@cryer/star-cli 0.3.12 → 0.3.14

package/README.md

@@ -1,6 +1,6 @@
 <div align="center">
 
-<img src="assets/icon.png" alt="Star CLI" width="320" />
+<img src="assets/star-cli-logo.svg" alt="Star CLI" width="480" />
 
 **An AI agent command-line interface written in TypeScript**
 
@@ -17,7 +17,7 @@ Multi-model LLM access · streaming terminal UI · tool calling · permission co
 
 </div>
 
-Features: streaming REPL with slash commands (+ autocomplete) · OpenAI / Anthropic / OpenAI-compatible providers with an interactive `/connect` onboarding wizard · built-in fs / bash / web tools with a permission gate · git integration (`/commit` drafts Conventional Commits messages, `/diff` shows a colored working-tree diff, repo status sent as a per-turn request-scoped reminder so it never breaks the prompt cache) · prompt caching on Anthropic providers (cache breakpoints on the stable prefix and the history tail; OpenAI-style protocols cache server-side automatically) with a status-bar hit rate · token thrift: re-reading an unchanged file answers with a one-line note, and auxiliary calls (summaries, titles) can run on a cheaper `smallModel` · plan mode with read-only research and plan approval · thinking spinner with dim reasoning preview · diff preview on write/edit approval · `@file` mentions · `!cmd` shell passthrough · custom slash commands from Markdown files · conversation compaction (`/compact`) · session persistence and resume with auto-generated titles (`/resume`, `star -r`) · subagent delegation for focused subtasks · lifecycle hooks (`PreToolUse`/`PostToolUse`/`Stop` shell commands from config) · file-write snapshots with `/undo` and checkpoint rollback with `/rewind` · TODO task tracking · background shell tasks with status-bar visibility (`/tasks`) · terminal bell on long turns and background-task completion · Markdown session export (`/export`) · `/init` + `/doctor` project scaffolding and environment checks · cost estimation · update notifier · `--json` NDJSON output for scripting.
+Features: streaming REPL with slash commands (+ autocomplete) · OpenAI / Anthropic / OpenAI-compatible providers with an interactive `/connect` onboarding wizard · built-in fs / bash / web tools with a permission gate · git integration (`/commit` drafts Conventional Commits messages, `/diff` shows a colored working-tree diff, repo status sent as a per-turn request-scoped reminder so it never breaks the prompt cache) · prompt caching on Anthropic providers (cache breakpoints on the stable prefix and the history tail; OpenAI-style protocols cache server-side automatically) with a status-bar hit rate · token thrift: re-reading an unchanged file answers with a one-line note, and auxiliary calls (summaries, titles) can run on a cheaper `smallModel` · plan mode with read-only research and plan approval · thinking spinner with dim reasoning preview · diff preview on write/edit approval · `@file` mentions · `!cmd` shell passthrough · custom slash commands from Markdown files · stale-context elision (old tool outputs, tool-call arguments and attached images are replaced in place with placeholders — tunable via `elision*`) · conversation compaction (`/compact`) · session persistence and resume with auto-generated titles (`/resume`, `star -r`) · subagent delegation for focused subtasks · lifecycle hooks (`PreToolUse`/`PostToolUse`/`Stop` shell commands from config) · file-write snapshots with `/undo` and checkpoint rollback with `/rewind` · TODO task tracking · background shell tasks with status-bar visibility (`/tasks`) · terminal bell on long turns and background-task completion · Markdown session export (`/export`) · `/init` + `/doctor` project scaffolding and environment checks · cost estimation · update notifier · `--json` NDJSON output for scripting.
 
