@aipack-ai/eval 1.1.6

package/dist/index.d.ts

@@ -0,0 +1,24 @@
+/**
+ * @aipack-ai/eval —— aipack Eval 体系（M1：mock 模式 + 规则评分器）
+ *
+ * 快速上手：
+ *   import { loadCases, runEval, renderMarkdown } from '@aipack-ai/eval';
+ *   const { cases } = await loadCases();           // 默认 eval/cases
+ *   const report = await runEval(cases, { mode: 'mock' });
+ *   console.log(renderMarkdown(report));
+ *
+ * CLI：
+ *   pnpm --filter @aipack-ai/eval eval -- --suite tool-calling
+ */
+export type { CaseOrigin, MockToolCall, MockAssistantTurn, MockScript, AgentInput, ExpectedToolCall, ExpectedResult, RuleScorerType, ScorerType, ScorerConfig, EvalCase, ToolCallRecord, RunTrace, ScoreResult, CaseResult, SuiteSummary, EvalReport, RunConfig, BaselineFile, BaselineComparison, } from './src/core/types';
+export { createMockStreamFn } from './src/core/mock-stream';
+export { createMockTools, STANDARD_MOCK_TOOLS, } from './src/core/mock-tools';
+export type { StandardMockToolName } from './src/core/mock-tools';
+export { extractTrajectory, countTurns } from './src/core/trajectory';
+export { expectedToScorers, resolveScorers, scoreTrace, RULE_SCORER_TYPES, runRuleScorer, partialMatch, } from './src/core/scorer';
+export type { CaseScore } from './src/core/scorer';
+export { validateEvalCase } from './src/core/validate';
+export { loadCases, defaultCasesDir, packageRoot, } from './src/core/loader';
+export type { LoadCasesResult } from './src/core/loader';
+export { runEval } from './src/core/runner';
+export { renderMarkdown, writeReport, writeBaseline, readBaseline, reportToBaseline, compareBaseline, finalizeReport, } from './src/core/report';

package/dist/index.js

@@ -0,0 +1,26 @@
+/**
+ * @aipack-ai/eval —— aipack Eval 体系（M1：mock 模式 + 规则评分器）
+ *
+ * 快速上手：
+ *   import { loadCases, runEval, renderMarkdown } from '@aipack-ai/eval';
+ *   const { cases } = await loadCases();           // 默认 eval/cases
+ *   const report = await runEval(cases, { mode: 'mock' });
+ *   console.log(renderMarkdown(report));
+ *
+ * CLI：
+ *   pnpm --filter @aipack-ai/eval eval -- --suite tool-calling
+ */
+// ─── Mock 层（fixture replay）─────────────────────────────────────
+export { createMockStreamFn } from './src/core/mock-stream.js';
+export { createMockTools, STANDARD_MOCK_TOOLS, } from './src/core/mock-tools.js';
+// ─── 轨迹提取 ─────────────────────────────────────────────────────
+export { extractTrajectory, countTurns } from './src/core/trajectory.js';
+// ─── 评分器 ───────────────────────────────────────────────────────
+export { expectedToScorers, resolveScorers, scoreTrace, RULE_SCORER_TYPES, runRuleScorer, partialMatch, } from './src/core/scorer/index.js';
+// ─── 校验 / 加载 ──────────────────────────────────────────────────
+export { validateEvalCase } from './src/core/validate.js';
+export { loadCases, defaultCasesDir, packageRoot, } from './src/core/loader.js';
+// ─── Runner 与报告 ────────────────────────────────────────────────
+export { runEval } from './src/core/runner.js';
+export { renderMarkdown, writeReport, writeBaseline, readBaseline, reportToBaseline, compareBaseline, finalizeReport, } from './src/core/report.js';
+//# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AA0BH,kEAAkE;AAClE,OAAO,EAAE,kBAAkB,EAAE,MAAM,wBAAwB,CAAC;AAC5D,OAAO,EACL,eAAe,EACf,mBAAmB,GACpB,MAAM,uBAAuB,CAAC;AAG/B,iEAAiE;AACjE,OAAO,EAAE,iBAAiB,EAAE,UAAU,EAAE,MAAM,uBAAuB,CAAC;AAEtE,kEAAkE;AAClE,OAAO,EACL,iBAAiB,EACjB,cAAc,EACd,UAAU,EACV,iBAAiB,EACjB,aAAa,EACb,YAAY,GACb,MAAM,mBAAmB,CAAC;AAG3B,iEAAiE;AACjE,OAAO,EAAE,gBAAgB,EAAE,MAAM,qBAAqB,CAAC;AACvD,OAAO,EACL,SAAS,EACT,eAAe,EACf,WAAW,GACZ,MAAM,mBAAmB,CAAC;AAG3B,kEAAkE;AAClE,OAAO,EAAE,OAAO,EAAE,MAAM,mBAAmB,CAAC;AAC5C,OAAO,EACL,cAAc,EACd,WAAW,EACX,aAAa,EACb,YAAY,EACZ,gBAAgB,EAChB,eAAe,EACf,cAAc,GACf,MAAM,mBAAmB,CAAC"}

