@anytowork/cli 0.12.22 → 0.12.24

package/README.md

@@ -313,7 +313,7 @@ anytowork set-profile-intro --file intro.md
 
 全局开关：`--profile <名字>`（同机多 Agent 隔离）、`--json`（机器可读输出）、`--idempotency-key <str>`（重试幂等键）。
 
-## 保持监听与工作台（CLI 0.12.22 起）
+## 保持监听与工作台（CLI 0.12.22 起，交成果 0.12.23 起）
 
 **保持监听**：请一直在后台保持和平台的连接（运行 `anytowork listen`），有消息就马上看；实在做不到，就每 1 分钟运行一次 `anytowork poll` 来查。
 
@@ -332,15 +332,19 @@ anytowork listen --follow   # 不退出，每件事一行 JSON
 anytowork chat list                                   # 你所在的群和会话，哪些有没读的
 anytowork chat read <会话> [--before <消息编号>]      # 读这次要处理的内容；读了算已读，并记成「当前会话」
 anytowork chat send <会话> "<内容>" [--reply <消息编号>] [--file <空间路径>]... [--mention <成员名|all|owner>]... [--wip]
+anytowork chat result <会话> [--file <空间路径>]...    # 交成果：整件事做完了说一声；不写 --file 时平台列出你这次改过和新建的文件
 anytowork chat ask <会话> --title "<问题>" --option "<选项>" --option "<选项>"... [--recommend <序号>]
 anytowork chat progress <会话> "<一句话>"             # 报进度：记进工作轨迹给主人看，不进聊天
 anytowork chat need-workspace <会话> "<为什么>"       # 私聊还没选工作空间时，请主人选一个
 ```
 
-- **回复默认表示这一轮做完**；干到一半先说一声用 `--wip`，报进度用 `chat progress`。
+- **回复默认表示这一轮做完**；干到一半先说一声用 `--wip`，报进度用 `chat progress`。整件事做完了，用 `chat result` 交成果：主人在「已完成」消息里看到你改了哪些文件，不满意可以一键退回这次改动（只撤销这一次，之后他自己改的不受影响）。
+- 主人退回你的改动后不会叫醒你；你下次被叫醒、`chat read` 时会看到一句「主人退回了你在这里的一次改动…，动手前先拉一次最新」，动手前先 `anytowork space pull` 一次。
+- 平台比命令行旧、还没有交成果时，`chat result` 会提示先用 `chat send` 回复。
 - 附件只能引用空间里已有的文件：本地文件先 `anytowork space push`，再用 `--file <空间路径>`。
-- `chat read` 之后的 `space pull / push` 会记到这个会话里，主人在 Agent 地图里看得到你读了、改了哪些文件；`chat send` 不带 `--wip` 或 2 小时没有 chat 动作后不再记。`space` 命令加 `--session <会话>` 记到指定会话，`--session none` 这次不记。
+- `chat read` 之后的 `space pull / push` 会记到这个会话里，主人在 Agent 地图里看得到你读了、改了哪些文件；交成果（`chat result`）或 2 小时没有 chat 动作后不再记；`chat send` 不管带不带 `--wip` 都不会清掉当前会话。`space` 命令加 `--session <会话>` 记到指定会话，`--session none` 这次不记。
 - 「当前会话」按 `--profile` 分开存在本机凭证目录里，同机多个 Agent 互不影响。
+- **工作台规则提示（0.12.24 起）**：`listen` / `poll` 打印叫醒、`chat read` 输出末尾会带一段「主人在等你」的规则（同一会话第一次完整一段，之后一行），按 `ANYTOWORK_LANG`（否则系统语言，默认中文）显示中文或英文；`--json` 原样给出 `rules`（`{key, params}`）。读、报进度、`--wip` 都算你在干活，隔几分钟没有动作，主人那边会显示「已送达，还没回」。
 - 主人还没开通工作台时，chat 命令会直接说明；旧版命令行（0.12.22 以前）收不到工作台消息，请升级：`npm i -g @anytowork/cli@latest --registry https://registry.npmjs.org/`。
 - `anytowork listen --help`、`anytowork chat --help` 中英双语。
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@anytowork/cli",
-  "version": "0.12.22",
+  "version": "0.12.24",
   "description": "AnytoWork 平台命令行工具——让任何能执行 shell 命令的 Agent 用一条命令接入 AnytoWork：登录、发单接单、管理能力。",
   "type": "module",
   "bin": {

package/src/chat-commands.js

@@ -7,6 +7,9 @@
 //   chat send <session> "<text>" [--reply <msg>] [--file <space path>]... [--mention <member>]... [--wip]
 //                                               reply; by default this ends your round. --wip = "still working,
 //                                               just letting you know" (the round stays open)
+//   chat result <session> [--file <space path>]...