 ## Requirements
 
@@ -58,6 +58,17 @@ contextMaxTokens = 100000         # context window in tokens (per-model override
 # compactThresholdTokens = 80000  # optional auto-compaction trigger in tokens; defaults to the
                                   # context window and is clamped to it (top-level only — there is
                                   # no per-model override). /compact always compacts, regardless.
+# Stale-content elision: once estimated context usage reaches
+# elisionThresholdFraction of the compaction threshold, bulk older than
+# elisionKeepRecentMessages messages (stale tool results, old tool-call
+# arguments, old attached images) is replaced in place with a placeholder —
+# message count and order are unchanged, so /undo markers stay valid and
+# compaction often never fires. Smaller-than-elisionMinTokens results are
+# kept (cheaper than re-fetching). Set the fraction to 1 to elide only at
+# the compaction point (effectively off), 0 to elide even in short sessions.
+elisionThresholdFraction = 0.5
+elisionKeepRecentMessages = 24
+elisionMinTokens = 250
 # Max model round-trips ("steps") per turn between progress checkpoints — one
 # step is a reply plus the tool calls it asked for; retries and auto-continue
 # nudges don't count. Reaching the cap while tools still execute resets it
@@ -186,6 +197,7 @@ contextMaxTokens = 128000
 - `STAR_HOME` — overrides `~/.star-cli` as the data home (config, sessions, `.env`, memory, debug log).
 - `STAR_DEBUG` — any non-empty value enables stream diagnostics appended to `~/.star-cli/debug.log`, one JSON object per line.
 - `STAR_NO_NOTIFY=1` — disables the terminal bell without touching the config.
+- `STAR_NO_MOTION=1` — disables spinner and terminal-title animation for reduced-motion or low-frequency terminals.
 
 
 ## CLI flags
@@ -262,7 +274,7 @@ The REPL rings the terminal bell (BEL) so you can switch windows while the agent
 
 ## Session budget
 
-Setting `sessionBudgetUsd` in the config caps spend per REPL session (the active model needs `promptPrice`/`completionPrice` so the cost can be computed). When a finished turn pushes the session cost past 80% of the cap, a one-time warning appears; past 100%, further prompts are blocked with an error while slash commands keep working — raise the limit in the config or start fresh with `/new`.
+Setting `sessionBudgetUsd` in the config caps spend per REPL session. Usage and cost are recorded per model when each main, auxiliary or subagent request finishes, so `/model` never reprices earlier history; each model needs `promptPrice`/`completionPrice` for an exact amount, and unpriced requests are reported separately. When recorded cost passes 80% of the cap, a one-time warning appears; past 100%, further prompts are blocked while slash commands keep working — raise the limit or start fresh with `/new`.
 
 ## Plan mode
 
@@ -362,7 +374,7 @@ The `subagent` tool (`exec` level, so it is hidden in plan mode and denied in re
 
 Permission modes: `ask` (reads allowed, writes/exec ask) · `auto` (everything allowed except the hard-denied rules above) · `readonly` (read-only) · `yolo` (allow everything, never ask — **all safety checks disabled**, use at your own risk) · `plan` (read-only research with a plan-approval flow, see above — session-only). Switch at runtime with `/permission` (persisted to the config file), per session with `/plan` or `Shift+Tab`, or at startup with `--permission-mode`. On top of the modes, the `[permissions]` config block holds persistent `allow` and `deny` rule lists (same `bash(npm test *)` syntax); deny rules are evaluated right after the hard safety rules and beat both `auto` mode and allow-rules, while `yolo` bypasses everything.
 
-Every successful `write_file` / `edit_file` first snapshots the file's previous content (in-memory, last 50 writes per session); `/undo` reverts the last turn's snapshots — restoring old content, or deleting files the turn created — but only after showing a preview of the retraction and per-file revert diffs and getting a `y`/`n` confirmation.
+Every successful `write_file` / `edit_file` first snapshots the file's previous content (in-memory, last 50 writes per session). A write is refused when existing content is too large or unreadable and therefore cannot be restored safely. `/undo` reverts the last turn's snapshots — restoring old content, or deleting files the turn created — but only after showing a preview and getting confirmation. Whole-tree Git snapshots also honor custom `permissions.sensitive` rules and disable Git attribute/filter/encoding conversion so restores preserve bytes.
 