package/dist/src/cli.d.ts

@@ -0,0 +1,12 @@
+#!/usr/bin/env node
+/**
+ * packages/eval/src/cli.ts - aipack-eval 命令行入口
+ *
+ * 用法：
+ *   aipack-eval run [--suite <name>]... [--cases-dir <dir>] [--repeats <n>]
+ *                   [--report-dir <dir>] [--baseline <path>]
+ *                   [--update-baseline] [--threshold <0.02>]
+ *
+ * 退出码：0 = 全过且无回归；1 = 有失败或回归；2 = 用法错误。
+ */
+export {};

package/dist/src/cli.js

@@ -0,0 +1,149 @@
+#!/usr/bin/env node
+/**
+ * packages/eval/src/cli.ts - aipack-eval 命令行入口
+ *
+ * 用法：
+ *   aipack-eval run [--suite <name>]... [--cases-dir <dir>] [--repeats <n>]
+ *                   [--report-dir <dir>] [--baseline <path>]
+ *                   [--update-baseline] [--threshold <0.02>]
+ *
+ * 退出码：0 = 全过且无回归；1 = 有失败或回归；2 = 用法错误。
+ */
+import { loadCases } from './core/loader.js';
+import { runEval } from './core/runner.js';
+import { finalizeReport, renderMarkdown } from './core/report.js';
+const USAGE = `用法: aipack-eval run [options]
+
+选项:
+  --suite <name>         只跑指定套件（可多次）
+  --cases-dir <dir>      用例目录（缺省 <包根>/eval/cases）
+  --report-dir <dir>     报告输出目录（缺省 ./eval-results；'none' 不写盘）
+  --baseline <path>      baseline 文件（门禁对比）
+  --update-baseline      用本次结果更新 baseline
+  --threshold <x>        通过率回归阈值（缺省 0.02）
+  --repeats <n>          每用例重复次数（缺省 1）
+  -h, --help             显示帮助`;
+function parseArgs(argv) {
+    const args = {
+        suites: [],
+        updateBaseline: false,
+        threshold: 0.02,
+    };
+    let i = 0;
+    while (i < argv.length) {
+        const a = argv[i];
+        const next = () => {
+            const v = argv[++i];
+            if (v === undefined)
+                throw new Error(`选项 ${a} 缺少参数值`);
+            return v;
+        };
+        try {
+            switch (a) {
+                case 'run':
+                    args.command = 'run';
+                    break;
+                case '--suite':
+                    args.suites.push(next());
+                    break;
+                case '--cases-dir':
+                    args.casesDir = next();
+                    break;
+                case '--report-dir':
+                    args.reportDir = next();
+                    break;
+                case '--baseline':
+                    args.baselinePath = next();
+                    break;
+                case '--update-baseline':
+                    args.updateBaseline = true;
+                    break;
+                case '--threshold': {
+                    const v = Number(next());
+                    if (!Number.isFinite(v) || v < 0 || v > 1)
+                        throw new Error('阈值须为 0~1');
+                    args.threshold = v;
+                    break;
+                }
+                case '--repeats': {
+                    const v = Number(next());
+                    if (!Number.isInteger(v) || v < 1)