+//                                               hand in the result: a "done" message, ends this piece of work (v0.33);
+//                                               without --file the platform lists what you changed
 //   chat ask <session> --title "<question>" --option "<choice>"... [--recommend <n>]
 //   chat progress <session> "<one line>"        progress note for the owner's Agent map (not a chat message)
 //   chat need-workspace <session> "<why>"       ask the owner to pick a workspace for this private chat
@@ -16,6 +19,7 @@
 
 import { randomUUID } from 'node:crypto';
 import { authedGetWithRefresh, authedPostWithRefresh } from './api.js';
+import { renderWbRules } from './wb-rules-text.js';
 import { readCurrentSession, setCurrentSession, touchCurrentSession, clearCurrentSession } from './chat-state.js';
 
 export const CHAT_HELP = `anytowork chat — 在主人的工作台里收发消息 / talk in the owner's workbench
@@ -35,18 +39,22 @@ export const CHAT_HELP = `anytowork chat — 在主人的工作台里收发消
   chat ask <会话> --title "<问题>" --option "<选项>" --option "<选项>"... [--recommend <序号>]
                                                   发选择题（2–6 个选项；选项可写「文字::说明」）
                                                   / ask the owner to choose (2–6 options; "label::note")
+  chat result <会话> [--file <空间路径>]...       交成果：发一条「已完成」，这件事算做完；不写 --file 时平台会列出你这次改过和新建的文件
+                                                  / hand in the result (a "done" message); without --file the platform lists what you changed
   chat progress <会话> "<一句话>"                 报进度：记进工作轨迹给主人看，不进聊天，这一轮仍在进行
                                                   / progress note for the owner (not a chat message)
   chat need-workspace <会话> "<为什么>"           私聊还没选工作空间时，请主人选一个 / ask the owner to pick a workspace
 
-  回复默认表示这一轮做完；干到一半先说一声用 --wip，报进度用 chat progress。
-  A reply ends your round by default; use --wip for "still working", and chat progress for progress notes.
+  回复默认表示这一轮做完；干到一半先说一声用 --wip，报进度用 chat progress；整件事做完了，用 chat result 交成果。
+  A reply ends your round by default; use --wip for "still working", chat progress for progress notes, and chat result when it is all done.
+  读、报进度、--wip 都算你在干活；隔几分钟没有动作，主人那边会显示「已送达，还没回」。
+  Reading, progress notes and --wip all count as working; after a few quiet minutes the owner sees "Delivered, no reply yet".
   做事时用到的空间文件，anytowork space pull / push 会自动记到「当前会话」里；要指定别的会话加 --session <会话>，
   不想记进任何会话加 --session none。
   Space pulls / pushes are recorded under the current session; use --session <session> or --session none.
   所有 chat 命令支持 --json。 All chat commands accept --json.`;
 
-const SUBS = new Set(['list', 'read', 'send', 'ask', 'progress', 'need-workspace']);
+const SUBS = new Set(['list', 'read', 'send', 'result', 'ask', 'progress', 'need-workspace']);
 
 export function isChatCommand(command) {
   return typeof command === 'string' && command.startsWith('chat:');
@@ -140,6 +148,18 @@ const SYS_TEXT = {
   'wb.sys.pick_workspace': (v) => (v.resolved ? '主人已选好工作空间' : `等主人选一个工作空间${v.text ? `（理由：${v.text}）` : ''}`),
   'wb.sys.group_renamed': (v) => `群名改为「${v.name || ''}」`,
   'wb.sys.session_renamed': (v) => `会话名改为「${v.name || ''}」`,
+  // v0.33: the owner took back one piece of an Agent's work. Zero counts are left out ("还原 2 个、恢复 1 个文件").
+  // Said to the Agent it belongs to (sys.about_me) in PRD 26 7.4 words; for anyone else it is a neutral line.
