing on individual status-line frames (session start, between requests), which would make the context segment flash. The cache falls back to the last valid value for the same session and model so the display stays stable — a stale value from a different session or model is never shown
+- Refactored command dispatch into `configCommand.js` / `modelCommand.js`, and extracted a shared `bar.js` and a context-scoped `cache.js`. No behavior change
+- Hardcoded `glm-4.7` fallback remains as a defensive guard against `data/models.json` I/O failure
+
 ## 1.2.0
 
 - Added interactive TUI configuration: `glm-quota-line configure` — live preview, per-component toggle and style, global options

package/README.en.md

@@ -116,6 +116,21 @@ Pacing analysis calculates theoretical budget based on elapsed work days:
 - Yellow — usage speed 1.1x–1.3x theoretical
 - Red — usage speed > 1.3x theoretical
 
+### reset-format — Reset time format
+
+Controls how the reset time is displayed in the status line.
+
+| Value | Description | Example |
+|---|---|---|
+| `time` (default) | Show the reset time point | `reset 14:47` |
+| `countdown` | Show the remaining countdown | `reset 52m` / `reset 2d 5h` |
+
+```bash
+glm-quota-line config set reset-format countdown
+```
+
+The weekly quota also shows the reset info (time point or countdown) alongside its progress bar.
+
 ### --json — JSON output
 
 Output structured JSON format (terminal mode only, ignored in status line mode):
@@ -188,6 +203,21 @@ Auth source priority (highest to lowest):
 2. Environment variables `ANTHROPIC_AUTH_TOKEN` / `ANTHROPIC_BASE_URL`
 3. `env` field in `~/.claude/settings.json`
 
+### reset — Restore defaults
+
+One-shot reset of user config. Install state is preserved (status line / hook are not removed).
+
+```bash
+glm-quota-line config reset            # reset all user config (interactive confirm)
+glm-quota-line config reset --models   # reset only custom model mappings (modelMap)
+glm-quota-line config reset --yes      # skip confirmation (scripts / CI)
+```
+
+- Default scope: all global config keys (theme/display/style/work-days/minimalist/raw-values/reset-format/auth-token/base-url), component layout (lines), and custom model mappings (modelMap).
+- `--models` limits the reset to modelMap only; other config is left untouched.
+- In non-interactive sessions (no TTY) `--yes` is required, otherwise it errors out — to avoid accidental deletion.
+- After reset the status line still works — model mappings fall back to the bundled table (`data/models.json`).
+
 ### configure — Interactive TUI setup
 
 Launch a terminal UI with live preview to adjust component toggles, styles, and global options.
@@ -211,6 +241,31 @@ Components:
 | `reset` | Reset time | — | ✓ |
 | `ctx` | Context window usage | ✓ (bar/text) | ✓ |
 
+### model — Context window management
+
+Manage the context window size mapping per model. The context window usage percentage is computed from this mapping.
+
+```bash
+glm-quota-line model list                    # list all models and their context window size
+glm-quota-line model get <model-id>          # show the size of a given model
+glm-quota-line model set <model-id> <size>   # set a model size (e.g. 300K or 300000)
+glm-quota-line model remove <model-id>       # remove a custom mapping (built-in model reverts to default)
+```
+
+Example `list` output:
+
+```
+glm-4.7        200K
+glm-5.1        200K
+glm-5.2        300K *
+```
+
+The `*` marker indicates a user-configured mapping.
+
+The model mapping is a two-layer system: the bundled `data/models.json` is the default base, and the `modelMap` in `~/.claude/glm-quota-line.json` is the user overlay (merged key-by-key at runtime). **When Zhipu releases a new model, you don't need to wait for an npm package update — just run `model set <new-model> <size>` to write the local overlay and it takes effect immediately.**
+
+To clear all custom model mappings and return to the bundled default table, run `glm-quota-line config reset --models --yes`.
+
 ## Recommended Combinations
 
 | Use case | Config |
@@ -219,9 +274,11 @@ Components:
 | Light terminal | `style=bar`, `theme=light` |
 | Minimal setup | `style=compact`, `theme=mono` |
 | Track usage | `style=bar`, `display=used` |
+| Countdown reset | `reset-format=countdown` |
 | Scripting | `--json` |
 | 6-day work week | `work-days=6` |
 | Minimal values | `minimalist=true` |
+| Custom model | `glm-quota-line model set glm-5.2 300K` |
 | Interactive setup | `glm-quota-line configure` |
 