+                        throw new Error('repeats 须为正整数');
+                    args.repeats = v;
+                    break;
+                }
+                case '-h':
+                case '--help':
+                    return { error: 'HELP' };
+                default:
+                    return { error: `未知参数: ${a}\n${USAGE}` };
+            }
+        }
+        catch (e) {
+            return { error: `${e.message}\n${USAGE}` };
+        }
+        i += 1;
+    }
+    if (args.command !== 'run')
+        return { error: `缺少子命令 'run'\n${USAGE}` };
+    return args;
+}
+async function main() {
+    const parsed = parseArgs(process.argv.slice(2));
+    if ('error' in parsed) {
+        if (parsed.error === 'HELP') {
+            console.log(USAGE);
+            return 0;
+        }
+        console.error(parsed.error);
+        return 2;
+    }
+    const { cases, errors } = await loadCases(parsed.casesDir);
+    if (errors.length > 0) {
+        console.error('用例加载错误:');
+        for (const e of errors)
+            console.error(`  - ${e}`);
+        return 1;
+    }
+    if (cases.length === 0) {
+        console.error('没有可运行的用例');
+        return 1;
+    }
+    const config = {
+        mode: 'mock',
+        suites: parsed.suites,
+        repeats: parsed.repeats,
+        casesDir: parsed.casesDir,
+        reportDir: parsed.reportDir === 'none' ? undefined : parsed.reportDir,
+        baselinePath: parsed.baselinePath,
+        updateBaseline: parsed.updateBaseline,
+        regressionThreshold: parsed.threshold,
+    };
+    console.error(`加载 ${cases.length} 个用例${parsed.suites.length ? `（套件: ${parsed.suites.join(', ')}）` : ''}，mode=mock`);
+    const report = await runEval(cases, config);
+    const finalized = await finalizeReport(report, config);
+    console.log(renderMarkdown(report));
+    if (finalized.jsonPath)
+        console.error(`\n报告已写入: ${finalized.jsonPath}`);
+    if (config.updateBaseline && finalized.baseline === undefined) {
+        console.error(`baseline 已更新: ${config.baselinePath}`);
+    }
+    let exitCode = report.totals.passed === report.totals.cases ? 0 : 1;
+    if (finalized.comparison && !finalized.comparison.ok) {
+        exitCode = 1;
+        const { overallDelta, regressions } = finalized.comparison;
+        console.error(`\n⚠️  baseline 回归（整体 ${overallDelta >= 0 ? '+' : ''}${(overallDelta * 100).toFixed(1)}%）:`);
+        for (const r of regressions) {
+            console.error(`  - ${r.suite}: ${(r.baseline * 100).toFixed(1)}% → ${(r.current * 100).toFixed(1)}%（${(r.delta * 100).toFixed(1)}%）`);
+        }
+    }
+    else if (finalized.comparison) {
+        console.error(`\nbaseline 门禁通过（${(finalized.comparison.overallDelta * 100).toFixed(1)}%）`);
+    }
+    return exitCode;
+}
+main().then((code) => process.exit(code), (e) => {
+    console.error(e);
+    process.exit(1);
+});
+//# sourceMappingURL=cli.js.map