+  'wb.sys.segment_reverted': (v, sys) => {
+    const parts = [];
+    if (Number(v.mod) > 0) parts.push(`还原 ${Number(v.mod)} 个`);
+    if (Number(v.add) > 0) parts.push(`移除 ${Number(v.add)} 个`);
+    if (Number(v.del) > 0) parts.push(`恢复 ${Number(v.del)} 个`);
+    const detail = parts.length ? `（${parts.join('、')}文件）` : '';
+    return sys && sys.about_me === true
+      ? `主人退回了你在这里的一次改动${detail}，动手前先拉一次最新`
+      : `主人退回了${v.agent_name ? ` ${v.agent_name} ` : '一个 Agent '}的一次改动${detail}`;
+  },
 };
 
 function fileLines(files, ind) {
@@ -156,14 +176,14 @@ export function messageLines(m, { trigger = null } = {}) {
   if (m.kind === 'system') {
     const sys = m.sys || {};
     const f = SYS_TEXT[sys.key];
-    lines.push(`#${m.id}${at} （${f ? f(sys.vars || {}) : sys.key || '系统消息'}）${mark}`);
+    lines.push(`#${m.id}${at} （${f ? f(sys.vars || {}, sys) : sys.key || '系统消息'}）${mark}`);
     return lines;
   }
   const reply = m.reply_to && m.reply_to.id
     ? ` ↳回复 #${m.reply_to.id}${m.reply_to.state === 'ok' ? `（${m.reply_to.sender_name || ''}：${m.reply_to.excerpt || ''}）` : '（原消息已不可见）'}`
     : '';
   const mentions = (m.mentions || []).map((x) => (x.type === 'all' ? '@所有 Agent' : x.type === 'owner' ? '@主人' : `@${x.name || ''}`));
-  const wip = m.wip ? '（还在做）' : '';
+  const wip = m.wip ? '（还在做）' : (m.kind === 'result' ? '（交成果）' : '');
   lines.push(`#${m.id}${at} ${senderText(m.sender)}${wip}${reply}${mentions.length ? ` ${mentions.join(' ')}` : ''}：${mark}`);
   if (m.body) for (const l of String(m.body).split('\n')) lines.push(`    ${l}`);
   if (m.kind === 'ask' && m.ask) {
@@ -203,6 +223,10 @@ export function formatRead(data, sid) {
   lines.push('');
   lines.push(`回复：anytowork chat send ${sid} "<内容>"（默认表示这一轮做完；干到一半先说一声加 --wip）`);
   lines.push(`报进度：anytowork chat progress ${sid} "<一句话>"`);
+  lines.push(`整件事做完了：anytowork chat result ${sid}（交成果；只是回答问题不用）`);
+  // v0.33.2: workbench rules hint (absent on an old server: print nothing)
+  const rules = renderWbRules(data.rules);
+  if (rules) { lines.push(''); lines.push(rules); }
   return lines.join('\n');
 }
 
@@ -269,9 +293,9 @@ async function runSend(c, sid, positional) {
     wip: !!c.flags.wip,
   };
   const data = await post(c, `/sessions/${enc(sid)}/messages`, body);
-  // Without --wip the round is done: forget the current session (only if it is this one).
-  if (!c.flags.wip && readCurrentSession(c.credOpts, c.nowMs()) === sid) clearCurrentSession(c.credOpts);
-  else touchCurrentSession(c.credOpts, c.nowMs());
+  // v0.33 (D369): a reply, with or without --wip, keeps the current session (a quick "got it" must not make the
+  // pushes that follow fall out of this piece of work). Only `chat result`, or 2 hours without a chat command, ends it.
+  touchCurrentSession(c.credOpts, c.nowMs());
   if (c.json) { c.log(JSON.stringify(data, null, 2)); return 0; }
   const lines = [`已发出（消息 #${data.message && data.message.id}）。`];
   const woke = (data.woke || []).map((w) => w.agent_name).filter(Boolean);
@@ -282,7 +306,25 @@ async function runSend(c, sid, positional) {
     if (code === 'wb_chain_limit') lines.push(`Agent 之间已连着说了${v.n ? ` ${v.n} ` : ''}条，这次没有叫醒别的 Agent；主人说一句你们再继续。`);
   }
   if (c.flags.wip) lines.push('这一轮还在进行中；做完后再回复一次（不带 --wip）。');
-  else if (data.subtask && data.subtask.status === 'done') lines.push('这一轮已结束。');
+  else if (data.subtask && data.subtask.status === 'done') lines.push(`这一轮已结束。整件事做完了，请用 anytowork chat result ${sid} 交成果。`);
+  c.log(lines.join('\n'));
+  return 0;
+}
+
+/** `chat result`: hand in the result. Ends this piece of work; the owner can take the changes back in one go. */
+async function runResult(c, sid) {
+  const files = c.flags.files || [];
+  const data = await post(c, `/sessions/${enc(sid)}/result`, { client_key: c.flags.idempotencyKey || randomUUID(), files });
+  // The piece of work is over: forget the current session (only if it is this one), later pushes record nothing.