 ## Command Reference
@@ -234,18 +291,45 @@ glm-quota-line version
 glm-quota-line check-update
 glm-quota-line configure
 glm-quota-line config show
-glm-quota-line config set <style|display|theme|auth-token|base-url|work-days|minimalist|raw-values> <value>
+glm-quota-line config set <style|display|theme|auth-token|base-url|work-days|minimalist|raw-values|reset-format> <value>
 glm-quota-line config unset <key>
+glm-quota-line config reset [--models] [--yes]
+glm-quota-line model list
+glm-quota-line model get <model-id>
+glm-quota-line model set <model-id> <size>
+glm-quota-line model remove <model-id>
 ```
 
 Run `glm-quota-line --help` for full descriptions.
 
+## For Agents / Automation
+
+If you are an AI agent (Claude Code / Cursor / Codex) or a script author, the following surface is guaranteed stable:
+
+```bash
+# Get the full machine-readable schema of every command in one call
+# (name / summary / sideEffect / args / examples)
+glm-quota-line commands --json
+
+# Focused help for a single command or command group
+glm-quota-line model --help
+glm-quota-line config set --help
+```
+
+Conventions:
+- **Exit codes**: `0` on success, `1` on error.
+- **Side-effect tagging**: every command is tagged `read` (safe for automation), `write` (persists config / model map), `mutating` (modifies Claude Code integration), or `interactive` (requires a TTY).
+- **Non-blocking**: no command blocks waiting for input (`config reset` needs `--yes` in non-interactive sessions; `configure` prints a hint instead of launching the TUI when there is no TTY).
+- **Read query**: `glm-quota-line --json` emits structured quota data (5h / week / MCP + reset times).
+
+Typical agent workflow: run `commands --json` to discover capabilities → invoke the command you need → parse stdout.
+
 ## Notes
 
 - Shows `TOKENS_LIMIT` and `MCP_LIMIT` / `TIME_LIMIT` quotas
 - Weekly quota progress bar shadow (▒) is a daily pacing baseline: the theoretical budget based on elapsed work days — the more the filled bar (█) extends past the shadow, the faster you're consuming
 - Context window usage is shown by default; toggle via `glm-quota-line configure`
-- Context window usage is calculated from raw tokens when model mapping is available, falls back to API-provided percentages
+- Context window usage is computed from raw tokens using the local model map; if the current model id is not in the map, the context segment is not shown (no guessing)
 - Each component can be individually toggled and styled via `glm-quota-line configure`
 - Missing auth returns `GLM | auth expired`; API failures return `GLM | quota unavailable`
 - `install` does not replace an unmanaged status line unless `--force` is used

package/README.md

@@ -46,7 +46,7 @@ glm-quota-line install
 安装完成。Claude Code 底部状态栏会自动显示配额：
 
 ```
-GLM Lite █████████░ 91% | W ▒▒▒▒░░░░░ 47% | 14:47 | ctx ███░░░ 45% (glm-4.7/200K)
+GLM Lite █████████░ 91% | W ▒▒▒▒░░░░░ 47% 11:10 | 14:47 | ctx ███░░░ 45% (glm-4.7/200K)
 ```
 
 也可以直接在终端运行 `glm-quota-line` 快速查看用量，无需启动 Claude Code。
@@ -66,9 +66,9 @@ glm-quota-line check-update
 
 | 值 | 说明 | 示例 |
 |---|---|---|
-| `bar`（默认） | 进度条可视化 | `GLM Lite █████████░ 91% \| W ▒▒▒▒░░░░░ 47% \| 14:47 \| ctx ███░░░ 45%` |
-| `text` | 完整文本 | `GLM Lite \| 5h 91% \| week 47% \| reset 14:47 \| ctx 45%` |
-| `compact` | 紧凑模式 | `GLM 5h 91% W 47% \| 14:47` |
+| `bar`（默认） | 进度条可视化 | `GLM Lite █████████░ 91% \| W ▒▒▒▒░░░░░ 47% 11:10 \| 14:47 \| ctx ███░░░ 45%` |
+| `text` | 完整文本 | `GLM Lite \| 5h 91% \| week 47% 11:10 \| reset 14:47 \| ctx 45%` |
+| `compact` | 紧凑模式 | `GLM 5h 91% W 47% 11:10 \| 14:47` |
 