 ## Checkpoints and /rewind
 
@@ -372,7 +384,7 @@ Checkpoints are persisted under `~/.star-cli/sessions/<id>/checkpoints/` (an `in
 
 ## Sessions
 
-Sessions persist under `~/.star-cli/sessions/<id>/` (messages as JSONL + `meta.json`). `/resume` with no argument opens an interactive picker of this directory's sessions (most recently active first, message counts, relative times, a preview of the first message, and a `current` marker on the live session; `/resume --all` covers every directory with each session's cwd shown), while `/resume <id>` and `star -r <id>` jump straight to one — both accept the short id shown in the list. `/fork` branches the live session: every message is copied into a new session titled `Fork of …` which becomes the active one, so you can explore a different direction while the original stays intact (checkpoints and rewind history are not carried over). `/search <query>` full-text-searches every stored session — message text plus tool-call arguments and results — and lists matching sessions with a snippet, ready for `/resume <id>`. `star -c` (`--continue`) jumps straight back into the most recently active session for the current directory, in the REPL and in print mode alike; when the directory has no sessions it says so and starts a fresh one. `-r` and `-c` are mutually exclusive, and a bare `star -r` prints the session list instead of erroring. Sessions are created lazily — opening the REPL and exiting without chatting leaves nothing on disk, and print mode (`-p`) doesn't create a session unless resuming with `-r` or `-c`. Token usage is accumulated in `meta.json`, so `/cost` reflects resumed history too. `/usage` is the global counterpart of `/cost` (which stays session-scoped): it aggregates every session's `meta.json` across all directories into a dashboard — grand totals, a per-day bar chart of the last 14 days with usage, and a per-model breakdown with a $ estimate for models that have pricing configured (models without pricing are listed but excluded from the total). Per-day numbers come from a `usageByDay` bucket map recorded from now on; usage recorded before that field existed only has grand totals and shows up as a single "earlier usage" line. `/new` abandons the in-memory conversation and attaches a fresh lazy session store (the old session stays on disk); deletion is a separate explicit action — `/clear-sessions` removes the sessions recorded for the current directory, `/clear-sessions --all` wipes every session directory (including ones with an unreadable `meta.json`), and both keep the live session; `star --clear-sessions [all]` does the same non-interactively and exits.
+Sessions persist under `~/.star-cli/sessions/<id>/` (messages as JSONL + `meta.json`). `/resume` with no argument opens an interactive picker of this directory's sessions (most recently active first, message counts, relative times, a preview of the first message, and a `current` marker on the live session; `/resume --all` covers every directory with each session's cwd shown), while `/resume <id>` and `star -r <id>` jump straight to one — both accept the short id shown in the list. `/fork` branches the live session: every message is copied into a new session titled `Fork of …` which becomes the active one, so you can explore a different direction while the original stays intact (checkpoints and rewind history are not carried over). `/search <query>` full-text-searches every stored session — message text plus tool-call arguments and results — and lists matching sessions with a snippet, ready for `/resume <id>`. `star -c` (`--continue`) jumps straight back into the most recently active session for the current directory, in the REPL and in print mode alike; when the directory has no sessions it says so and starts a fresh one. `-r` and `-c` are mutually exclusive, and a bare `star -r` prints the session list instead of erroring. Session content is still created lazily — opening the REPL and exiting without chatting leaves no `meta.json` or `messages.jsonl`, and print mode (`-p`) doesn't create a session unless resuming with `-r` or `-c`. A writable session holds an exclusive lifecycle lock until close; another process cannot resume or delete the active session, and stale locks from crashed processes are reclaimed automatically. Token usage and actual cost are accumulated per model in `meta.json`, so `/cost` reflects resumed and mixed-model history accurately. `/usage` is the global counterpart of `/cost` (which stays session-scoped): it aggregates every session's `meta.json` across all directories into a dashboard — grand totals, a per-day bar chart of the last 14 days with usage, and a per-model breakdown with a $ estimate for models that have pricing configured (models without pricing are listed but excluded from the total). Per-day numbers come from a `usageByDay` bucket map recorded from now on; usage recorded before that field existed only has grand totals and shows up as a single "earlier usage" line. `/new` abandons the in-memory conversation and attaches a fresh lazy session store (the old session stays on disk); deletion is a separate explicit action — `/clear-sessions` removes the sessions recorded for the current directory, `/clear-sessions --all` wipes every session directory (including ones with an unreadable `meta.json`), and both keep the live session; `star --clear-sessions [all]` does the same non-interactively and exits.
 
