dsh-global-todo 0.16.3 → 0.17.0

package/CHANGELOG.md

@@ -1,3 +1,33 @@
+## [0.17.0] — 2026-10-24
+
+Documentation and prompt text: the confirmation gate is the headline now, not the third bullet.
+Nothing in the write path changed — no field, no route, no validation rule — so upgrading is safe
+for existing stores.
+
+- **A release at all, because npm freezes documentation at publish time.** The README, `docs/` and
+  the plugin-page locales in the tarball are a snapshot; leading with the gate only reaches the npm
+  page and the installed docs through a version.
+
+- **Both READMEs lead with the gate.** It is claim 01 (was 03) and it names **both doors**: an
+  agent's `global_todo_done` (or a status change to `done`), and the host closing a dispatched todo
+  after a clean turn. A new early section — 「完成确认：没有一条记录能自己变成「完成」」 / *Confirmation:
+  no record finishes itself* — carries the two-door table, the two answers and the clean-turn rule;
+  the old 0.7 subsection and its row are gone rather than duplicated, and the comparison table gained
+  a *Finishing* row.
+- **The landing page, the npm description and the plugin-page locales open with the same sentence**:
+  page title, `meta`/`og`/`twitter` descriptions, hero lede, claim 01 (was 03), and the compare
+  table. `package.json` gained `human-in-the-loop`, `confirmation` and `approval`, which is how
+  someone looking for this actually searches.
+- **`docs/DESIGN.md` documents `review` for the first time.** It lived in `lib/model.js` and in no
+  design doc: there is now a *Done is a claim, from either door* section, a `review` field row, and
+  the `links` row finally admits `kind: 'image'` (0.15 shipped it).
+- **The model-facing half says it too.** The `global_todo_done` and `global_todo_update`
+  descriptions call a completion a claim; the done result prints 「这条挂在「待确认」…」; `todoLine()`
+  marks pending records in every rendering, so the model cannot read one as settled; and the injected
+  guidance tells the model to reply 「已标记完成，等你确认」 instead of 「已经结束了」. The stale
+  「插件也会在你安静下来后自动划掉」 line is gone — that idle-window heuristic was removed in 0.3.0
+  and the sentence outlived it by thirteen minor versions, telling the model the opposite of the rule.
+
 ## [0.16.3] — 2026-10-23
 
 Documentation release: the npm page finally shows the current install instructions.

package/README.md

@@ -5,14 +5,17 @@
 **A cross-session todo list for DeepSeek Harness: one file you own, written by you and your
 agent, and handed to a conversation in one click.**
 
-Three things it does that a session-local plan cannot:
-
-1. **One JSON file is the whole list.** `cat`, diff, commit it, put it on a NAS, carry it to
+Three things it does that a session-local plan cannot — the first is the one it is known for:
+
+1. **An agent saying "done" is a claim, whichever door it comes through.** Its own report — the
+   tool call, a status change — and the host closing the todo after a clean turn are both just
+   claims: the record waits in **待确认 / Needs review** until a person confirms it, or sends it
+   back with its completion cleared. A session-local plan never asks that question: the agent
+   rewrites the table and the work is "finished".
+2. **One JSON file is the whole list.** `cat`, diff, commit it, put it on a NAS, carry it to
    another machine. Nothing lives inside a session log.
-2. **Hand-off is a button, not a paragraph.** Four modes, the working directory checked, and
+3. **Hand-off is a button, not a paragraph.** Four modes, the working directory checked, and
    the todo's own **steps** arrive as a numbered plan.
-3. **An agent saying "done" is a claim.** It lands in **待确认 / Needs review** until a person
-   confirms it — or sends it back to the list.
 