package/dist/src/cli.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"cli.js","sourceRoot":"","sources":["../../src/cli.ts"],"names":[],"mappings":";AACA;;;;;;;;;GASG;AAEH,OAAO,EAAE,SAAS,EAAE,MAAM,eAAe,CAAC;AAC1C,OAAO,EAAE,OAAO,EAAE,MAAM,eAAe,CAAC;AACxC,OAAO,EAAE,cAAc,EAAE,cAAc,EAAE,MAAM,eAAe,CAAC;AAc/D,MAAM,KAAK,GAAG;;;;;;;;;;8BAUgB,CAAC;AAE/B,SAAS,SAAS,CAAC,IAAc;IAC/B,MAAM,IAAI,GAAY;QACpB,MAAM,EAAE,EAAE;QACV,cAAc,EAAE,KAAK;QACrB,SAAS,EAAE,IAAI;KAChB,CAAC;IACF,IAAI,CAAC,GAAG,CAAC,CAAC;IACV,OAAO,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC;QACvB,MAAM,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,MAAM,IAAI,GAAG,GAAW,EAAE;YACxB,MAAM,CAAC,GAAG,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC;YACpB,IAAI,CAAC,KAAK,SAAS;gBAAE,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;YACtD,OAAO,CAAC,CAAC;QACX,CAAC,CAAC;QACF,IAAI,CAAC;YACH,QAAQ,CAAC,EAAE,CAAC;gBACV,KAAK,KAAK;oBACR,IAAI,CAAC,OAAO,GAAG,KAAK,CAAC;oBACrB,MAAM;gBACR,KAAK,SAAS;oBACZ,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC;oBACzB,MAAM;gBACR,KAAK,aAAa;oBAChB,IAAI,CAAC,QAAQ,GAAG,IAAI,EAAE,CAAC;oBACvB,MAAM;gBACR,KAAK,cAAc;oBACjB,IAAI,CAAC,SAAS,GAAG,IAAI,EAAE,CAAC;oBACxB,MAAM;gBACR,KAAK,YAAY;oBACf,IAAI,CAAC,YAAY,GAAG,IAAI,EAAE,CAAC;oBAC3B,MAAM;gBACR,KAAK,mBAAmB;oBACtB,IAAI,CAAC,cAAc,GAAG,IAAI,CAAC;oBAC3B,MAAM;gBACR,KAAK,aAAa,CAAC,CAAC,CAAC;oBACnB,MAAM,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC;oBACzB,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC;wBAAE,MAAM,IAAI,KAAK,CAAC,UAAU,CAAC,CAAC;oBACvE,IAAI,CAAC,SAAS,GAAG,CAAC,CAAC;oBACnB,MAAM;gBACR,CAAC;gBACD,KAAK,WAAW,CAAC,CAAC,CAAC;oBACjB,MAAM,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC;oBACzB,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC;wBAAE,MAAM,IAAI,KAAK,CAAC,eAAe,CAAC,CAAC;oBACpE,IAAI,CAAC,OAAO,GAAG,CAAC,CAAC;oBACjB,MAAM;gBACR,CAAC;gBACD,KAAK,IAAI,CAAC;gBACV,KAAK,QAAQ;oBACX,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC;gBAC3B;oBACE,OAAO,EAAE,KAAK,EAAE,SAAS,CAAC,KAAK,KAAK,EAAE,EAAE,CAAC;YAC7C,CAAC;QACH,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,OAAO,EAAE,KAAK,EAAE,GAAI,CAAW,CAAC,OAAO,KAAK,KAAK,EAAE,EAAE,CAAC;QACxD,CAAC;QACD,CAAC,IAAI,CAAC,CAAC;IACT,CAAC;IACD,IAAI,IAAI,CAAC,OAAO,KAAK,KAAK;QAAE,OAAO,EAAE,KAAK,EAAE,gBAAgB,KAAK,EAAE,EAAE,CAAC;IACtE,OAAO,IAAI,CAAC;AACd,CAAC;AAED,KAAK,UAAU,IAAI;IACjB,MAAM,MAAM,GAAG,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;IAChD,IAAI,OAAO,IAAI,MAAM,EAAE,CAAC;QACtB,IAAI,MAAM,CAAC,KAAK,KAAK,MAAM,EAAE,CAAC;YAC5B,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACnB,OAAO,CAAC,CAAC;QACX,CAAC;QACD,OAAO,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