 After the first turn of a session (the first user message gets its first assistant reply), a lightweight background LLM request — same model, capped at 20 output tokens — summarizes the first user message into a short title (≤50 chars, matching the message's language) and stores it in `meta.json`, where the `/resume` list picks it up. Title generation is fire-and-forget: it never blocks the REPL or print-mode output, it runs only when the session has no title yet (resumed sessions keep their existing title; untitled older sessions get one after their next turn), and if the request fails the title simply stays empty — untitled sessions show as `(无标题)` in the list. An unfinished request is abandoned when the process exits. It applies wherever a session is attached, including print mode when resuming with `-r`/`-c`.
 

package/README.zh-CN.md

@@ -1,6 +1,6 @@
 <div align="center">
 
-<img src="assets/icon.png" alt="Star CLI" width="320" />
+<img src="assets/star-cli-logo.svg" alt="Star CLI" width="480" />
 
 **用 TypeScript 编写的 AI 代理命令行界面**
 
@@ -17,7 +17,7 @@
 
 </div>
 
-功能特性：带斜杠命令（+ 自动补全）的流式 REPL · OpenAI / Anthropic / OpenAI 兼容服务商，配交互式 `/connect` 接入向导 · 内置 fs / bash / web 工具，带权限门禁 · git 集成（`/commit` 起草 Conventional Commits 提交信息，`/diff` 显示彩色工作区 diff，仓库状态以每回合请求级 reminder 发送，不会破坏提示词缓存）· Anthropic 服务商的提示词缓存（在稳定前缀与历史尾部打缓存断点；OpenAI 系协议由服务端自动缓存），状态栏显示命中率 · token 节俭：重复读取未变化的文件只回一行说明，压缩摘要、标题等辅助调用可配置更便宜的 `smallModel` · 计划模式：只读调研 + 计划批准 · 带暗色推理预览的思考动画 · 写入/编辑批准时的 diff 预览 · `@file` 引用 · `!cmd` shell 直通 · 用 Markdown 文件自定义斜杠命令 · 对话压缩（`/compact`）· 会话持久化与恢复，自动生成标题（`/resume`、`star -r`）· 子代理委派，处理专注的子任务 · 生命周期钩子（来自配置的 `PreToolUse`/`PostToolUse`/`Stop` shell 命令）· 文件写入快照 `/undo` 与检查点回滚 `/rewind` · TODO 任务跟踪 · 状态栏可见的后台 shell 任务（`/tasks`）· 长回合与后台任务完成时的终端响铃 · Markdown 会话导出（`/export`）· `/init` + `/doctor` 项目脚手架与环境检查 · 成本估算 · 更新提醒 · 便于脚本化的 `--json` NDJSON 输出。
+功能特性：带斜杠命令（+ 自动补全）的流式 REPL · OpenAI / Anthropic / OpenAI 兼容服务商，配交互式 `/connect` 接入向导 · 内置 fs / bash / web 工具，带权限门禁 · git 集成（`/commit` 起草 Conventional Commits 提交信息，`/diff` 显示彩色工作区 diff，仓库状态以每回合请求级 reminder 发送，不会破坏提示词缓存）· Anthropic 服务商的提示词缓存（在稳定前缀与历史尾部打缓存断点；OpenAI 系协议由服务端自动缓存），状态栏显示命中率 · token 节俭：重复读取未变化的文件只回一行说明，压缩摘要、标题等辅助调用可配置更便宜的 `smallModel` · 计划模式：只读调研 + 计划批准 · 带暗色推理预览的思考动画 · 写入/编辑批准时的 diff 预览 · `@file` 引用 · `!cmd` shell 直通 · 用 Markdown 文件自定义斜杠命令 · 陈旧上下文省略（旧的工具输出、工具调用参数和附加图片会被原地替换为占位符——可用 `elision*` 键调节）· 对话压缩（`/compact`）· 会话持久化与恢复，自动生成标题（`/resume`、`star -r`）· 子代理委派，处理专注的子任务 · 生命周期钩子（来自配置的 `PreToolUse`/`PostToolUse`/`Stop` shell 命令）· 文件写入快照 `/undo` 与检查点回滚 `/rewind` · TODO 任务跟踪 · 状态栏可见的后台 shell 任务（`/tasks`）· 长回合与后台任务完成时的终端响铃 · Markdown 会话导出（`/export`）· `/init` + `/doctor` 项目脚手架与环境检查 · 成本估算 · 更新提醒 · 便于脚本化的 `--json` NDJSON 输出。
 
