@danceiny/gotry 0.0.1-rc.9 → 0.2.0-rc.28

package/dist/data/sf-golden-manifest.json

@@ -0,0 +1,103 @@
+{
+  "comment": "sf-live-benchmark golden(issue #21 验收「字段准确率 ≥90%」对照源)。本表=公开班期 + 价格区间,不绑任何 vendor;字段评分子集硬命中(query_id/from/to/currency/source/verdict)+ 时间/价格/班次软命中(±60min/±15%/子串包含)。sf-01 fixture 来自 ts/scripts/session-benchmark.ts:26-42 已知期望;其余 7 条以公开班期知识写时间窗口与价格带(具体班次由 session 真实落,匹配子串算 correct)。",
+  "threshold": 0.9,
+  "query_manifest_sha256": "3be197830e5a6aa6e1309684a1fce4a736ad2a845c67b6f2f1b8a4ad3ccf977a",
+  "matches": [
+    {
+      "query_id": "sf-01",
+      "from": "上海",
+      "to": "丽江",
+      "date": "2026-10-01",
+      "window_dep_local": { "earliest": "06:00", "latest": "22:00" },
+      "window_arr_local": { "earliest": "10:00", "latest": "23:59" },
+      "duration_min": { "min": 180, "max": 720 },
+      "price_cny": { "min": 800, "max": 4000 },
+      "transport_hint": "直飞或 1 段中转,承运含 MU/HO/9C/CA/CZ 之一",
+      "known_flights": ["HO5577", "MU6145", "9C8779"]
+    },
+    {
+      "query_id": "sf-02",
+      "from": "北京",
+      "to": "大理",
+      "date": "2026-10-02",
+      "window_dep_local": { "earliest": "06:00", "latest": "22:00" },
+      "window_arr_local": { "earliest": "10:00", "latest": "23:59" },
+      "duration_min": { "min": 240, "max": 900 },
+      "price_cny": { "min": 800, "max": 4500 },
+      "transport_hint": "直飞稀,多经昆明/成都中转 1-2 段,承运含 CA/MU/3U/CZ",
+      "known_flights": ["CA1441", "MU5712", "3U8831"]
+    },
+    {
+      "query_id": "sf-03",
+      "from": "上海",
+      "to": "三亚",
+      "date": "2026-11-11",
+      "window_dep_local": { "earliest": "06:00", "latest": "22:00" },
+      "window_arr_local": { "earliest": "10:00", "latest": "23:59" },
+      "duration_min": { "min": 180, "max": 600 },
+      "price_cny": { "min": 500, "max": 3000 },
+      "transport_hint": "旺季直飞多,承运含 9C/MU/HU/CZ/CA",
+      "known_flights": ["9C8779", "MU5377", "HU7177", "CZ6766"]
+    },
+    {
+      "query_id": "sf-04",
+      "from": "广州",
+      "to": "昆明",
+      "date": "2026-12-20",
+      "window_dep_local": { "earliest": "06:00", "latest": "22:00" },
+      "window_arr_local": { "earliest": "09:00", "latest": "23:59" },
+      "duration_min": { "min": 150, "max": 600 },
+      "price_cny": { "min": 400, "max": 2500 },
+      "transport_hint": "直飞密集,承运含 CZ/MU/3U/8L",
+      "known_flights": ["CZ3497", "MU5738", "3U8805", "8L9628"]
+    },
+    {
+      "query_id": "sf-05",
+      "from": "深圳",
+      "to": "成都",
+      "date": "2026-10-05",
+      "window_dep_local": { "earliest": "06:00", "latest": "22:00" },
+      "window_arr_local": { "earliest": "09:00", "latest": "23:59" },
+      "duration_min": { "min": 150, "max": 600 },
+      "price_cny": { "min": 400, "max": 2500 },
+      "transport_hint": "直飞密集,承运含 CZ/MU/CA/3U/ZH",
+      "known_flights": ["CZ3453", "MU5402", "CA4314", "3U8744"]
+    },
+    {
+      "query_id": "sf-06",
+      "from": "杭州",
+      "to": "厦门",
+      "date": "2026-10-06",
+      "window_dep_local": { "earliest": "06:00", "latest": "22:00" },