 ![The panel in use: list, detail, steps, review, inbox, reminders, selection](docs/screenshots/tour.gif)
 
@@ -39,14 +42,44 @@ one record:
 | When it was added | not tracked | **`createdAt`, immutable, shown everywhere** |
 | Context | title + state | description + **documents and addresses** |
 | Hand-off | — | **one click: reuse the bound conversation or open a session** |
+| Finishing | whatever the agent writes | **waits in Needs review for a person** |
 | Storage | session event log | one JSON file you can `cat`, edit, diff and back up |
 
+## Confirmation: no record finishes itself
+
+This is the rule the plugin is known for, and the one that separates it from every list that lets
+the agent report its own progress: **an agent's "done" is a sentence it said.** The record is
+written as `done` and carries `review: { pending: true }`; the row shows a **Needs review** badge
+and the header a chip with the count, until a person answers.
+
+Both doors are gated, because both are only a claim:
+
+| Where "done" came from | The record | Who answers |
+|---|---|---|
+| the agent's `global_todo_done`, or a status change to `done` | `done` **+ pending review** | a person |
+| the dispatched conversation ended cleanly, so the host closed it (`turn/end` = `completed`) | `done` **+ pending review** | a person |
+| you clicked done in the panel | `done` | already a person |
+
+- **Two answers.** *Confirm done* drops the mark and records who confirmed it; *Send back* returns
+  the todo to `open`, clears `doneAt` and the completion record, and adds a `reopened` entry.
+  Nothing is deleted: title, description, steps, attachments and notes all stay.
+- **The automatic close goes through the same gate.** What the host writes on the agent's behalf is
+  thinner than what the agent could have written — only the closing message and the files that turn
+  changed — so it has even less business counting by itself.
+- **Only clean turns are candidates.** `error`, `aborted`, `interrupted` and `max-tokens` leave the
+  todo open. A failed job wearing a "done" label is the one failure mode that would make the whole
+  list untrustworthy.
+
+There is exactly one implementation of the rule: `completeTodo`, `applyPatch` and a status change in
+`lib/model.js` share the same `by === 'agent'` test, and the panel and the ten tools all go through
+it.
+
 ## What 0.9 and 0.10 added (and the two just before)
 
-Six surfaces these READMEs had never explained. Session read-back and evidence came in 0.9;
-Markdown and the weekly report in 0.10; steps in 0.8 and the review mark in 0.7, which until now
-only existed in the changelog. Each changes what a record can carry, or what you can do with it
-after the fact.
+Five surfaces these READMEs had never explained. Session read-back and evidence came in 0.9;
+Markdown and the weekly report in 0.10; steps in 0.8. Each changes what a record can carry, or what
+you can do with it after the fact. (The review mark came in 0.7 and now opens this document, instead
+of hiding among the new features.)
 
 ### Steps: a plan inside the todo
 
@@ -78,20 +111,6 @@ The tail is bounded and in memory: 200 conversations, 400 characters each, fille
 `assistant/message`. Nothing is written to disk, so a Host restart forgets it. A missing tail is
 normal — this process has not seen that conversation speak.
 
-### Review: "done" is a claim
-
-When an agent finishes (`global_todo_done`, or the watcher on a clean turn), the record becomes
-done **and** carries `review: { pending: true }`. The row gets a **Needs review** badge and the
-header a **Review** chip with the count.
-
-| Answer | Effect |
-|---|---|
-| **Confirm done** | stays done, the mark is dropped, the activity trail records who confirmed |
-| **Send back** | back to `open`; `doneAt` and the completion record are cleared; a `reopened` activity entry is added |
-
-Both answers are recorded as the person's action. Sending back deletes nothing: title,
-description, steps, attachments and progress notes stay.
-
 ### Markdown in and out
 