 ## 环境要求
 
@@ -58,6 +58,15 @@ contextMaxTokens = 100000         # 上下文窗口（token 数，下方可按
 # compactThresholdTokens = 80000  # 可选的自动压缩触发阈值（token 数）；默认为上下文窗口本身，
                                   # 且会被钳制到窗口以内（仅顶层键——没有按模型覆盖）。
                                   # /compact 无视此阈值，总是执行压缩。
+# 陈旧内容省略（elision）：当估算的上下文用量达到压缩阈值的
+# elisionThresholdFraction 倍时，早于最近 elisionKeepRecentMessages 条消息的大块内容
+#（陈旧的工具结果、旧的工具调用参数、早先附加的图片）会被原地替换为占位符——
+# 消息数量和顺序不变，因此 /undo 标记仍然有效，且往往因此不再需要整轮压缩。
+# 小于 elisionMinTokens 的结果会保留（重新获取反而更贵）。
+# fraction 设为 1 表示只在压缩点才省略（基本等于关闭），设为 0 表示短会话也省略。
+elisionThresholdFraction = 0.5
+elisionKeepRecentMessages = 24
+elisionMinTokens = 250
 # 每回合在进度检查点之间允许的最大模型往返（"step"）数——一个 step 是
 # 一次回复加上它请求的所有工具调用；重试和 auto-continue nudge 不占 step。
 # 到达上限时若工具仍在执行，会发提示并重置额度而不是终止回合；0 表示完全不限。
@@ -181,6 +190,7 @@ contextMaxTokens = 128000
 - `STAR_HOME` —— 把数据目录从 `~/.star-cli` 改为其他位置（配置、会话、`.env`、记忆、调试日志）。
 - `STAR_DEBUG` —— 任意非空值即启用流诊断日志，追加写入 `~/.star-cli/debug.log`，每行一个 JSON 对象。
 - `STAR_NO_NOTIFY=1` —— 不改配置即可禁用终端响铃。
+- `STAR_NO_MOTION=1` —— 禁用 spinner 和终端标题动画，适合减少动态效果或低频远程终端。
 
 
 ## CLI 参数
@@ -257,7 +267,7 @@ REPL 会响终端铃（BEL），这样代理工作时你可以切到别的窗口
 
 ## 会话预算
 
-在配置中设置 `sessionBudgetUsd` 可为每个 REPL 会话设置花费上限（活动模型需要配置 `promptPrice`/`completionPrice` 才能计算成本）。已完成的回合使会话成本超过上限的 80% 时，会出现一次性警告；超过 100% 时，后续提示会被阻止并报错，斜杠命令仍可用——在配置中提高限额，或用 `/new` 重新开始。
+在配置中设置 `sessionBudgetUsd` 可为每个 REPL 会话设置花费上限。主请求、辅助请求和子代理完成时都会按实际所用模型分别记录 token 与成本，因此 `/model` 不会按新价格重算历史；每个模型需要配置 `promptPrice`/`completionPrice` 才能精确计价，未定价请求会单独标明。已记录成本超过上限的 80% 时会出现一次性警告；超过 100% 时后续提示会被阻止，斜杠命令仍可用——在配置中提高限额，或用 `/new` 重新开始。
 