+      "window_arr_local": { "earliest": "09:00", "latest": "23:59" },
+      "duration_min": { "min": 90, "max": 360 },
+      "price_cny": { "min": 300, "max": 2000 },
+      "transport_hint": "短途直飞,承运含 GJ/MU/CZ/CA",
+      "known_flights": ["GJ7153", "MU5520", "CZ6955"]
+    },
+    {
+      "query_id": "sf-07",
+      "from": "西安",
+      "to": "桂林",
+      "date": "2026-10-07",
+      "window_dep_local": { "earliest": "06:00", "latest": "22:00" },
+      "window_arr_local": { "earliest": "09:00", "latest": "23:59" },
+      "duration_min": { "min": 150, "max": 600 },
+      "price_cny": { "min": 400, "max": 2500 },
+      "transport_hint": "直飞较稀,多中转 1 段,承运含 MU/CZ/9C/HU",
+      "known_flights": ["JD5143", "MU2176", "CZ6319"]
+    },
+    {
+      "query_id": "sf-08",
+      "from": "重庆",
+      "to": "贵阳",
+      "date": "2026-10-08",
+      "window_dep_local": { "earliest": "06:00", "latest": "22:00" },
+      "window_arr_local": { "earliest": "09:00", "latest": "23:59" },
+      "duration_min": { "min": 60, "max": 300 },
+      "price_cny": { "min": 250, "max": 1800 },
+      "transport_hint": "极短途直飞,承运含 G5/CZ/MU/GY",
+      "known_flights": ["G52667", "CZ5817", "MU2146", "GY7122"]
+    }
+  ]
+}

package/dist/data/sf-static-routes.json

@@ -0,0 +1,91 @@
+{
+  "schema_version": "sf-static-routes.v1",
+  "source": {
+    "name": "OpenFlights routes.dat",
+    "url": "https://raw.githubusercontent.com/jpatokal/openflights/4b969f8e91eb800c45f0e0e2355a0fbb93de27e4/data/routes.dat",
+    "revision": "4b969f8e91eb800c45f0e0e2355a0fbb93de27e4",
+    "license": "ODbL-1.0",
+    "retrieved_at": "2026-08-30",
+    "limitations": [
+      "route/carrier coverage only",
+      "no schedule",
+      "no price",
+      "no availability"
+    ]
+  },
+  "routes": [
+    {
+      "query_id": "sf-01",
+      "from_city": "上海",
+      "to_city": "丽江",
+      "route_kind": "direct",
+      "legs": [
+        { "from_iata": "SHA", "to_iata": "LJG", "carrier_codes": ["MU"] }
+      ]
+    },
+    {
+      "query_id": "sf-02",
+      "from_city": "北京",
+      "to_city": "大理",
+      "route_kind": "transfer",
+      "legs": [
+        { "from_iata": "PEK", "to_iata": "KMG", "carrier_codes": ["8L", "CA", "CZ", "HU", "MU", "ZH"] },
+        { "from_iata": "KMG", "to_iata": "DLU", "carrier_codes": ["8L", "CZ", "MU"] }
+      ]
+    },
+    {
+      "query_id": "sf-03",
+      "from_city": "上海",
+      "to_city": "三亚",
+      "route_kind": "direct",
+      "legs": [
+        { "from_iata": "PVG", "to_iata": "SYX", "carrier_codes": ["9C", "CZ", "FM", "HO", "MU"] }
+      ]
+    },
+    {
+      "query_id": "sf-04",
+      "from_city": "广州",
+      "to_city": "昆明",
+      "route_kind": "direct",
+      "legs": [
+        { "from_iata": "CAN", "to_iata": "KMG", "carrier_codes": ["CA", "CZ", "KY", "MU", "ZH"] }
+      ]
+    },
+    {
+      "query_id": "sf-05",
+      "from_city": "深圳",
+      "to_city": "成都",
+      "route_kind": "direct",
+      "legs": [
+        { "from_iata": "SZX", "to_iata": "CTU", "carrier_codes": ["3U", "CA", "CZ", "EU", "HU", "TV", "ZH"] }
