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

package/dist/scripts/typed-contract-canary.js

@@ -0,0 +1,176 @@
+/**
+ * D-30 普通模型 canary 运行器(issue #112 验收,tool-orchestration-design §4⑥)。
+ *
+ * 量什么:普通 LLM 面对投影后的工具 JSON Schema,能否**一次成型**产出合法参数——
+ * 这是 typed 参数契约迁移(D-30)要解决的核心问题(blob 时代强弱模型差距全部
+ * 暴露在「猜参数形状」上)。
+ *
+ * 怎么量:真实 LLM(env 三件套 LLM_API_KEY/LLM_BASE_URL/LLM_MODEL,或仓库 .env)
+ * + 工具投影 schema(defineTool 后的 parameters,与模型所见完全一致)+ tool_choice
+ * 强制单工具 → 模型出参 → `validateJsonSchemaValue`(与 dsh defineTool execute 前
+ * 同款校验)→ 违例为零 = 该例一次成型。
+ *
+ * 不执行工具:零检索副作用、零站点请求——canary 只量参数形态,不量业务结果。
+ *
+ * 达标线:工具调用一次成型率 ≥90%(10 例容 1 失);语义正确性(城市/日期/kind)
+ * 作为二级信号同报,不进门。
+ *
+ * 运行(在仓根或 ts/):LLM 三件套已配 → npx tsx ts/scripts/typed-contract-canary.ts
+ */
+import assert from 'node:assert/strict';
+import { readFileSync, existsSync, mkdtempSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { join, dirname } from 'node:path';
+import { fileURLToPath } from 'node:url';
+import { validateJsonSchemaValue } from '@deepseek-ai/dsh-tools';
+import { apply } from '../src/index.js';
+const ROOT = join(dirname(fileURLToPath(import.meta.url)), '..', '..');
+// --- LLM 三件套:env 优先,回退仓库 .env(只取值,永不回显) ---
+function loadEnvTriple() {
+    const env = process.env;
+    let key = env.LLM_API_KEY ?? '';
+    let base = env.LLM_BASE_URL ?? '';
+    let model = env.LLM_MODEL ?? '';
+    const envFile = join(ROOT, '.env');
+    if (existsSync(envFile)) {
+        for (const line of readFileSync(envFile, 'utf-8').split('\n')) {
+            const m = /^([A-Z_]+)=(.*)$/.exec(line.trim());
+            if (!m)
+                continue;
+            const [, k, v] = m;
+            if (k === 'LLM_API_KEY' && !key)
+                key = v.trim();
+            if (k === 'LLM_BASE_URL' && !base)
+                base = v.trim();
+            if (k === 'LLM_MODEL' && !model)
+                model = v.trim();
+        }
+    }
+    assert.ok(key && base && model, 'LLM_API_KEY/LLM_BASE_URL/LLM_MODEL 必须已配(env 或 .env)');
+    return { key, base, model };
+}
+// --- 捕获插件注册表(隔离 stateRoot:canary 不写任何共享状态;sessionAccess=off 免审批) ---
+const stateRoot = mkdtempSync(join(tmpdir(), 'gotry-canary-state-'));
+const registered = [];
+const ctx = {
+    tools: { register: (t) => registered.push(t) },
+    systemPrompt: { variable: () => () => '' },
+    on: () => () => { },
+};
+apply(ctx, { stateRoot, timeoutMs: 30_000, hbcliBin: 'hbcli-not-on-path', sessionAccess: 'off' });
+const byName = (n) => {
+    const t = registered.find(x => x.name === n);
+    assert.ok(t, `工具 ${n} 应已注册`);
+    return t;
+};
+const CASES = [
+    { label: '机·票(城市对+日期)', tool: 'gotry_flyai_search', prompt: '帮我查一下 10 月 1 日从上海飞丽江的机票' },
+    { label: '火车票', tool: 'gotry_flyai_search', prompt: '10 月 1 日上海到昆明的火车票帮我查一下' },
+    { label: '酒店(带档期)', tool: 'gotry_flyai_search', prompt: '查一下大理 10 月 1 日入住、10 月 3 日退房的酒店' },
+    { label: '酒店(摸底,无档期)', tool: 'gotry_flyai_search', prompt: '先看看大理有哪些酒店' },
+    { label: '天气预报', tool: 'gotry_weather_check', prompt: '大理这几天天气怎么样' },
+    { label: '酒店实时源(hbcli)', tool: 'gotry_hotel_search', prompt: '查一下大理的酒店,入住下周五,住三晚' },
+    { label: '会话面机票(登录态)', tool: 'gotry_session_search', prompt: '用我的携程账号查 10 月 1 日上海到丽江的机票' },
+    { label: '目的地候选库', tool: 'gotry_anything_search', prompt: '大连在你们的目的地库里吗?帮我查一下' },
+    { label: 'GitHub 搜索', tool: 'gotry_github_search', prompt: '在 GitHub 上搜一下 agent-reach 这个项目' },
+    { label: '网页读取', tool: 'gotry_web_search', prompt: '帮我读一下 https://example.com 这个页面的内容' },
+];
+async function callLlmMessages(cfg, toolName, messages) {
+    const tool = byName(toolName);