 ## 计划模式
 
@@ -357,7 +367,7 @@ star -p "what's wrong in this UI?" --image screenshot.png --image mockup.jpg
 
 权限模式：`ask`（读放行，写/执行询问）· `auto`（除上述硬性拒绝规则外全部放行）· `readonly`（只读）· `yolo`（放行一切，从不询问——**所有安全检查关闭**，风险自负）· `plan`（只读调研加计划审批流程，见上文——仅会话内）。运行时用 `/permission` 切换（持久化到配置文件），会话内用 `/plan` 或 `Shift+Tab`，启动时用 `--permission-mode`。在模式之上，`[permissions]` 配置块持有持久的 `allow` 和 `deny` 规则列表（同样的 `bash(npm test *)` 语法）；deny 规则在硬性安全规则之后评估，胜过 `auto` 模式和 allow 规则，而 `yolo` 绕过一切。
 
-每次成功的 `write_file` / `edit_file` 会先快照文件的旧内容（内存中，每会话保留最近 50 次写入）；`/undo` 还原上一回合的快照——恢复旧内容，或删除该回合创建的文件——但只在展示撤回预览与逐文件还原 diff 并获得 `y`/`n` 确认之后。
+每次成功的 `write_file` / `edit_file` 会先快照文件的旧内容（内存中，每会话保留最近 50 次写入）。如果旧内容过大或不可读、无法安全恢复，写入会被拒绝。`/undo` 还原上一回合的快照——恢复旧内容，或删除该回合创建的文件——但只在展示预览并获得确认之后。整树 Git 快照同样遵循自定义 `permissions.sensitive` 规则，并禁用 Git 属性、过滤器和编码转换，以确保按原始字节恢复。
 
 ## 检查点与 /rewind
 
@@ -367,7 +377,7 @@ star -p "what's wrong in this UI?" --image screenshot.png --image mockup.jpg
 