QAC5B,OAAO,CAAC,CAAC;IACX,CAAC;IAED,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,GAAG,MAAM,SAAS,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;IAC3D,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACtB,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;QACzB,KAAK,MAAM,CAAC,IAAI,MAAM;YAAE,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;QAClD,OAAO,CAAC,CAAC;IACX,CAAC;IACD,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvB,OAAO,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC;QAC1B,OAAO,CAAC,CAAC;IACX,CAAC;IAED,MAAM,MAAM,GAAc;QACxB,IAAI,EAAE,MAAM;QACZ,MAAM,EAAE,MAAM,CAAC,MAAM;QACrB,OAAO,EAAE,MAAM,CAAC,OAAO;QACvB,QAAQ,EAAE,MAAM,CAAC,QAAQ;QACzB,SAAS,EAAE,MAAM,CAAC,SAAS,KAAK,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,SAAS;QACrE,YAAY,EAAE,MAAM,CAAC,YAAY;QACjC,cAAc,EAAE,MAAM,CAAC,cAAc;QACrC,mBAAmB,EAAE,MAAM,CAAC,SAAS;KACtC,CAAC;IAEF,OAAO,CAAC,KAAK,CACX,MAAM,KAAK,CAAC,MAAM,OAAO,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,QAAQ,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,YAAY,CACrG,CAAC;IACF,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;IAC5C,MAAM,SAAS,GAAG,MAAM,cAAc,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAEvD,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC;IAEpC,IAAI,SAAS,CAAC,QAAQ;QAAE,OAAO,CAAC,KAAK,CAAC,YAAY,SAAS,CAAC,QAAQ,EAAE,CAAC,CAAC;IACxE,IAAI,MAAM,CAAC,cAAc,IAAI,SAAS,CAAC,QAAQ,KAAK,SAAS,EAAE,CAAC;QAC9D,OAAO,CAAC,KAAK,CAAC,iBAAiB,MAAM,CAAC,YAAY,EAAE,CAAC,CAAC;IACxD,CAAC;IAED,IAAI,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,MAAM,KAAK,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAEpE,IAAI,SAAS,CAAC,UAAU,IAAI,CAAC,SAAS,CAAC,UAAU,CAAC,EAAE,EAAE,CAAC;QACrD,QAAQ,GAAG,CAAC,CAAC;QACb,MAAM,EAAE,YAAY,EAAE,WAAW,EAAE,GAAG,SAAS,CAAC,UAAU,CAAC;QAC3D,OAAO,CAAC,KAAK,CAAC,wBAAwB,YAAY,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,YAAY,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC;QAC3G,KAAK,MAAM,CAAC,IAAI,WAAW,EAAE,CAAC;YAC5B,OAAO,CAAC,KAAK,CACX,OAAO,CAAC,CAAC,KAAK,KAAK,CAAC,CAAC,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CACvH,CAAC;QACJ,CAAC;IACH,CAAC;SAAM,IAAI,SAAS,CAAC,UAAU,EAAE,CAAC;QAChC,OAAO,CAAC,KAAK,CAAC,mBAAmB,CAAC,SAAS,CAAC,UAAU,CAAC,YAAY,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;IAC7F,CAAC;IAED,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,IAAI,EAAE,CAAC,IAAI,CACT,CAAC,IAAI,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAC5B,CAAC,CAAC,EAAE,EAAE;IACJ,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;IACjB,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAClB,CAAC,CACF,CAAC"}