+      ]
+    },
+    {
+      "query_id": "sf-06",
+      "from_city": "杭州",
+      "to_city": "厦门",
+      "route_kind": "direct",
+      "legs": [
+        { "from_iata": "HGH", "to_iata": "XMN", "carrier_codes": ["CA", "CZ", "MF", "MU", "SC"] }
+      ]
+    },
+    {
+      "query_id": "sf-07",
+      "from_city": "西安",
+      "to_city": "桂林",
+      "route_kind": "direct",
+      "legs": [
+        { "from_iata": "XIY", "to_iata": "KWL", "carrier_codes": ["CA", "CZ", "GS", "JD", "MU", "NS", "ZH"] }
+      ]
+    },
+    {
+      "query_id": "sf-08",
+      "from_city": "重庆",
+      "to_city": "贵阳",
+      "route_kind": "direct",
+      "legs": [
+        { "from_iata": "CKG", "to_iata": "KWE", "carrier_codes": ["CA", "CZ", "G5", "GS", "SC"] }
+      ]
+    }
+  ]
+}

package/dist/scripts/action-cache-tests.js

@@ -1,3 +1,8 @@
+/**
+ * action-cache 自愈层确定性测试(RFC P2;run-all §25):
+ * 命中/指纹失配/TTL 过期/变量化 key/LRU 淘汰/损坏文件容错/持久化跨实例/幂等回写。
+ * 全纯函数 + 临时目录,零浏览器零网络。
+ */
 import { mkdtempSync, rmSync, writeFileSync } from 'node:fs';
 import { tmpdir } from 'node:os';
 import { join } from 'node:path';
@@ -8,7 +13,8 @@ function assert(cond, label, detail) {
     if (cond) {
         pass += 1;
         console.log(`  ok - ${label}`);
-    } else {
+    }
+    else {
         fail += 1;
         console.log(`  FAIL - ${label}${detail !== undefined ? ' :: ' + String(detail) : ''}`);
     }
@@ -17,91 +23,50 @@ const iso = mkdtempSync(join(tmpdir(), 'gotry-actioncache-'));
 const path = actionCachePath(iso);
 const T0 = new Date('2026-08-28T10:00:00Z');
 let clock = T0.getTime();
-const now = ()=>new Date(clock);
+const now = () => new Date(clock);
 const site = 'ctrip-flight';
 console.log('A. templateKey(参数变量化)');
-const k1 = templateKey(site, 'search', {
-    from: '上海',
-    to: '丽江',
-    date: '2026-10-01'
-});
-const k2 = templateKey(site, 'search', {
-    from: '上海',
-    to: '丽江',
-    date: '2026-11-11'
-});
+const k1 = templateKey(site, 'search', { from: '上海', to: '丽江', date: '2026-10-01' });
+const k2 = templateKey(site, 'search', { from: '上海', to: '丽江', date: '2026-11-11' });
 assert(k1 === k2 && k1 === 'ctrip-flight:search:上海|丽江|%date%', '同型查询共享模板 key(日期变量化),跨日期复用');
 console.log('B. 命中与指纹失配(被动失效)');
-const c = new ActionCache(path, {
-    now
-});
+const c = new ActionCache(path, { now });
 const fpA = fingerprint('batchSearch v1 shape {"flightItineraryList":[]}');
-const fpB = fingerprint('batchSearch v2 shape {"itineraries":[]}');
-c.record(k1, fpA, {
-    entryUrl: 'https://flights.ctrip.com/online/list/oneway-sha-ljg?depdate=%date%',
-    parser: 'batchSearch.v1'
-}, site);
+const fpB = fingerprint('batchSearch v2 shape {"itineraries":[]}'); // 接口改版
+c.record(k1, fpA, { entryUrl: 'https://flights.ctrip.com/online/list/oneway-sha-ljg?depdate=%date%', parser: 'batchSearch.v1' }, site);
 clock += 60_000;
 assert(c.lookup(k1, fpA, site)?.locator.parser === 'batchSearch.v1', '指纹一致 → 命中并回放确定性载荷');
 assert(c.lookup(k1, fpA, site)?.hits === 2, '命中计 hits');