 The JSON export is the round-trip format. Markdown is the one you read, diff and edit by hand

package/README.zh.md

@@ -4,11 +4,13 @@
 
 **一份跨会话的待办清单：一个属于你的文件，你和 Agent 读写同一份，一键交给某个对话继续干。**
 
-三件会话级计划做不到的事：
+三件会话级计划做不到的事——第一件是这个插件最要紧的一条：
 
-1. **一个 JSON 文件就是全部**。能 `cat`、能 diff、能提交进仓库、能放 NAS、换机带走——不藏在任何会话日志里。
-2. **交接是一个按钮，不是一段话**。四种模式、会先检查工作目录，待办自己的**步骤**会作为编号计划一起交出去。
-3. **Agent 说「做完了」不算数**。它先进**待确认**，人点头才算完成，或者被退回清单。
+1. **Agent 说「做完了」不算数，两个入口都只是声明**。它自己在工具里报完成是一种，交给的会话干净收尾、
+   宿主替它收尾是另一种；两条路都只把记录送进**待确认**。人点「确认完成」才算完成，点「退回清单」
+   就回到清单里、完成记录一并清掉。会话级计划没人问这一句：Agent 把表一改，事情就算结束了。
+2. **一个 JSON 文件就是全部**。能 `cat`、能 diff、能提交进仓库、能放 NAS、换机带走——不藏在任何会话日志里。
+3. **交接是一个按钮，不是一段话**。四种模式、会先检查工作目录，待办自己的**步骤**会作为编号计划一起交出去。
 
 ![面板实况：列表、详情、步骤、待确认、收件箱、提醒、批量选择](docs/screenshots/tour.gif)
 
@@ -33,13 +35,38 @@ DSH 自带的 `todo_write` 属于**某一个 agent 会话**：整表替换、三
 | 加入时间 | 不记录 | **`createdAt` 只写一次，处处可见** |
 | 上下文 | 标题 + 状态 | 描述 + **文档与地址** |
 | 交给 Agent | 无 | **一键：复用已绑定会话，或新开一个** |
+| 标成完成 | Agent 改了就算 | **进「待确认」，人点头才算完成** |
 | 存储 | 会话事件日志 | 一个 JSON 文件，能 `cat`、能手改、能 diff、能备份 |
 
+## 完成确认：没有一条记录能自己变成「完成」
+
+这是它最容易被记住的一条，也是它和所有「让 Agent 自己报进度」的清单最不一样的地方：
+**Agent 说「做完了」，只是它说了这句话。** 记录写完是 `done`，同时带上
+`review: { pending: true }`；行上有「待确认」徽标，顶部有带计数的筛选，直到人回答为止。
+
+两个入口都受同一条规则管，因为它们同样只是声明：
+
+| 「做完了」从哪来 | 记录变成 | 谁来回答 |
+|---|---|---|
+| Agent 调 `global_todo_done`，或把状态改成 `done` | `done` **+ 待确认** | 人 |
+| 交给的会话干净收尾，宿主自动收尾（`turn/end` 是 `completed`） | `done` **+ 待确认** | 人 |
+| 你自己在面板里点完成 | `done` | 已经是人 |
+
+- **回答只有两个**：「确认完成」清掉标记，动态里记下是谁确认的；「退回清单」回到 `open`，
+  清空 `doneAt` 与完成记录，动态里加一条 `reopened`。标题、描述、步骤、附件、进展一条不删。
+- **自动收尾也走这个门**。宿主替 Agent 写的那份完成记录本来就更薄——只有最后一段回复和这一轮
+  改过的文件——它更没理由自己算数。
+- **只有干净收尾才是候选**：`error`、`aborted`、`interrupted`、`max-tokens` 一律留着未完成。
+  一轮失败的任务挂着「完成」标签，是让整份清单不可信的唯一起点。
+
+规则在代码里只有一个实现：`completeTodo`、`applyPatch` 和状态改写共用 `lib/model.js` 里那个
+`by === 'agent'` 判断，面板和十个工具都走它。
+
 ## 0.9 / 0.10 补了什么（以及再往前两版）
 
-六个这两份 README 一直没讲的能力：会话回读与原话是 0.9 的，Markdown 与本周报告是 0.10 的，
-步骤是 0.8 的，待确认标记是 0.7 的——后两个此前只写在 CHANGELOG 里。每一个都改变了记录能装什么，
-或者改变了事后能拿它做什么。
+五件这两份 README 一直没讲的能力：会话回读与原话是 0.9 的，Markdown 与本周报告是 0.10 的，
+步骤是 0.8 的。每一个都改变了记录能装什么，或者改变了事后能拿它做什么。
+（还有一件是 0.7 的「待确认」，它现在就是开头那条规则，不在这里当新功能讲。）
 