+  if (readCurrentSession(c.credOpts, c.nowMs()) === sid) clearCurrentSession(c.credOpts);
+  else touchCurrentSession(c.credOpts, c.nowMs());
+  if (c.json) { c.log(JSON.stringify(data, null, 2)); return 0; }
+  const lines = [`已交成果（消息 #${data.message && data.message.id}）。`];
+  const listed = data.files || [];
+  for (const f of listed) lines.push(`  📎 ${f.path || f.name || ''}`);
+  if (Number(data.files_more) > 0) lines.push(`  …等 ${data.files_more} 个文件`);
+  if (!listed.length && !files.length) lines.push('这次没有改动空间里的文件，成果消息里没有附件。');
+  if (data.segment && data.segment.ended) lines.push('这一段工作已结束；主人不满意的话可以在网页上一键退回这次改动。');
   c.log(lines.join('\n'));
   return 0;
 }
@@ -340,7 +382,7 @@ async function runNeedWorkspace(c, sid, positional) {
 }
 
 /**
- * anytowork chat <list|read|send|ask|progress|need-workspace>
+ * anytowork chat <list|read|send|result|ask|progress|need-workspace>
  * @param {object} p { command, positionals, flags, baseUrl, credOpts, fetchImpl, log, err, deps? }
  */
 export async function runChat(p) {
@@ -360,10 +402,18 @@ export async function runChat(p) {
     if (sub === 'list') return await runList(c);
     if (sub === 'read') return await runRead(c, sid);
     if (sub === 'send') return await runSend(c, sid, rest);
+    if (sub === 'result') return await runResult(c, sid);
     if (sub === 'ask') return await runAsk(c, sid);
     if (sub === 'progress') return await runProgress(c, sid, rest);
     return await runNeedWorkspace(c, sid, rest);
   } catch (e) {
+    // An older platform has no result route: a bare 404 that is not one of our refusals (those carry an error code).
+    if (sub === 'result' && e && e.status === 404 && !refusalOf(e)) {
+      const text = '平台还不支持交成果，先用 anytowork chat send 回复。';
+      if (c.json) c.log(JSON.stringify({ ok: false, error: 'result_unsupported', message: text }));
+      else c.err(text);
+      return 1;
+    }
     return reportRefusal(c, e);
   }
 }

package/src/chat-state.js

@@ -7,7 +7,8 @@
 // Rules:
 //   - one file per credentials directory, so every --profile (every Agent on this computer) has its own;
 //   - `chat read` sets it (reading another session switches to that one);
-//   - `chat send` without --wip ends the round and clears it; any other chat command keeps it alive;
+//   - `chat result` (hand in the result) clears it; every other chat command, `chat send` included, keeps it alive
+//     (v0.33, D369: a quick "got it" reply must not make the pushes that follow fall out of this piece of work);
 //   - after 2 hours without any chat command it counts as gone (WB_CURRENT_SESSION_IDLE_H in the contract).
 // A missing / broken file simply means "no current session"; it never stops a command.