 ```bash
 glm-quota-line config set style compact
@@ -125,8 +125,8 @@ glm-quota-line config set minimalist true
 ```
 
 效果对比：
-- 关闭：`GLM Lite █████████░ 91% | W ▒▒▒▒░░░░░ 47% | 14:47`
-- 开启：`█████████░ 91% | ▒▒▒▒░░░░░ 47% | 14:47`
+- 关闭：`GLM Lite █████████░ 91% | W ▒▒▒▒░░░░░ 47% 11:10 | 14:47`
+- 开启：`█████████░ 91% | ▒▒▒▒░░░░░ 47% 11:10 | 14:47`
 
 ### raw-values — 原始数值
 
@@ -136,6 +136,21 @@ glm-quota-line config set minimalist true
 glm-quota-line config set raw-values true
 ```
 
+### reset-format — 重置时间格式
+
+控制状态栏中重置时间的显示格式。
+
+| 值 | 说明 | 示例 |
+|---|---|---|
+| `time`（默认） | 显示重置时间点 | `reset 14:47` |
+| `countdown` | 显示剩余倒计时 | `reset 52m` / `reset 2d 5h` |
+
+```bash
+glm-quota-line config set reset-format countdown
+```
+
+周配额也会同步显示重置信息（时间点或倒计时）。
+
 ### --json — JSON 输出
 