package/dist/src/core/loader.d.ts

@@ -0,0 +1,21 @@
+/**
+ * packages/eval/src/core/loader.ts - 用例加载器
+ *
+ * 递归扫描用例目录（缺省 <包根>/eval/cases），解析 JSON 为 EvalCase，
+ * 加载即校验（fail fast），重复 id 报错。
+ */
+import type { EvalCase } from './types';
+/** 包根目录：从当前文件向上查找最近的 package.json（源码与 dist 产物通用） */
+export declare function packageRoot(): string;
+export declare function defaultCasesDir(): string;
+export interface LoadCasesResult {
+    cases: EvalCase[];
+    /** 相对用例目录的文件路径（诊断用） */
+    sources: Map<string, string>;
+    errors: string[];
+}
+/**
+ * 加载用例目录。校验失败的文件进入 errors（不中断整体加载），
+ * 但重复 id 一律报错（防套件意外覆盖）。
+ */
+export declare function loadCases(casesDir?: string): Promise<LoadCasesResult>;

package/dist/src/core/loader.js

@@ -0,0 +1,91 @@
+/**
+ * packages/eval/src/core/loader.ts - 用例加载器
+ *
+ * 递归扫描用例目录（缺省 <包根>/eval/cases），解析 JSON 为 EvalCase，
+ * 加载即校验（fail fast），重复 id 报错。
+ */
+import { readdir, readFile, stat } from 'node:fs/promises';
+import { existsSync } from 'node:fs';
+import { fileURLToPath } from 'node:url';
+import { dirname, join, resolve } from 'node:path';
+import { validateEvalCase } from './validate.js';
+/** 包根目录：从当前文件向上查找最近的 package.json（源码与 dist 产物通用） */
+export function packageRoot() {
+    let dir = dirname(fileURLToPath(import.meta.url));
+    for (let i = 0; i < 10; i += 1) {
+        if (existsSync(join(dir, 'package.json')))
+            return dir;
+        const parent = dirname(dir);
+        if (parent === dir)
+            break;
+        dir = parent;
+    }
+    // 兜底：回到源码相对位置（src/core → 上两级）
+    return resolve(dirname(fileURLToPath(import.meta.url)), '..', '..');
+}
+export function defaultCasesDir() {
+    return join(packageRoot(), 'eval', 'cases');
+}
+async function* walk(dir) {
+    const entries = await readdir(dir, { withFileTypes: true });
+    for (const e of entries) {
+        const full = join(dir, e.name);
+        if (e.isDirectory())
+            yield* walk(full);
+        else if (e.isFile() && (e.name.endsWith('.json') || e.name.endsWith('.jsonc'))) {
+            yield full;
+        }
+    }
+}
+/**
+ * 加载用例目录。校验失败的文件进入 errors（不中断整体加载），
+ * 但重复 id 一律报错（防套件意外覆盖）。
+ */
+export async function loadCases(casesDir) {
+    const dir = resolve(casesDir ?? defaultCasesDir());
+    const cases = [];
+    const sources = new Map();
+    const errors = [];
+    const seenIds = new Map();
+    let dirExists = true;
+    try {
+        await stat(dir);
+    }
+    catch {
+        dirExists = false;
+    }
+    if (!dirExists) {
+        return { cases, sources, errors: [`用例目录不存在: ${dir}`] };
+    }
+    for await (const file of walk(dir)) {
+        const rel = file.slice(dir.length + 1);
+        let parsed;
+        try {
+            parsed = JSON.parse(await readFile(file, 'utf-8'));
+        }
+        catch (e) {
+            errors.push(`${rel}: JSON 解析失败 - ${e.message}`);
+            continue;
+        }
+        // 支持单文件单用例或数组
+        const list = Array.isArray(parsed) ? parsed : [parsed];
+        for (const item of list) {
+            const c = item;
+            const caseErrors = validateEvalCase(c);
+            if (caseErrors.length > 0) {
+                errors.push(`${rel}: ${caseErrors.join('; ')}`);
+                continue;
+            }
+            const dup = seenIds.get(c.id);
+            if (dup) {
+                errors.push(`${rel}: id '${c.id}' 与 ${dup} 重复`);
+                continue;
+            }
+            seenIds.set(c.id, rel);
+            cases.push(c);
+            sources.set(c.id, rel);
+        }
+    }
+    return { cases, sources, errors };
+}
+//# sourceMappingURL=loader.js.map