+    const url = `${cfg.base.replace(/\/+$/, '')}/chat/completions`;
+    const body = {
+        model: cfg.model,
+        messages,
+        tools: [{ type: 'function', function: { name: tool.name, description: tool.description, parameters: tool.parameters } }],
+        tool_choice: { type: 'function', function: { name: tool.name } },
+        temperature: 0,
+    };
+    const res = await fetch(url, {
+        method: 'POST',
+        headers: { 'content-type': 'application/json', authorization: `Bearer ${cfg.key}` },
+        body: JSON.stringify(body),
+        signal: AbortSignal.timeout(60_000),
+    });
+    const bodyText = await res.text();
+    assert.equal(res.status, 200, `LLM HTTP ${res.status}: ${bodyText.slice(0, 200)}`);
+    const j = JSON.parse(bodyText);
+    const msg = j.choices?.[0]?.message;
+    const tc = msg.tool_calls?.[0];
+    if (!tc?.function?.arguments) {
+        const text = String(msg.content ?? '');
+        assert.ok(/年份|日期|确认|2024|2025|2026/.test(text), `模型既未调工具也未按拒绝指引向用户确认: ${text.slice(0, 120)}`);
+        return { args: undefined, callId: 'text-reply', text, textReply: true };
+    }
+    return { args: JSON.parse(tc.function.arguments), callId: tc.id ?? 'canary-call', textReply: false };
+}
+async function main() {
+    const cfg = loadEnvTriple();
+    console.log(`typed-contract canary:model=${cfg.model} 已配(值不回显);用例 ${CASES.length} 条,达标线 ≥90% 一次成型\n`);
+    let pass = 0;
+    const failures = [];
+    for (const c of CASES) {
+        const tool = byName(c.tool);
+        process.stdout.write(`- ${c.label} → ${c.tool} … `);
+        try {
+            const { args } = await callLlmMessages(cfg, c.tool, [
+                { role: 'system', content: '你是 GoTry 旅行助手。需要查数据时必须调用提供的工具,参数从用户话里取,不要编造。' },
+                { role: 'user', content: c.prompt },
+            ]);
+            const violations = validateJsonSchemaValue(tool.parameters, args);
+            if (violations.length === 0) {
+                pass++;
+                console.log(`PASS 一次成型 ${JSON.stringify(args).slice(0, 120)}`);
+            }
+            else {
+                failures.push(`${c.label}: schema 违例 ${violations.join(';')}`);
+                console.log(`FAIL schema 违例: ${violations.join(';').slice(0, 160)}`);
+            }
+        }
+        catch (e) {
+            failures.push(`${c.label}: ${e.message.slice(0, 160)}`);
+            console.log(`FAIL ${e.message.slice(0, 140)}`);
+        }
+    }
+    // ---------------------------------------------------------------
+    // 验收②(issue #140):畸形/越界调用被宿主权拒绝后,模型按 violations 自纠。
+    // 确定性播种:预置一次带过去日期的被拒调用与结构化拒绝结果(tool 角色),
+    // 验证模型从拒绝中自纠出「schema 合规 + 日期非过去」的参数。只校验不执行(零副作用)。
+    // ---------------------------------------------------------------
+    {
+        const tool = byName('gotry_flyai_search');
+        const today = new Date().toISOString().slice(0, 10);
+        try {
+            const messages = [
+                { role: 'system', content: '你是 GoTry 旅行助手。需要查数据时必须调用提供的工具,参数从用户话里取,不要编造。' },
+                { role: 'user', content: '帮我查一下 10 月 1 日从上海飞丽江的机票' },
+                { role: 'assistant', content: null, tool_calls: [{ id: 'seed-reject', type: 'function', function: { name: 'gotry_flyai_search', arguments: JSON.stringify({ kind: 'flight', from: '上海', to: '丽江', date: '2024-10-01' }) } }] },
+                { role: 'tool', tool_call_id: 'seed-reject', content: JSON.stringify({ ok: false, verdict: 'error', summary: `未发起查询:日期 2024-10-01 已是过去(今天 ${today}),过去不存在在售机票。多为用户时间表达未带年份所致——向用户确认年份(或按未来最近的同月日修正)后再查。` }) },
+            ];
+            const reply = await callLlmMessages(cfg, 'gotry_flyai_search', messages);