 ### 步骤：待办里的一份计划
 
@@ -67,18 +94,6 @@ DSH 自带的 `todo_write` 属于**某一个 agent 会话**：整表替换、三
 tail 有上限、只在内存里：200 个会话、每个 400 字，来自每条 `assistant/message`。不落盘，宿主一重启就忘。
 查不到 tail 是正常情况——这个进程没见那个对话说过话。
 
-### 待确认：Agent 的「做完了」只是声明
-
-Agent 收尾时（`global_todo_done`，或干净一轮后的自动收尾），记录会变成已完成，**同时**带上
-`review: { pending: true }`。行上有「待确认」徽标，顶部多一个带计数的「待确认」筛选。
-
-| 你的回答 | 结果 |
-|---|---|
-| **确认完成** | 保持已完成，标记清除，动态里记下是谁确认的 |
-| **退回清单** | 回到 `open`，`doneAt` 与完成记录清空，动态里加一条 `reopened` |
-
-两个回答都按**人**的动作记录。退回不删除任何东西：标题、描述、步骤、附件、进展记录都还在。
-
 ### Markdown 进与出
 
 JSON 导出是往返格式；Markdown 是拿来读、diff、手改的那份：

package/docs/DESIGN.md

@@ -18,7 +18,7 @@ so the file stays readable, hand-editable and diffable.
 | `priority` | `low` \| `normal` \| `high` \| `urgent` | `urgent` was added in front of the first three; `normal` stays the default |
 | `dueAt` | ISO-8601 string \| `null` | canonicalised to UTC on write |
 | `tags` | string[] | ≤ 20, each ≤ 32, de-duplicated case-insensitively |
-| `links` | `{ kind: 'url' \| 'file', value, title }[]` | documents and addresses |
+| `links` | `{ kind: 'url' \| 'file' \| 'image', value, title }[]` | documents, addresses and images |
 | `createdAt` | epoch ms | **written once, never rewritten** |
 | `updatedAt` | epoch ms | last accepted change |
 | `createdBy` | `human` \| `agent` | which side added it |
@@ -29,12 +29,32 @@ so the file stays readable, hand-editable and diffable.
 | `originSessionId` | string | the conversation that wrote it; the panel groups by this |
 | `lastDispatch` | `{ at, mode, ok, sessionId?, error? }` \| `null` | last hand-off, including its failure reason |
 | `completion` | `{ at, by, summary, sessionId, artifacts }` \| `null` | what the finishing turn said and produced; cleared on reopen |
+| `review` | `{ pending, at, by }` \| `null` | an agent's completion waiting for a person (*Done is a claim, from either door*) |
 | `notes` | `{ at, by, text }[]` | append-only progress notes, capped at `LIMITS.notes` |
 | `activity` | `{ at, by, kind, summary? }[]` | trimmed to `activityLimit` (default 50) |
 
 The activity kinds are `created`, `updated`, `status`, `done`, `reopened`, `linked`,
 `unlinked`, `dispatched`, `dispatch-failed`, `noted`, `completed`.
 