 输出结构化 JSON 格式（仅终端模式，状态栏模式忽略）：
@@ -169,6 +184,21 @@ glm-quota-line --json
 }
 ```
 
+### reset — 恢复出厂配置
+
+一键重置用户配置,保留安装状态(不卸载状态栏/hook)。
+
+```bash
+glm-quota-line config reset            # 交互确认后重置全部用户配置
+glm-quota-line config reset --models   # 只清空自定义模型映射(modelMap)
+glm-quota-line config reset --yes      # 跳过确认(脚本/CI 用)
+```
+
+- 默认重置范围:所有全局配置项(theme/display/style/work-days/minimalist/raw-values/reset-format/auth-token/base-url)、组件布局(lines)、自定义模型映射(modelMap)。
+- `--models` 限定只清 modelMap,其它配置不动。
+- 非交互环境(无 TTY)必须加 `--yes`,否则报错退出,避免误删。
+- 重置后状态栏仍正常工作——模型映射会回退到包内默认表(`data/models.json`)。
+
 ### auth-token / base-url — 自定义鉴权
 
 当 Claude Code 运行在代理或网关后面时，注入的 `ANTHROPIC_AUTH_TOKEN` / `ANTHROPIC_BASE_URL` 可能不是实际值，可手动覆盖：
@@ -198,7 +228,7 @@ glm-quota-line configure
 
 操作方式：
 - `↑↓` 选择组件，`Enter` 进入编辑，`Tab` 切换样式，`Space` 开关显示
-- `g` 进入全局选项（主题、显示模式、极简模式、原始数值）
+- `g` 进入全局选项（主题、显示模式、极简模式、原始数值、重置时间格式）
 - `s` 保存，`q` 退出
 
 组件列表：
@@ -211,6 +241,31 @@ glm-quota-line configure
 | `reset` | 重置时间 | — | ✓ |
 | `ctx` | 上下文窗口用量 | ✓ (bar/text) | ✓ |
 
+### model — 模型上下文窗口管理
+
+管理模型的上下文窗口大小映射。上下文窗口用量百分比基于此映射计算。
+
+```bash
+glm-quota-line model list                    # 查看所有模型及其上下文窗口大小
+glm-quota-line model get <model-id>          # 查看指定模型的大小
+glm-quota-line model set <model-id> <size>   # 设置模型大小（如 300K 或 300000）
+glm-quota-line model remove <model-id>       # 移除自定义映射（内置模型恢复默认值）
+```
+
+`list` 输出示例：
+
+```
+glm-4.7        200K
+glm-5.1        200K
+glm-5.2        300K *
+```
+
+`*` 标记表示用户自定义的映射。
+
+模型映射采用双层结构:包内 `data/models.json` 作为默认底层,`~/.claude/glm-quota-line.json` 里的 `modelMap` 作为用户覆盖层(运行时逐键覆盖)。**智谱发布新模型时,无需等待 npm 包更新——直接 `model set <新模型> <大小>` 写入本地覆盖层即可立刻生效。**
+
+想清空所有自定义模型映射、回到包内默认表:执行 `glm-quota-line config reset --models --yes`。
+
 ## 推荐搭配
 
 | 使用场景 | 配置 |
@@ -219,9 +274,11 @@ glm-quota-line configure
 | 浅色终端 | `style=bar`, `theme=light` |
 | 极简工作流 | `style=compact`, `theme=mono` |
 | 关注已用量 | `style=bar`, `display=used` |
+| 倒计时重置 | `reset-format=countdown` |
 | 脚本调用 | `--json` |
 | 6 天工作制 | `work-days=6` |
 | 极简数值 | `minimalist=true` |
+| 自定义模型 | `glm-quota-line model set glm-5.2 300K` |
 | 交互式配置 | `glm-quota-line configure` |
 
 ## 命令参考
@@ -234,12 +291,38 @@ glm-quota-line version
 glm-quota-line check-update
 glm-quota-line configure
 glm-quota-line config show
-glm-quota-line config set <style|display|theme|auth-token|base-url|work-days|minimalist|raw-values> <value>
+glm-quota-line config set <style|display|theme|auth-token|base-url|work-days|minimalist|raw-values|reset-format> <value>
 glm-quota-line config unset <key>
+glm-quota-line config reset [--models] [--yes]
+glm-quota-line model list
+glm-quota-line model get <model-id>
+glm-quota-line model set <model-id> <size>
+glm-quota-line model remove <model-id>
 ```
 
 运行 `glm-quota-line --help` 查看完整说明。
 
+## 给 Agent / 自动化
+
+如果你是 AI agent(Claude Code / Cursor / Codex)或脚本作者,以下接口保证稳定:
+
+```bash
+# 一次调用拿到所有命令的 machine-readable schema(name / summary / sideEffect / args / examples)
+glm-quota-line commands --json
+
+# 查看单个命令或命令组的聚焦帮助
+glm-quota-line model --help
+glm-quota-line config set --help
+```
+
+约定:
+- **退出码**:成功 `0`,失败 `1`。
+- **副作用标注**:每个命令标为 `read`(只读,自动化安全)、`write`(改配置/模型映射)、`mutating`(改 Claude Code 集成)、`interactive`(需 TTY)。
+- **非阻塞**:所有命令都不会卡住等待输入(仅 `config reset` 在非交互环境需 `--yes`,`configure` 在非 TTY 时打印提示而非启动 TUI)。
+- **只读查询**:`glm-quota-line --json` 输出结构化配额数据(5h/week/MCP + 重置时间)。
+
+典型 agent 工作流:先 `commands --json` 了解能力 → 调用需要的命令 → 读取 stdout 解析结果。
+
 ## 说明
 