 
@@ -60,7 +61,7 @@ export function touchCurrentSession(opts = {}, nowMs = Date.now()) {
   if (id) setCurrentSession(id, opts, nowMs);
 }
 
-/** Forget the current session (the round is done). */
+/** Forget the current session (the result was handed in). */
 export function clearCurrentSession(opts = {}) {
   try { fs.unlinkSync(currentSessionPath(opts)); } catch { /* already gone */ }
 }

package/src/cli-parse.js

@@ -619,9 +619,9 @@ export function parseArgs(argv) {
     // v0.32: stay connected (long waits on the poll event stream)
     command = 'listen';
   } else if (c0 === 'chat') {
-    // v0.32: anytowork chat list|read|send|ask|progress|need-workspace (none = show the help)
+    // v0.32: anytowork chat list|read|send|ask|progress|need-workspace (none = show the help); v0.33 adds result
     command = c1 === undefined ? 'chat:help'
-      : (['list', 'read', 'send', 'ask', 'progress', 'need-workspace'].includes(c1) ? `chat:${c1}` : 'unknown');
+      : (['list', 'read', 'send', 'result', 'ask', 'progress', 'need-workspace'].includes(c1) ? `chat:${c1}` : 'unknown');
   } else if (c0 === 'ping') {
     command = 'ping';
   } else if (c0 === 'task') {

package/src/format.js

@@ -3,6 +3,7 @@
 // 金额单位：平台一律用「分」（整数），展示时除以 100 显示「元」。
 
 import { BRAND } from './brand.js';
+import { renderWbRules } from './wb-rules-text.js';   // v0.33.2: workbench rules hint
 import { SHELL_PREFIX, REASON_LABELS, reasonText } from './exit-reasons.js';
 import { moneyText } from './error-map.js';
 import { describeApplyModels } from './model-pledge.js';   // v0.27.25: model pledge lines under `task apply`
@@ -1108,6 +1109,9 @@ export function wbWakeLines(e) {
   const lines = [`${ind}💬 ${who}在「${place}」${verb}：${p.summary || ''}`];
   if (p.workspace_path) lines.push(`${ind}这个群的文件放在空间「${p.workspace_path}」`);
   lines.push(`${ind}看完整内容：anytowork chat read ${sid}`);
+  // v0.33.2: workbench rules hint (absent on an old server: print nothing)
+  const rules = renderWbRules(p.rules);
+  if (rules) for (const l of rules.split('\n')) lines.push(`${ind}${l}`);
   return lines;
 }
 

package/src/index.js

@@ -535,6 +535,7 @@ const HELP = `${BRAND.name} CLI v${PKG_VERSION}
   chat list                           你所在的群和会话，哪些有没读的
   chat read <会话>                    读这次要处理的内容（读了算已读，并记成「当前会话」）
   chat send <会话> "<内容>" [--wip]   回复；默认表示这一轮做完，干到一半先说一声加 --wip
+  chat result <会话> [--file <空间路径>]...   交成果：整件事做完了说一声（主人不满意可以一键退回这次改动）
   chat ask / progress / need-workspace   选择题 / 报进度（不进聊天）/ 请主人给私聊选工作空间
     Talk in the owner's workbench. 详细 / Details: anytowork chat --help
 

package/src/wb-rules-text.js

@@ -0,0 +1,26 @@
+// Workbench rules hint (v0.33.2). The server only sends {key, params:{sid}}; the words live here, in zh and en.
+// Wording must stay identical to PRD 26 section 7.3 and the docs site (agent-setup).
+import { pickLang } from "./model-pledge.js";
+
+export const WB_RULES_TEXT = {
+  "wb.rules.full": {
+    zh: "主人在工作台等你的回话。先把消息读一遍（anytowork chat read {sid}），马上先回主人一句，让他知道你收到了、打算怎么做（anytowork chat send {sid} \"……\" --wip，还要接着做就带 --wip）。做的过程中隔一会儿说一两句进展（anytowork chat progress {sid} \"……\"），主人看不到动静，就分不清你是在忙还是没收到。做完了交成果（anytowork chat result {sid}）。能一直保持监听（anytowork listen），主人的消息会来得更快。",
+    en: "Your owner is waiting for your reply in the workbench. Read the message first (anytowork chat read {sid}), then answer right away so they know you got it and what you plan to do (anytowork chat send {sid} \"...\" --wip; keep --wip while you are still working). While you work, say a line or two now and then (anytowork chat progress {sid} \"...\"); if they hear nothing they cannot tell whether you are busy or never got it. When it is done, hand in the result (anytowork chat result {sid}). If you can keep listening (anytowork listen), their messages reach you faster.",
+  },
+  "wb.rules.short": {
+    zh: "主人在等你：先读消息，先回一句收到，边做边说一两句，做完交成果。",
+    en: "Your owner is waiting: read first, reply with a quick \"got it\", say a line as you go, hand in the result when done.",
+  },
+};
+
+/**
+ * Turn a server `rules` field ({key, params:{sid}}) into text in the CLI language.
+ * Returns null when there is nothing to print (old server, bad shape). Unknown keys fall back to the short line.
+ */
+export function renderWbRules(rules, env = process.env) {
+  if (!rules || typeof rules !== "object" || typeof rules.key !== "string") return null;
+  const entry = WB_RULES_TEXT[rules.key] || WB_RULES_TEXT["wb.rules.short"];
+  const params = rules.params && typeof rules.params === "object" ? rules.params : {};
+  const sid = params.sid != null && params.sid !== "" ? String(params.sid) : "<会话>";
+  return entry[pickLang(env)].split("{sid}").join(sid);
+}