 assert(c.lookup(k1, fpB, site) === null, '接口改版(指纹失配)→ miss 不误放(错误的缓存点击比慢更糟)');
-assert(c.lookup(templateKey(site, 'search', {
-    from: '北京',
-    to: '大理',
-    date: '2026-10-01'
-}), fpA, site) === null, 'key 不同 → miss');
+assert(c.lookup(templateKey(site, 'search', { from: '北京', to: '大理', date: '2026-10-01' }), fpA, site) === null, 'key 不同 → miss');
 console.log('C. miss 回写与幂等');
-c.record(k1, fpB, {
-    entryUrl: 'oneway-bjs-dlu?depdate=%date%',
-    parser: 'batchSearch.v2'
-}, site);
+c.record(k1, fpB, { entryUrl: 'oneway-bjs-dlu?depdate=%date%', parser: 'batchSearch.v2' }, site);
 assert(c.lookup(k1, fpB, site)?.locator.parser === 'batchSearch.v2', '改版后回写新载荷,下次命中');
 const hitsBefore = c.lookup(k1, fpB, site).hits;
-c.record(k1, fpB, {
-    parser: 'batchSearch.v2'
-}, site);
+c.record(k1, fpB, { parser: 'batchSearch.v2' }, site);
 assert(c.lookup(k1, fpB, site).hits === hitsBefore + 1, '重复 record 幂等(hits/createdAt 保留,不重置)');
 console.log('D. TTL 过期(48h 默认;注入时钟)');
 clock += 49 * 3600_000;
 assert(c.lookup(k1, fpB, site) === null && c.size(site) === 0, '超 48h → 过期删除条目');
 console.log('E. LRU 淘汰(每站点上限)');
-const c2 = new ActionCache(path + '.lru', {
-    now,
-    maxEntriesPerSite: 3
-});
-for(let i = 0; i < 5; i++){
+const c2 = new ActionCache(path + '.lru', { now, maxEntriesPerSite: 3 });
+for (let i = 0; i < 5; i++) {
     const key = `ctrip-flight:search:城${i}|丽江|%date%`;
-    c2.record(key, fingerprint(`shape-${i}`), {
-        i: String(i)
-    }, site);
+    c2.record(key, fingerprint(`shape-${i}`), { i: String(i) }, site);
     clock += 1_000;
 }
 assert(c2.size(site) === 3, '超容 LRU 淘汰至 3 条');
 console.log('F. 持久化跨实例 + 损坏文件容错');
-const c3 = new ActionCache(path + '.p', {
-    now
-});
-c3.record('meituan:search:大理|民宿|%date%', fingerprint('mt-shape'), {
-    url: 'https://www.meituan.com/...'
-}, 'meituan');
+const c3 = new ActionCache(path + '.p', { now });
+c3.record('meituan:search:大理|民宿|%date%', fingerprint('mt-shape'), { url: 'https://www.meituan.com/...' }, 'meituan');
 clock += 5_000;
-const c4 = new ActionCache(path + '.p', {
-    now
-});
+const c4 = new ActionCache(path + '.p', { now });
 assert(c4.lookup('meituan:search:大理|民宿|%date%', fingerprint('mt-shape'), 'meituan') !== null, '跨实例:落盘后新实例可命中');
 writeFileSync(path + '.broken', '{not json', 'utf8');
-const c5 = new ActionCache(path + '.broken', {
-    now
-});
+const c5 = new ActionCache(path + '.broken', { now });
 assert(c5.size() === 0 && c5.lookup('any', 'any', site) === null, '损坏 JSON 按空缓存,不抛错');
 console.log('G. fingerprint 敏感度');
 assert(fingerprint('abc') !== fingerprint('abd') && fingerprint('abc') === fingerprint('abc'), '微改即变,同文稳定');
-rmSync(iso, {
-    recursive: true,
-    force: true
-});
+rmSync(iso, { recursive: true, force: true });
 console.log(`\nACTION-CACHE: ${pass} pass, ${fail} fail`);
 process.exit(fail > 0 ? 1 : 0);
-
-
-//# sourceURL=/Users/bytedance/work/gotry/ts/scripts/action-cache-tests.ts

package/dist/scripts/agent-planning-turn-deadline-e2e.js

@@ -0,0 +1,242 @@
+/**
+ * Offline headless E2E for the agent turn boundary (ADR-24 v2).