 - 展示 `TOKENS_LIMIT` 配额和 `MCP_LIMIT` / `TIME_LIMIT` 配额
@@ -247,6 +330,7 @@ glm-quota-line config unset <key>
 - 状态栏默认显示上下文窗口用量和模型信息，可通过 `glm-quota-line configure` 交互式关闭
 - 每个组件可单独控制开关和样式，通过 `glm-quota-line configure` 交互式调整
 - 上下文窗口使用率优先从原始 token 数计算，当模型映射不可用时回退到 API 提供的百分比
+- 模型上下文窗口大小可通过 `glm-quota-line model` 命令自定义，支持查看、设置、移除和批量导入
 - 鉴权缺失返回 `GLM | auth expired`；接口异常返回 `GLM | quota unavailable`
 - 非 GLM 提供商（非智谱 AI 端点）会自动禁用配额查询
 - `install` 默认不会覆盖非本工具管理的状态栏，除非使用 `--force`

package/data/models.json

@@ -0,0 +1,10 @@
+{
+  "schemaVersion": 1,
+  "source": "https://open.bigmodel.cn/dev/api#glm",
+  "models": {
+    "glm-4.5-air": 128000,
+    "glm-4.7": 200000,
+    "glm-5-turbo": 200000,
+    "glm-5.2": 1000000
+  }
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "glm-quota-line",
-  "version": "1.2.0",
+  "version": "1.3.0",
   "description": "Zhipu GLM Coding Plan quota monitor for Claude Code status line.",
   "type": "module",
   "license": "MIT",
@@ -29,6 +29,7 @@
   },
   "files": [
     "src",
+    "data",
     "README.md",
     "README.en.md",
     "LICENSE",

package/src/claude/sessionStart.js

@@ -1,5 +1,4 @@
 import { loadConfig } from "../shared/config.js";
-import { clearCtxCache } from "../core/quota/cache.js";
 import { resolveQuotaStatus } from "../core/quota/service.js";
 import { readStatusLineInput } from "./input.js";
 import { readToolConfig } from "./settings.js";
@@ -16,10 +15,6 @@ export async function refreshQuotaOnSessionStart(options = {}) {
     sessionId: hookInput?.session_id || ""
   };
 
-  if (hookInput?.source === "compact" || hookInput?.source === "clear" || hookInput?.source === "startup") {
-    await clearCtxCache(config.cacheFilePath).catch(() => {});
-  }
-
   return resolveQuotaStatusFn(config, {
     now: options.now,
     fetchImpl: options.fetchImpl,

package/src/claude/settings.js

@@ -5,6 +5,7 @@ import {
   TOOL_CONFIG_MANAGED_BY,
   TOOL_CONFIG_SCHEMA_VERSION,
   isValidDisplayMode,
+  isValidResetFormat,
   isValidStatusStyle,
   isValidTheme,
   isValidWorkDays
@@ -69,6 +70,10 @@ function normalizeToolConfig(config) {
     normalized.rawValues = base.rawValues;
   }
 
+  if (isValidResetFormat(base.resetFormat)) {
+    normalized.resetFormat = base.resetFormat;
+  }
+
   const authToken = normalizeOptionalString(base.authToken);
   if (authToken) {
     normalized.authToken = authToken;
@@ -84,6 +89,19 @@ function normalizeToolConfig(config) {
     normalized.lines = base.lines;
   }
 
+  // Preserve modelMap (custom model context window sizes)
+  if (base.modelMap && typeof base.modelMap === "object") {
+    const validMap = {};
+    for (const [modelId, size] of Object.entries(base.modelMap)) {
+      if (typeof modelId === "string" && modelId && typeof size === "number" && size > 0 && Number.isFinite(size)) {
+        validMap[modelId] = size;
+      }
+    }
+    if (Object.keys(validMap).length > 0) {
+      normalized.modelMap = validMap;
+    }
+  }
+
   return normalized;
 }
 