+## Done is a claim, from either door
+
+An agent "finishes" work in two places, and both are only claims. `global_todo_done` — or a status
+change to `done` through `applyPatch` — is the model saying so. The dispatch watcher (*Finishing by
+itself, without lying*) is the host inferring it from a clean `turn/end`. Either way `lib/model.js`
+sets `review: { pending: true, at, by }` when the writer is an agent, and the record keeps that mark
+until a person answers: `reviewTodo` clears it and records the person in the activity trail, or
+sending it back returns the todo to `open` with `doneAt` and `completion` cleared and everything
+else intact.
+
+Why one gate instead of trusting the better-evidenced path: a single wrong `done` makes the list
+worthless, and the two paths differ in how much evidence they carry, not in who is entitled to
+decide. The mirror matters as much — a person's own click never sets the mark, because marking your
+own work as unconfirmed would be noise. The badge means exactly one thing: *an agent said finished,
+no person has answered yet.*
+
+It is one function, not a convention. The panel's done button, `global_todo_done`, a bulk complete
+and the watcher all go through the same service path, so no writer can opt out of the gate.
+
 ## Storage
 
 ```

package/docs/FAQ.md

@@ -2,6 +2,10 @@
 
 14 个真问题，答案都能在代码里核对。文中路径相对仓库根；`$DSH_HOME` 默认是 `~/.dsh`。
 
+> **这个插件的中心规则：完成要人点头。** Agent 在工具里报完成，和交给的会话干净收尾后宿主自动
+> 收尾，都只把记录送进「待确认」；你点「确认完成」才算完成，点「退回清单」就回到清单里。为什么
+> 两条路都要人回答，见[第 3 问](#3-agent-把一条标成完成之后会发生什么)。
+
 ## 1. 我的数据到底存在哪？卸载会删掉吗？
 
 默认 `$DSH_HOME/global-todo/todos.json`，可以用插件配置的 `storePath` 指到任意绝对路径

package/docs/QUICKSTART.md

@@ -3,6 +3,9 @@
 第一次装 dsh-global-todo 的完整路径：装上、打开面板、写第一条带步骤的待办、交给一个对话、弄清
 「待确认」是什么，最后知道数据在哪、怎么备份。不解释设计，只说每一步点哪里。
 
+先说清它最要紧的一条规则，后面每一步都绕着它：**Agent 说「做完了」不算完成**——它自己在工具里
+报的、或者交给的会话干净收尾后宿主自动收尾的，都只进「待确认」，你点头才算（§5）。
+
 ## 0. 前提
 
 | 需要 | 检查 |

package/docs/ROADMAP.md

@@ -134,7 +134,8 @@ re-litigated every time someone tries to "clean it up".
 **External ingest with an inbox** — `POST /ingest` takes source-tagged, confidence-scored items
 behind a Bearer token; low-confidence records wait in an `inbox` tag view with 确认 / 丢弃 and a
 seven-day sweep, so an external source cannot silently fill the list.
-**An agent's "done" is a claim** — finishing as an agent sets `review = { pending }`; a person
+**An agent's "done" is a claim** — finishing as an agent sets `review = { pending }`, whether the
+model reported it or the dispatch watcher closed a clean turn; a person
 confirms or sends it back through `POST /review` (`reviewTodo()` in `lib/model.js`, `review()` in
 `lib/service.js`). Confirming keeps the record finished; rejecting reopens it and drops the
 completion record, because "not actually done" is the only honest reading. The panel shows a

package/lib/index.js

@@ -37,7 +37,7 @@ import { createStore, resolveStorePath } from './store.js'
 import { createTools, normalizeToolGroups, selectTools } from './tools.js'
 
 /** Must equal `package.json`'s version; `npm run check` enforces that. */
-export const VERSION = '0.16.3'
+export const VERSION = '0.17.0'
 
 /** Plugin id. Also the Loader row id in `cordis.patch.yml`. */