+ *
+ * Exercises a clean installed-package binary on the PRODUCT path:
+ *   bin/gotry-inner.js -> dist/src/index.js -> dsh headless -> local SSE relay
+ *
+ * 场景是失败轨迹的合成重放:deep-planning 类用户消息(10.3 婚礼 + 十几天
+ * + 请假 + IRW)进入路由,relay 把首个 planner 响应拖过硬阈(阈值经 env
+ * 压到秒级,exit 仍由路由决定为 handoff)→ 首个工具派发被拒并落
+ * `gotry_turn_handoff.v1` 工单 → 工具 schema 同步抑制 → 下一请求
+ * text-only → 二进制以含 ETA 语义的 final 收尾。工单落盘走
+ * GOTRY_TURN_HANDOFF_ROOT 钉死的隔离根,不触碰真实 gotry-state
+ * (巡检状态纪律);DSH_HOME 与 child cwd 均为临时目录,唯一模型端点
+ * 是 loopback。
+ */
+import assert from 'node:assert/strict';
+import { createServer } from 'node:http';
+import { existsSync, mkdtempSync, readdirSync, readFileSync, rmSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+import { spawn } from 'node:child_process';
+const ROOT = join(import.meta.dirname, '..', '..');
+const TS_DIR = join(import.meta.dirname, '..');
+const TIMEOUT_MS = 90_000;
+const TOOL = 'gotry_artifacts_list';
+const SOFT_MS = 5_000;
+const HARD_MS = 10_000;
+const RELAY_STALL_MS = HARD_MS + 1_000;
+const HANDOFF = 'TURN_DEADLINE_HANDOFF';
+const HANDOFF_SCHEMA = 'gotry_turn_handoff.v1';
+/** 失败轨迹的合成等价消息:必须命中 deep-planning 路由。 */
+const DEEP_MESSAGE = '2026年我还有6天IRW额度,我准备再请几天假,在国内待个十几天。'
+    + '其中10.3要去湖南衡阳参加同学婚礼。请根据我的实际情况和偏好,给我安排行程';
+function sse(data) {
+    return `data: ${JSON.stringify(data)}\n\n`;
+}
+function toolResponse(idSuffix) {
+    // 0.2.0-rc.2 线型:Messages 流事件(message_start/content_block_*/message_stop)。
+    return sse({ type: 'message_start', message: { usage: { input_tokens: 10 } } })
+        + sse({ type: 'content_block_start', index: 0, content_block: { type: 'tool_use', id: `deadline-call-${idSuffix}`, name: TOOL, input: {} } })
+        + sse({ type: 'content_block_delta', index: 0, delta: { type: 'input_json_delta', partial_json: JSON.stringify({ query: { limit: 1 } }) } })
+        + sse({ type: 'content_block_stop', index: 0 })
+        + sse({ type: 'message_delta', delta: { stop_reason: 'tool_use' }, usage: { output_tokens: 5 } })
+        + sse({ type: 'message_stop' }) + 'data: [DONE]\n\n';
+}
+function textResponse(text, _id = 'deadline-final') {
+    return sse({ type: 'message_start', message: { usage: { input_tokens: 10 } } })
+        + sse({ type: 'content_block_start', index: 0, content_block: { type: 'text', text } })
+        + sse({ type: 'content_block_stop', index: 0 })
+        + sse({ type: 'message_delta', delta: { stop_reason: 'end_turn' }, usage: { output_tokens: 4 } })
+        + sse({ type: 'message_stop' }) + 'data: [DONE]\n\n';
+}
+async function startRelay() {
+    const bodies = [];
+    const plannerBodies = [];
+    const server = createServer((req, res) => {
+        // 0.2.0-rc.2 起 SDK 调 /v1/messages;保留 /chat/completions 兼容旧线型。
+        if (req.method !== 'POST' || !(req.url?.endsWith('/chat/completions') || req.url?.endsWith('/messages'))) {
+            res.writeHead(404).end('not found');
+            return;
+        }
+        const chunks = [];
+        req.on('data', chunk => chunks.push(Buffer.from(chunk)));
+        req.on('end', () => {
+            let body;
+            try {
+                body = JSON.parse(Buffer.concat(chunks).toString('utf8'));
+            }
+            catch {
+                body = {};
+            }
+            bodies.push(body);