package/dist/src/core/loader.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"loader.js","sourceRoot":"","sources":["../../../src/core/loader.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,kBAAkB,CAAC;AAC3D,OAAO,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AACrC,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AACzC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAEnD,OAAO,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAE9C,oDAAoD;AACpD,MAAM,UAAU,WAAW;IACzB,IAAI,GAAG,GAAG,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;IAClD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC;QAC/B,IAAI,UAAU,CAAC,IAAI,CAAC,GAAG,EAAE,cAAc,CAAC,CAAC;YAAE,OAAO,GAAG,CAAC;QACtD,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC;QAC5B,IAAI,MAAM,KAAK,GAAG;YAAE,MAAM;QAC1B,GAAG,GAAG,MAAM,CAAC;IACf,CAAC;IACD,8BAA8B;IAC9B,OAAO,OAAO,CAAC,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC;AACtE,CAAC;AAED,MAAM,UAAU,eAAe;IAC7B,OAAO,IAAI,CAAC,WAAW,EAAE,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC;AAC9C,CAAC;AAED,KAAK,SAAS,CAAC,CAAC,IAAI,CAAC,GAAW;IAC9B,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,GAAG,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;IAC5D,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;QACxB,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC;QAC/B,IAAI,CAAC,CAAC,WAAW,EAAE;YAAE,KAAK,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;aAClC,IAAI,CAAC,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC,EAAE,CAAC;YAC/E,MAAM,IAAI,CAAC;QACb,CAAC;IACH,CAAC;AACH,CAAC;AASD;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,SAAS,CAAC,QAAiB;IAC/C,MAAM,GAAG,GAAG,OAAO,CAAC,QAAQ,IAAI,eAAe,EAAE,CAAC,CAAC;IACnD,MAAM,KAAK,GAAe,EAAE,CAAC;IAC7B,MAAM,OAAO,GAAG,IAAI,GAAG,EAAkB,CAAC;IAC1C,MAAM,MAAM,GAAa,EAAE,CAAC;IAC5B,MAAM,OAAO,GAAG,IAAI,GAAG,EAAkB,CAAC;IAE1C,IAAI,SAAS,GAAG,IAAI,CAAC;IACrB,IAAI,CAAC;QACH,MAAM,IAAI,CAAC,GAAG,CAAC,CAAC;IAClB,CAAC;IAAC,MAAM,CAAC;QACP,SAAS,GAAG,KAAK,CAAC;IACpB,CAAC;IACD,IAAI,CAAC,SAAS,EAAE,CAAC;QACf,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,YAAY,GAAG,EAAE,CAAC,EAAE,CAAC;IACzD,CAAC;IAED,IAAI,KAAK,EAAE,MAAM,IAAI,IAAI,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC;QACnC,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QACvC,IAAI,MAAe,CAAC;QACpB,IAAI,CAAC;YACH,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,CAAC;QACrD,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,MAAM,CAAC,IAAI,CAAC,GAAG,GAAG,iBAAkB,CAAW,CAAC,OAAO,EAAE,CAAC,CAAC;YAC3D,SAAS;QACX,CAAC;QAED,cAAc;QACd,MAAM,IAAI,GAAc,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;QAClE,KAAK,MAAM,IAAI,IAAI,IAAI,EAAE,CAAC;YACxB,MAAM,CAAC,GAAG,IAAgB,CAAC;YAC3B,MAAM,UAAU,GAAG,gBAAgB,CAAC,CAAC,CAAC,CAAC;YACvC,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBAC1B,MAAM,CAAC,IAAI,CAAC,GAAG,GAAG,KAAK,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;gBAChD,SAAS;YACX,CAAC;YACD,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;YAC9B,IAAI,GAAG,EAAE,CAAC;gBACR,MAAM,CAAC,IAAI,CAAC,GAAG,GAAG,SAAS,CAAC,CAAC,EAAE,OAAO,GAAG,KAAK,CAAC,CAAC;gBAChD,SAAS;YACX,CAAC;YACD,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,GAAG,CAAC,CAAC;YACvB,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YACd,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,GAAG,CAAC,CAAC;QACzB,CAAC;IACH,CAAC;IAED,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC;AACpC,CAAC"}

package/dist/src/core/mock-stream.d.ts

@@ -0,0 +1,16 @@
+/**
+ * packages/eval/src/core/mock-stream.ts - Mock StreamFn（fixture replay）
+ *
+ * 通过 RuntimeOptions.streamFn 注入脚本化的 LLM 响应，零改动 Runtime。
+ * 每次 streamFn 调用消费 MockScript.turns 中的一条；轮次耗尽后按
+ * fallbackText / infiniteTool 兜底。
+ */
+import type { StreamFn } from '@aipack-ai/agent';
+import type { MockScript } from './types';
+/**
+ * 创建脚本式 StreamFn：第 n 次调用返回第 n 轮脚本。
+ *
+ * - 轮次耗尽且未配置 infiniteTool：返回 stopReason 'stop' 的兜底文本轮
+ * - 配置 infiniteTool：持续返回该工具调用（模拟死循环，配合 maxTurns 断言）
+ */
+export declare function createMockStreamFn(script: MockScript): StreamFn;