@@ -132,3 +150,27 @@ export async function unsetToolConfigValue(key, configPath = getToolConfigPath()
   await writeToolConfig(current, configPath);
   return current;
 }
+
+// Reset user config to defaults. Preserves install metadata (does not uninstall
+// the status line), schemaVersion, and managedBy. With modelsOnly, only the
+// user modelMap overlay is cleared (bundled seed still applies at runtime).
+export async function resetToolConfig({ modelsOnly = false } = {}, configPath = getToolConfigPath()) {
+  const current = await readToolConfig(configPath);
+  const install =
+    current.install && typeof current.install === "object" ? current.install : {};
+
+  let reset;
+  if (modelsOnly) {
+    reset = { ...current };
+    delete reset.modelMap;
+  } else {
+    reset = {
+      schemaVersion: TOOL_CONFIG_SCHEMA_VERSION,
+      managedBy: TOOL_CONFIG_MANAGED_BY,
+      install
+    };
+  }
+
+  await writeToolConfig(reset, configPath);
+  return reset;
+}

package/src/cli/args.js

@@ -15,6 +15,8 @@ const FLAGS = [
   { names: ["--theme"], dest: "theme" },
   { names: ["--force"], dest: "force", boolean: true },
   { names: ["--json"], dest: "json", boolean: true },
+  { names: ["--yes"], dest: "yes", boolean: true },
+  { names: ["--models"], dest: "models", boolean: true },
   { names: ["--help", "-h"], dest: "help", boolean: true },
   { names: ["--version", "-v"], dest: "version", boolean: true }
 ];

package/src/cli/commands.js

@@ -1,5 +1,6 @@
 import {
   isValidDisplayMode,
+  isValidResetFormat,
   isValidStatusStyle,
   isValidTheme,
   isValidWorkDays
@@ -15,6 +16,10 @@ import { installClaudeStatusLine, uninstallClaudeStatusLine } from "../claude/in
 import { refreshQuotaOnSessionStart } from "../claude/sessionStart.js";
 import { checkForUpdates } from "./update.js";
 import { getPackageVersion } from "../shared/packageInfo.js";
+import { handleModelCommand } from "./modelCommand.js";
+import { handleConfigReset } from "./configCommand.js";
+import { renderCommandsTable } from "./help.js";
+import { COMMAND_REGISTRY, GLOBAL_FLAGS, ENVIRONMENT, allCommands } from "./registry.js";
 
 function isNonEmptyString(value) {
   return typeof value === "string" && value.trim().length > 0;
@@ -78,9 +83,17 @@ const CONFIG_KEYS = {
     validate: (v) => v === "true" || v === "false",
     invalidMessage: "Invalid raw-values. Use: true or false.",
     transform: (v) => v === "true"
+  },
+  "reset-format": {
+    property: "resetFormat",
+    validate: isValidResetFormat,
+    invalidMessage: "Invalid reset-format. Use: time or countdown."
   }
 };
 
+const SUPPORTED_CONFIG_KEYS_MESSAGE =
+  "Supported config keys: style, display, theme, auth-token, base-url, work-days, minimalist, raw-values\n";
+
 export async function handleCommand(args, output = process.stdout, dependencies = {}) {
   const getVersion = dependencies.getVersion || getPackageVersion;
   const runUpdateCheck = dependencies.runUpdateCheck || checkForUpdates;
@@ -91,6 +104,21 @@ export async function handleCommand(args, output = process.stdout, dependencies
     return true;
   }
 
+  if (command === "commands") {
+    if (args.json) {
+      const payload = {
+        version: await getVersion(),
+        commands: allCommands(),
+        globalFlags: GLOBAL_FLAGS,
+        environment: ENVIRONMENT
+      };
+      output.write(`${JSON.stringify(payload, null, 2)}\n`);
+    } else {
+      renderCommandsTable(output);
+    }
+    return true;
+  }
+
   if (command === "check-update") {
     const result = await runUpdateCheck();
     output.write(`glm-quota-line ${result.currentVersion}\n`);
@@ -163,7 +191,7 @@ export async function handleCommand(args, output = process.stdout, dependencies
     const configKey = CONFIG_KEYS[key];
     if (!configKey) {
       process.exitCode = 1;
-      output.write("Supported config keys: style, display, theme, auth-token, base-url, work-days, minimalist, raw-values\n");
+      output.write(SUPPORTED_CONFIG_KEYS_MESSAGE);
       return true;
     }
 