+            const violations = reply.args ? validateJsonSchemaValue(tool.parameters, reply.args) : [];
+            if (reply.textReply) {
+                assert.match(reply.text ?? '', /年份|日期|确认/, '文字形态应向用户确认年份/日期(契约允许的自纠)');
+                console.log('自纠回路(验收②):模型按拒绝指引转为向用户确认年份(文字形态,契约允许)');
+            }
+            else {
+                assert.equal(violations.length, 0, `自纠出参 schema 违例: ${violations.join(';')}`);
+                assert.ok(reply.args && reply.args.date && reply.args.date >= today, `自纠后日期应为今天或未来,实际 ${JSON.stringify(reply.args).slice(0, 80)}`);
+                console.log(`自纠回路(验收②):模型按拒绝自纠出参 → PASS(${JSON.stringify(reply.args).slice(0, 100)})`);
+            }
+        }
+        catch (e) {
+            failures.push(`验收② 自纠回路异常: ${e.message.slice(0, 140)}`);
+            console.log(`FAIL 自纠回路异常(计入失败,不中断报告): ${e.message.slice(0, 140)}`);
+        }
+    }
+    const rate = Math.round((pass / CASES.length) * 100);
+    console.log(`\n一次成型率:${pass}/${CASES.length} = ${rate}%(达标线 90%)`);
+    if (failures.length)
+        console.log('失败明细:\n' + failures.map(f => `  - ${f}`).join('\n'));
+    if (rate < 90) {
+        console.error('TYPED CONTRACT CANARY FAIL(不达标不合入)');
+        process.exit(1);
+    }
+    console.log('TYPED CONTRACT CANARY OK');
+}
+await main();

package/dist/scripts/tz-resolver-sanity.js

@@ -0,0 +1,50 @@
+// Quick sanity check for tz-resolver. Run with: TZ=UTC npx tsx scripts/tz-resolver-sanity.ts
+import { resolveOffsetForLocalDate } from '../src/tz-resolver.js';
+const cases = [
+    // Asia/Shanghai has no DST throughout the year
+    ['Asia/Shanghai', '2026-07-18', '14:45', 480],
+    ['Asia/Shanghai', '2026-01-01', '09:00', 480],
+    ['Asia/Tokyo', '2026-07-04', '09:00', 540],
+    ['Asia/Hong_Kong', '2026-07-18', '14:45', 480],
+    ['Asia/Bangkok', '2026-08-09', '23:55', 420],
+    ['Asia/Singapore', '2026-08-09', '23:55', 480],
+    ['Asia/Dubai', '2026-07-18', '00:45', 240],
+    // America/New_York DST: EST = -5, EDT = -4. 2026 spring-forward = 2026-03-08, fall-back = 2026-11-01
+    ['America/New_York', '2026-03-07', '14:00', -300], // day before spring forward -> EST
+    ['America/New_York', '2026-03-09', '14:00', -240], // day after spring forward -> EDT
+    ['America/New_York', '2026-06-15', '09:00', -240], // summer -> EDT
+    ['America/New_York', '2026-12-15', '09:00', -300], // winter -> EST
+    ['America/Los_Angeles', '2026-06-15', '14:00', -420],
+    ['UTC', '2026-07-18', '12:00', 0],
+    ['Europe/London', '2026-06-15', '14:00', 60],
+    ['Europe/London', '2026-01-15', '14:00', 0],
+];
+let ok = 0, fail = 0;
+for (const [zone, ymd, hhmm, expected] of cases) {
+    const r = resolveOffsetForLocalDate(zone, ymd, hhmm);
+    const pass = r.ok && r.offsetMin === expected;
+    if (pass)
+        ok++;
+    else {
+        fail++;
+        console.log('FAIL', zone, ymd, hhmm, 'expected', expected, 'got', r.ok ? r.offsetMin : r);
+    }
+}
+const negatives = [
+    ['Foo/Bar', '2026-07-18', '14:45', 'unknown_zone'],
+    ['America/New_York', '2026-03-08', '02:30', 'gap_nonexistent'], // spring forward day
+    ['America/New_York', '2026-11-01', '01:30', 'overlap_ambiguous'], // fall back day
+];
+for (const [zone, ymd, hhmm, expected] of negatives) {
+    const r = resolveOffsetForLocalDate(zone, ymd, hhmm);
+    const pass = !r.ok && r.kind === expected;
+    if (pass)
+        ok++;
+    else {
+        fail++;
+        console.log('FAIL', zone, ymd, hhmm, 'expected', expected, 'got', r.ok ? `ok(${r.offsetMin})` : r.kind);
+    }
+}
+console.log(`\n${ok}/${ok + fail} sanity ok`);
+if (fail > 0)
+    process.exit(1);