+            res.writeHead(200, { 'content-type': 'text/event-stream', 'cache-control': 'no-cache' });
+            if (toolNames(body).includes(TOOL)) {
+                plannerBodies.push(body);
+                const plannerIndex = plannerBodies.length;
+                // issue #205:已交接判定只认真实 role=tool 消息(persona (21) 的
+                // TURN_DEADLINE_HANDOFF 字面量也在 system 文本里,整包扫描会首轮误判
+                // 已交接 → 永不派发工具 → handoff 语义测试退化为纯文本收尾)。
+                // 新线型里工具结果以 user 消息携带 tool_result 块;旧线型是 role='tool'。
+                const realToolMessages = (body.messages ?? []).filter(message => message.role === 'tool' || (Array.isArray(message.content) && message.content.some((block) => block?.type === 'tool_result')));
+                const alreadyHandedOff = realToolMessages.some(message => JSON.stringify(message).includes(HANDOFF));
+                // 首个 planner 响应拖过硬阈:首个工具派发必然越过 wall-clock 边界。
+                // handoff 发生后,后续 planner 请求一律回 text-only final,收敛回路。
+                const stall = !alreadyHandedOff && plannerIndex === 1 ? RELAY_STALL_MS : 0;
+                setTimeout(() => {
+                    if (!alreadyHandedOff && plannerIndex === 1) {
+                        res.end(toolResponse(String(plannerIndex)));
+                    }
+                    else {
+                        res.end(textResponse('deadline E2E final answer'));
+                    }
+                }, stall);
+            }
+            else {
+                // Session-title and other auxiliary calls deliberately carry no tool
+                // schemas. Keep them out of the planner-call budget and answer text.
+                res.end(textResponse('deadline E2E final answer', 'deadline-auxiliary'));
+            }
+        });
+    });
+    await new Promise(resolve => server.listen(0, '127.0.0.1', resolve));
+    return {
+        port: server.address().port,
+        bodies,
+        plannerBodies,
+        close: () => new Promise(resolve => server.close(() => resolve())),
+    };
+}
+function allToolMessages(bodies) {
+    // 新线型:工具结果以 user 消息携带 tool_result 块;旧线型是 role='tool'。
+    return bodies.flatMap(body => (body.messages ?? []).filter(message => message.role === 'tool' || (Array.isArray(message.content) && message.content.some((block) => block?.type === 'tool_result'))));
+}
+function toolNames(body) {
+    return (body.tools ?? []).map(tool => {
+        const fn = tool.function;
+        return String(fn?.name ?? tool.name ?? '');
+    }).filter(Boolean);
+}
+async function main() {
+    for (const file of ['dist/src/index.js', 'dist/capabilities/artifacts.js']) {
+        assert.ok(existsSync(join(ROOT, file)), `missing ${file}; run node scripts/build-dist.mjs first`);
+    }
+    const relay = await startRelay();
+    const dshHome = mkdtempSync(join(tmpdir(), 'gotry-deadline-dsh-home-'));
+    const childCwd = mkdtempSync(join(tmpdir(), 'gotry-deadline-cwd-'));
+    // 巡检状态纪律:handoff 工单只落隔离根(source 模式 dsh cwd 是创始人真实
+    // 数据目录 ts/dsh-runtime,绝不让 E2E 往那里写)。
+    const handoffRoot = mkdtempSync(join(tmpdir(), 'gotry-deadline-handoff-'));
+    let exitCode = null;
+    let output = '';
+    const configuredBin = process.env.GOTRY_DEADLINE_E2E_BIN;
+    assert.ok(configuredBin && existsSync(configuredBin), 'GOTRY_DEADLINE_E2E_BIN must point to the clean installed-package binary');