 ## 会话
 
-会话持久化在 `~/.star-cli/sessions/<id>/` 下（消息为 JSONL + `meta.json`）。无参数的 `/resume` 打开当前目录会话的交互式选择器（最近活跃优先、消息数、相对时间、首条消息预览、当前会话带 `current` 标记；`/resume --all` 涵盖所有目录并显示各会话的 cwd），`/resume <id>` 和 `star -r <id>` 直接跳转到指定会话——两者都接受列表中显示的短 id。`/fork` 分叉当前会话：所有消息复制到一个标题为 `Fork of …` 的新会话并使之成为活动会话，因此你可以探索不同方向而原会话保持完好（检查点与回滚历史不随带）。`/search <query>` 对所有已存会话做全文搜索——消息文本加工具调用参数与结果——并列出带摘要片段的匹配会话，可直接 `/resume <id>`。`star -c`（`--continue`）直接回到当前目录最近活跃的会话，REPL 与 print 模式一致；目录没有会话时它会说明并开始新会话。`-r` 与 `-c` 互斥，裸 `star -r` 打印会话列表而不报错。会话是懒创建的——打开 REPL 不聊天就退出不会在磁盘留下任何东西，print 模式（`-p`）除非用 `-r` 或 `-c` 恢复，否则不创建会话。token 用量累积在 `meta.json` 中，因此 `/cost` 也反映恢复的历史。`/usage` 是 `/cost` 的全局对应物（`/cost` 保持会话作用域）：它聚合所有目录下每个会话的 `meta.json` 成一张面板——总计、最近 14 天有用量日期的按日柱状图，以及按模型细分并为配置了定价的模型给出美元估算（未配置定价的模型会列出但不计入总额）。按日数字来自此后记录的 `usageByDay` 桶映射；该字段出现之前记录的用量只有总计，显示为单独一行“earlier usage”（更早用量）。`/new` 丢弃内存中的对话并挂载一个全新的懒会话存储（旧会话保留在磁盘上）；删除是单独的显式操作——`/clear-sessions` 删除当前目录记录的会话，`/clear-sessions --all` 清空所有会话目录（包括 `meta.json` 不可读的），两者都保留当前活动会话；`star --clear-sessions [all]` 以非交互方式做同样的事并退出。
+会话持久化在 `~/.star-cli/sessions/<id>/` 下（消息为 JSONL + `meta.json`）。无参数的 `/resume` 打开当前目录会话的交互式选择器（最近活跃优先、消息数、相对时间、首条消息预览、当前会话带 `current` 标记；`/resume --all` 涵盖所有目录并显示各会话的 cwd），`/resume <id>` 和 `star -r <id>` 直接跳转到指定会话——两者都接受列表中显示的短 id。`/fork` 分叉当前会话：所有消息复制到一个标题为 `Fork of …` 的新会话并使之成为活动会话，因此你可以探索不同方向而原会话保持完好（检查点与回滚历史不随带）。`/search <query>` 对所有已存会话做全文搜索——消息文本加工具调用参数与结果——并列出带摘要片段的匹配会话，可直接 `/resume <id>`。`star -c`（`--continue`）直接回到当前目录最近活跃的会话，REPL 与 print 模式一致；目录没有会话时它会说明并开始新会话。`-r` 与 `-c` 互斥，裸 `star -r` 打印会话列表而不报错。会话内容仍是懒创建的——打开 REPL 不聊天就退出不会留下 `meta.json` 或 `messages.jsonl`，print 模式（`-p`）除非用 `-r` 或 `-c` 恢复，否则不创建会话。可写会话在关闭前持有独占生命周期锁，其他进程不能恢复或删除活动会话；崩溃遗留的锁会自动回收。token 用量和实际成本按模型累积到 `meta.json`，因此 `/cost` 能准确覆盖恢复历史和混合模型会话。`/usage` 是 `/cost` 的全局对应物（`/cost` 保持会话作用域）：它聚合所有目录下每个会话的 `meta.json` 成一张面板——总计、最近 14 天有用量日期的按日柱状图，以及按模型细分并为配置了定价的模型给出美元估算（未配置定价的模型会列出但不计入总额）。按日数字来自此后记录的 `usageByDay` 桶映射；该字段出现之前记录的用量只有总计，显示为单独一行“earlier usage”（更早用量）。`/new` 丢弃内存中的对话并挂载一个全新的懒会话存储（旧会话保留在磁盘上）；删除是单独的显式操作——`/clear-sessions` 删除当前目录记录的会话，`/clear-sessions --all` 清空所有会话目录（包括 `meta.json` 不可读的），两者都保留当前活动会话；`star --clear-sessions [all]` 以非交互方式做同样的事并退出。
 
 会话首个回合完成后（第一条用户消息得到第一条助手回复），一个轻量后台 LLM 请求——同一模型，上限 20 个输出 token——把第一条用户消息总结为短标题（≤50 字符，与消息语言一致）并存入 `meta.json`，`/resume` 列表从中读取。标题生成是即发即忘的：它从不阻塞 REPL 或 print 模式输出，只在会话尚无标题时运行（恢复的会话保留已有标题；无标题的旧会话在下一回合后获得标题），请求失败时标题保持为空——无标题会话在列表中显示为 `(无标题)`。进程退出时未完成的请求被放弃。只要挂载着会话它就适用，包括用 `-r`/`-c` 恢复的 print 模式。