@@ -187,7 +215,7 @@ export async function handleCommand(args, output = process.stdout, dependencies
     const configKey = CONFIG_KEYS[key];
     if (!configKey) {
       process.exitCode = 1;
-      output.write("Supported config keys: style, display, theme, auth-token, base-url, work-days, minimalist, raw-values\n");
+      output.write(SUPPORTED_CONFIG_KEYS_MESSAGE);
       return true;
     }
 
@@ -196,12 +224,20 @@ export async function handleCommand(args, output = process.stdout, dependencies
     return true;
   }
 
+  if (command === "config" && subcommand === "reset") {
+    return handleConfigReset(args, output, dependencies.configPath);
+  }
+
   if (command === "config") {
     process.exitCode = 1;
-    output.write("Supported config subcommands: show, set, unset\n");
+    output.write("Supported config subcommands: show, set, unset, reset\n");
     return true;
   }
 
+  if (command === "model") {
+    return handleModelCommand(subcommand, key, value, output, dependencies.configPath);
+  }
+
   if (command === "configure") {
     return false;
   }

package/src/cli/configCommand.js

@@ -0,0 +1,86 @@
+import readline from "node:readline/promises";
+
+import {
+  getToolConfigPath,
+  readToolConfig,
+  resetToolConfig
+} from "../claude/settings.js";
+
+// Keys considered "user config" for the reset summary (install/schema/managedBy are preserved).
+const RESET_USER_KEYS = [
+  "style",
+  "displayMode",
+  "theme",
+  "workDays",
+  "minimalist",
+  "rawValues",
+  "resetFormat",
+  "authToken",
+  "baseUrl"
+];
+
+export async function handleConfigReset(args, output, configPath) {
+  const modelsOnly = Boolean(args.models);
+  const skipConfirm = Boolean(args.yes);
+  const targetPath = configPath || getToolConfigPath();
+  const config = await readToolConfig(targetPath);
+
+  if (modelsOnly) {
+    const count = config.modelMap ? Object.keys(config.modelMap).length : 0;
+    if (count === 0) {
+      output.write("No custom model mappings to reset.\n");
+      return true;
+    }
+    if (!skipConfirm && !(await confirmReset(`Reset ${count} custom model mapping(s)? [y/N] `, output))) {
+      output.write("Aborted.\n");
+      return true;
+    }
+    await resetToolConfig({ modelsOnly: true }, targetPath);
+    output.write(`Reset ${count} custom model mapping(s).\nconfig: ${targetPath}\n`);
+    return true;
+  }
+
+  const setKeys = RESET_USER_KEYS.filter((k) => config[k] !== undefined);
+  const modelCount = config.modelMap ? Object.keys(config.modelMap).length : 0;
+  const hasLines = Array.isArray(config.lines) && config.lines.length > 0;
+
+  if (setKeys.length === 0 && modelCount === 0 && !hasLines) {
+    output.write("Nothing to reset — config is already at defaults.\n");
+    return true;
+  }
+
+  const parts = [];
+  if (setKeys.length > 0) {
+    parts.push(`${setKeys.length} config key(s) (${setKeys.join(", ")})`);
+  }
+  if (modelCount > 0) {
+    parts.push(`${modelCount} custom model mapping(s)`);
+  }
+  if (hasLines) {
+    parts.push("component layout (lines)");
+  }
+
+  if (!skipConfirm && !(await confirmReset(`This will remove: ${parts.join(", ")}. Continue? [y/N] `, output))) {
+    output.write("Aborted.\n");
+    return true;
+  }
+
+  await resetToolConfig({ modelsOnly: false }, targetPath);
+  output.write(`Reset ${parts.join(", ")} to defaults.\nconfig: ${targetPath}\n`);
+  return true;
+}
+
+async function confirmReset(prompt, output) {
+  if (!process.stdin.isTTY) {
+    process.exitCode = 1;
+    output.write("config reset is destructive; pass --yes to skip confirmation in non-interactive sessions.\n");
+    return false;
+  }
+  const rl = readline.createInterface({ input: process.stdin, output: process.stdout });
+  try {
+    const answer = (await rl.question(prompt)).trim().toLowerCase();
+    return answer === "y" || answer === "yes";
+  } finally {
+    rl.close();
+  }
+}