+    try {
+        const executableArgs = [DEEP_MESSAGE];
+        const childEnv = { ...process.env };
+        childEnv.DSH_HOME = dshHome;
+        // 可选宿主插件(calendar)挂载与否由 setup 状态面(~/.gotry/calendar.json)决定,
+        // 不吃 env(D-9 founder 纠偏)——E2E 隔离手段=HOME 指到临时目录(os.homedir()
+        // 尊重 $HOME),子进程看不到真实机器的 setup 态;dsh 配置面已由 DSH_HOME 单独隔离。
+        childEnv.HOME = dshHome;
+        childEnv.LLM_API_KEY = 'synthetic-e2e-key';
+        childEnv.LLM_BASE_URL = 'http://127.0.0.1:' + relay.port + '/v1';
+        childEnv.LLM_MODEL = 'synthetic-deadline-model';
+        childEnv.GOTRY_LOCALE = 'zh-CN';
+        childEnv.GOTRY_TURN_DEADLINE_SOFT_MS = String(SOFT_MS);
+        childEnv.GOTRY_TURN_DEADLINE_HARD_MS = String(HARD_MS);
+        childEnv.GOTRY_TURN_HANDOFF_ROOT = handoffRoot;
+        delete childEnv.GOTRY_LLM_MODEL;
+        delete childEnv.GOTRY_FORCE_TURN_DEADLINE;
+        const child = spawn(configuredBin, executableArgs, {
+            cwd: childCwd,
+            env: childEnv,
+            stdio: ['ignore', 'pipe', 'pipe'],
+        });
+        child.stdout.on('data', chunk => { output += chunk.toString(); });
+        child.stderr.on('data', chunk => { output += chunk.toString(); });
+        exitCode = await new Promise(resolve => {
+            const timer = setTimeout(() => { child.kill('SIGKILL'); resolve(null); }, TIMEOUT_MS);
+            child.once('exit', code => { clearTimeout(timer); resolve(code); });
+        });
+    }
+    finally {
+        // relay 故意不在此关闭:收集闭环段还要以同一 relay 做模型端点。
+        // Give the spawned binary a brief moment to release its temp files
+        // before we recursively remove the tmpdir.
+        await new Promise(resolve => setTimeout(resolve, 200));
+    }
+    assert.equal(exitCode, 0, `headless exit=${exitCode}; output tail=${output.slice(-1200)}`);
+    assert.ok(relay.plannerBodies.length >= 1, `expected at least one tool-enabled planner request, got ${relay.plannerBodies.length}; output tail=${output.slice(-12_000)}`);
+    // 首个工具派发越过硬阈:必须得到 handoff 语义的结构化拒绝,且没有任何
+    // 工具 body 真正执行。
+    const toolMessages = allToolMessages(relay.bodies);
+    const toolMessageDiagnostic = JSON.stringify(toolMessages).slice(-12_000);
+    const handoffResults = toolMessages.filter(message => JSON.stringify(message).includes(HANDOFF));
+    const successfulResults = toolMessages.filter(message => !JSON.stringify(message).includes(HANDOFF));
+    assert.ok(handoffResults.length >= 1, `expected at least one handoff result; messages=${toolMessageDiagnostic}`);
+    assert.equal(successfulResults.length, 0, `no tool body should run past the hard deadline; messages=${toolMessageDiagnostic}`);
+    assert.match(JSON.stringify(handoffResults[0]), /th-/, 'handoff result carries the ticket id');
+    // handoff 工单已落隔离根,字段完整(含用户原文与 ETA 承诺)。
+    const ticketDir = join(handoffRoot, 'gotry-state', 'turn-handoffs');
+    assert.ok(existsSync(ticketDir), `ticket dir missing: ${ticketDir}`);
+    const ticketFiles = readdirSync(ticketDir).filter(name => name.endsWith('.json'));
+    assert.equal(ticketFiles.length, 1, `expected exactly one handoff ticket, got ${ticketFiles.join(', ')}`);
+    const ticket = JSON.parse(readFileSync(join(ticketDir, ticketFiles[0]), 'utf8'));
+    assert.equal(ticket['schema'], HANDOFF_SCHEMA);