 export const name = 'dsh-global-todo'
@@ -104,7 +104,8 @@ const GUIDANCE = [
   '记的时候：标题一句话说清要做什么；把相关文档**绝对路径**或网址挂进 `links`（接手时会作为上下文一起交出去）；需要精确时间的写 `dueAt`。',
   '**大活拆步骤**：一件需要多轮才能做完的事，用 `steps` 记下 3–6 步（`global_todo_add` 的 `steps` 参数）——派发时这些步骤会作为编号计划原样交给接手的 Agent，比一段散文有用得多；做完一步就用 `global_todo_update` 把该步的 `done` 置为 true。',
   '记完用一句话说明记了什么（别复述整条描述）。做完用 `global_todo_done` 留痕，**不要删**。',
-  '**一轮收尾时顺手划掉**：如果清单里某条就是你刚做完的事，直接 `global_todo_done` 结束它，别留着占地方（插件也会在你安静下来后自动划掉，但你先划更准）。',
+  '**完成要人点头**：`global_todo_done`（或把状态改成 `done`）只是声明——记录会挂「待确认」，用户回答之后才算完成。所以在回复里说「已标记完成，等你确认」，别说「已经结束了」。',
+  '**一轮收尾时顺手划掉**：如果清单里某条就是你刚做完的事，直接 `global_todo_done` 结束它，并写清 `summary` 与 `artifacts`。交给会话的那条待办，宿主会在这一轮干净结束时自动收尾，但那样只留下最后一段回复和改动的文件——两条路都要人确认。',
   '做不到 / 卡住了：用 `global_todo_note` 追加一条进展，说明卡在哪、需要用户做什么——**不要**改写待办的 description（那是用户自己的文字）。',
   '用户说「抓一下待办 / 整理一下要跟进的」时：扫最近对话，用 `global_todo_capture` 一次登记多条，再回一句汇总。',
 ].join('\n')

package/lib/tools.js

@@ -229,6 +229,9 @@ export function todoLine(todo) {
   const mark = todo.status === 'done' ? 'x' : todo.status === 'doing' ? '~' : todo.status === 'cancelled' ? '-' : ' '
   const label = Number.isInteger(todo.seq) && todo.seq > 0 ? `#${todo.seq} ` : ''
   const parts = [`- [${mark}] ${label}${todo.title}  id=${todo.id}`, `加入于 ${new Date(todo.createdAt).toISOString()}`]
+  // An agent's "done" is a claim: every rendering that shows the mark shows that it is unanswered,
+  // so the model cannot read a completed record as settled and report it as finished.
+  if (todo.review?.pending === true) parts.push('待确认（人还没回答）')
   if (todo.priority !== 'normal') parts.push(PRIORITY_LABELS[todo.priority] ?? todo.priority)
   if (todo.dueAt !== null) parts.push(`截止 ${todo.dueAt}`)
   if (todo.tags.length > 0) parts.push(todo.tags.map((tag) => `#${tag}`).join(' '))
@@ -625,7 +628,8 @@ export function createTools(service, options = {}) {
     {
       name: 'global_todo_update',
       description:
-        '改一条全局待办。只传要改的字段；`dueAt: ""` 清空截止时间，`tags: []` 清空标签，`links` / `steps` 是整体替换（要单独加附件请用 global_todo_link），`status` 可以设为 archived（归档，不删除）。',
+        '改一条全局待办。只传要改的字段；`dueAt: ""` 清空截止时间，`tags: []` 清空标签，`links` / `steps` 是整体替换（要单独加附件请用 global_todo_link），`status` 可以设为 archived（归档，不删除）。'
+        + '`status` 设成 `done` 和 `global_todo_done` 一样只是声明：会挂「待确认」，用户点头才算完成。',
       parameters: {
         type: 'object',
         properties: {
@@ -801,8 +805,10 @@ export function createTools(service, options = {}) {
         + '标为完成时**务必带上 `summary`**（一句话说明你做了什么、结论是什么），'
         + '以及 `artifacts`（这次产出的文件路径或相关网址）——'
         + '清单上留下的完成记录就是这个，缺了它「已完成」等于什么都没说。'
+        + '**标完成只是声明**：记录会进「待确认」，用户点头才算完成，所以在回复里别把'
+        + '「我已标记完成」说成「这件事已经结束了」。'
         + '（交付给会话的待办即使你忘了调用，宿主也会在这一轮干净结束时自动收尾，'
-        + '但那样只有最后一段回复和改动文件，不如你主动写清楚。）',
+        + '但那样只有最后一段回复和改动文件，不如你主动写清楚；那条同样要人确认。）',
       parameters: {
         type: 'object',
         properties: {
@@ -838,6 +844,9 @@ export function createTools(service, options = {}) {
         render: (_args, value) => {
           if (value.ok !== true) return text(`状态没改成功：${value.error ?? value.reason ?? '未知原因'}`)
           const lines = [`已更新状态：${value.todo.status}`, todoLine(value.todo)]
+          if (value.todo.status === 'done' && value.todo.review?.pending === true) {
+            lines.push('', '这条挂在「待确认」：标完成只是声明，用户点头才算完成——回复里别把它说成已经收尾。')
+          }
           const completion = value.todo.completion
           if (completion !== null && completion !== undefined) {
             lines.push('', `完成于 ${new Date(completion.at).toISOString()}（${completion.by}）`)

package/locale/en.json

@@ -1,6 +1,6 @@
 {
   "meta": {
     "title": "Global Todos",
-    "description": "Cross-session todos for DeepSeek Harness. One JSON file you own; a person and an agent write the same records. Every item keeps when it was added, a description, attached documents and links, an optional plan of steps, and the sentence it came from. Hand it to a conversation in one click - the steps travel as a numbered plan - then read back what that conversation last said. An agent's \"done\" stays a claim until you confirm it."
+    "description": "Cross-session todos for DeepSeek Harness whose central rule is that an agent's \"done\" is not completion. Whether the agent reports it from a tool or the host infers it when a dispatched conversation ends cleanly, the record waits in Needs review: a person confirms it, or it goes back to the list. One JSON file you own, written by both sides, with created time, description, attached documents and links, a plan of steps and the sentence it came from; hand an item to a conversation in one click and read back what that conversation last said."
   }
 }

package/locale/zh.json

@@ -1,6 +1,6 @@
 {
   "meta": {
     "title": "全局待办",
-    "description": "DeepSeek Harness 的跨会话待办清单。一份属于你的 JSON 文件，人和 Agent 读写同一条记录：每条记下加入时间、描述、可挂文档与地址、可选的步骤计划，以及它来自哪句原话。一键交给某个对话继续干（步骤会作为编号计划一起交出去），随时回读那边最后说了什么；Agent 标的「完成」只是声明，你确认之后才算数。"
+    "description": "DeepSeek Harness 的跨会话待办清单，中心规则是：Agent 说「做完了」不算完成。它自己在工具里报完成，或交给的会话干净收尾后宿主自动收尾，两条路都只进「待确认」，人点头才算完成、退回就回清单。一份属于你的 JSON 文件，人和 Agent 读写同一条记录：加入时间、描述、文档与地址、步骤计划、原话；一键交给某个对话继续干，随时回读那边最后说了什么。"
   }
 }

package/package.json

@@ -1,8 +1,8 @@
 {
   "name": "dsh-global-todo",
-  "version": "0.16.3",
+  "version": "0.17.0",
   "icon": "./docs/icon.svg",
-  "description": "Cross-session todo list for DeepSeek Harness: one JSON file a person and an agent both write to, with step plans that ship as a numbered hand-off, session read-back, and a human confirmation before an agent's \"done\" counts.",
+  "description": "Cross-session todo list for DeepSeek Harness where an agent's \"done\" never settles itself: whether it is reported by the tool or inferred from a clean turn, the record waits in Needs review until a person confirms it. One JSON file both write, numbered step hand-offs, session read-back.",
   "type": "module",
   "license": "MIT",
   "author": "LUgABeEEE",
@@ -27,6 +27,9 @@
     "task-management",
     "handoff",
     "agent",
+    "human-in-the-loop",
+    "confirmation",
+    "approval",
     "productivity",
     "待办",
     "全局待办"
@@ -86,7 +89,8 @@
     "check:matrix": "node scripts/matrix.mjs",
     "market": "node tools/market.mjs",
     "market:check": "node tools/market.mjs --check",
-    "market:fresh": "node tools/market.mjs --fresh"
+    "market:fresh": "node tools/market.mjs --fresh",
+    "release:check": "node tools/release-check.mjs"
   },
   "dsh": {
     "manifestVersion": 1,