+    assert.equal(ticket['status'], 'open');
+    assert.ok(String(ticket['userMessage']).includes('婚礼'), 'ticket carries the verbatim deep-planning ask');
+    assert.ok(String(ticket['etaLabel']).includes('约 1 小时'));
+    // 回路收敛:最终请求不带任何继承工具 schema,二进制输出 final。
+    const finalRequest = relay.bodies.at(-1);
+    const finalNames = toolNames(finalRequest);
+    assert.ok(!finalNames.includes(TOOL), `final request inherited GoTry tool: ${finalNames.join(', ')}`);
+    assert.match(output, /deadline E2E final answer/);
+    // ---- 收集闭环:collector 以真打包二进制为 planner 回收这单工单。------
+    // 子会话(GOTRY_HANDOFF_CHILD=1)带满工具面打到 relay;relay 的全局
+    // plannerIndex 已 >1,立即回 text-only final → 子会话 converge 退出,
+    // collector 把最终答复结算为交付物。工单 open→settled,交付物可复访。
+    {
+        const ticketId = String(ticket['id']);
+        const collector = spawn(process.execPath, [
+            join(TS_DIR, 'node_modules', 'tsx', 'dist', 'cli.mjs'),
+            join(TS_DIR, 'scripts', 'turn-handoff-collect.ts'),
+            ticketId, handoffRoot,
+        ], {
+            cwd: TS_DIR,
+            env: {
+                ...process.env,
+                GOTRY_HANDOFF_PLANNER_BIN: configuredBin,
+                LLM_API_KEY: 'synthetic-e2e-key',
+                LLM_BASE_URL: 'http://127.0.0.1:' + relay.port + '/v1',
+                LLM_MODEL: 'synthetic-deadline-model',
+                GOTRY_LOCALE: 'zh-CN',
+            },
+            stdio: ['ignore', 'pipe', 'pipe'],
+        });
+        let collectorOut = '';
+        let collectorErr = '';
+        collector.stdout.on('data', chunk => { collectorOut += chunk.toString(); });
+        collector.stderr.on('data', chunk => { collectorErr += chunk.toString(); });
+        const collectorCode = await new Promise(resolve => {
+            const timer = setTimeout(() => { collector.kill('SIGKILL'); resolve(null); }, 90_000);
+            collector.once('exit', code => { clearTimeout(timer); resolve(code); });
+        });
+        assert.equal(collectorCode, 0, `collector exit=${collectorCode}; stderr tail=${collectorErr.slice(-800)}; stdout tail=${collectorOut.slice(-400)}`);
+        const terminal = JSON.parse(collectorOut.trim().split('\n').at(-1));
+        assert.equal(terminal['schema'], 'gotry_turn_handoff_terminal.v1');
+        assert.equal(terminal['status'], 'succeeded');
+        assert.ok(String(terminal['deliverable_path']).includes(`${ticketId}.deliverable.md`));
+        const settled = JSON.parse(readFileSync(join(ticketDir, `${ticketId}.json`), 'utf8'));
+        assert.equal(settled['status'], 'settled');
+        assert.ok(settled['settledAt'], 'settledAt recorded');
+        const deliverable = readFileSync(join(ticketDir, `${ticketId}.deliverable.md`), 'utf8');
+        assert.ok(deliverable.includes('deadline E2E final answer'), `deliverable carries planner final; got: ${deliverable.slice(-400)}`);
+    }
+    console.log(`agent turn deadline E2E: OK (exit=${exitCode}, plannerRequests=${relay.plannerBodies.length}, handoffResults=${handoffResults.length}, ticket=${ticketFiles[0]}, collected=settled)`);
+    await relay.close();
+    rmSync(dshHome, { recursive: true, force: true });
+    rmSync(childCwd, { recursive: true, force: true });
+    rmSync(handoffRoot, { recursive: true, force: true });
+}
+await main();