package/dist/src/core/mock-stream.js

@@ -0,0 +1,71 @@
+/**
+ * packages/eval/src/core/mock-stream.ts - Mock StreamFn（fixture replay）
+ *
+ * 通过 RuntimeOptions.streamFn 注入脚本化的 LLM 响应，零改动 Runtime。
+ * 每次 streamFn 调用消费 MockScript.turns 中的一条；轮次耗尽后按
+ * fallbackText / infiniteTool 兜底。
+ */
+const DEFAULT_USAGE = { input: 10, output: 5, total: 15 };
+/** 把脚本轮次转成框架层 AssistantMessage */
+function buildTurnMessage(turn, idBase) {
+    const content = [];
+    if (turn.text)
+        content.push({ type: 'text', text: turn.text });
+    for (const [i, tc] of (turn.toolCalls ?? []).entries()) {
+        content.push({
+            type: 'toolCall',
+            id: tc.id ?? `mock_tc_${idBase}_${i + 1}`,
+            name: tc.name,
+            arguments: tc.args ?? {},
+        });
+    }
+    const hasToolCalls = (turn.toolCalls ?? []).length > 0;
+    return {
+        role: 'assistant',
+        content,
+        stopReason: turn.stopReason ?? (hasToolCalls ? 'toolUse' : 'stop'),
+        usage: {
+            input: turn.usage?.input ?? DEFAULT_USAGE.input,
+            output: turn.usage?.output ?? DEFAULT_USAGE.output,
+            total: turn.usage?.total ?? DEFAULT_USAGE.total,
+        },
+        timestamp: Date.now(),
+    };
+}
+/**
+ * 创建脚本式 StreamFn：第 n 次调用返回第 n 轮脚本。
+ *
+ * - 轮次耗尽且未配置 infiniteTool：返回 stopReason 'stop' 的兜底文本轮
+ * - 配置 infiniteTool：持续返回该工具调用（模拟死循环，配合 maxTurns 断言）
+ */
+export function createMockStreamFn(script) {
+    let callIndex = 0;
+    return async function* () {
+        // 每次调用让出一次事件循环（模拟网络延迟）：
+        // 保证墙钟超时 / 并发调度有执行机会，行为更接近真实 provider
+        await new Promise((resolve) => setImmediate(resolve));
+        const turn = script.turns[callIndex];
+        if (turn) {
+            callIndex += 1;
+            yield { type: 'done', message: buildTurnMessage(turn, callIndex) };
+            return;
+        }
+        if (script.infiniteTool) {
+            callIndex += 1;
+            yield {
+                type: 'done',
+                message: buildTurnMessage({
+                    toolCalls: [
+                        { name: script.infiniteTool, args: script.infiniteToolArgs ?? {} },
+                    ],
+                }, callIndex),
+            };
+            return;
+        }
+        yield {
+            type: 'done',
+            message: buildTurnMessage({ text: script.fallbackText ?? 'done' }, callIndex + 1),
+        };
+    };
+}
+//# sourceMappingURL=mock-stream.js.map

package/dist/src/core/mock-stream.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"mock-stream.js","sourceRoot":"","sources":["../../../src/core/mock-stream.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AASH,MAAM,aAAa,GAAG,EAAE,KAAK,EAAE,EAAE,EAAE,MAAM,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,CAAC;AAE1D,kCAAkC;AAClC,SAAS,gBAAgB,CACvB,IAKC,EACD,MAAc;IAEd,MAAM,OAAO,GAAmB,EAAE,CAAC;IAEnC,IAAI,IAAI,CAAC,IAAI;QAAE,OAAO,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC;IAC/D,KAAK,MAAM,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,IAAI,EAAE,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC;QACvD,OAAO,CAAC,IAAI,CAAC;YACX,IAAI,EAAE,UAAU;YAChB,EAAE,EAAE,EAAE,CAAC,EAAE,IAAI,WAAW,MAAM,IAAI,CAAC,GAAG,CAAC,EAAE;YACzC,IAAI,EAAE,EAAE,CAAC,IAAI;YACb,SAAS,EAAE,EAAE,CAAC,IAAI,IAAI,EAAE;SACzB,CAAC,CAAC;IACL,CAAC;IAED,MAAM,YAAY,GAAG,CAAC,IAAI,CAAC,SAAS,IAAI,EAAE,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC;IACvD,OAAO;QACL,IAAI,EAAE,WAAW;QACjB,OAAO;QACP,UAAU,EAAE,IAAI,CAAC,UAAU,IAAI,CAAC,YAAY,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC;QAClE,KAAK,EAAE;YACL,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,KAAK,IAAI,aAAa,CAAC,KAAK;YAC/C,MAAM,EAAE,IAAI,CAAC,KAAK,EAAE,MAAM,IAAI,aAAa,CAAC,MAAM;YAClD,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,KAAK,IAAI,aAAa,CAAC,KAAK;SAChD;QACD,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;KACtB,CAAC;AACJ,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,kBAAkB,CAAC,MAAkB;IACnD,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,OAAO,KAAK,SAAS,CAAC;QACpB,wBAAwB;QACxB,sCAAsC;QACtC,MAAM,IAAI,OAAO,CAAO,CAAC,OAAO,EAAE,EAAE,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC;QAC5D,MAAM,IAAI,GAAG,MAAM,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;QACrC,IAAI,IAAI,EAAE,CAAC;YACT,SAAS,IAAI,CAAC,CAAC;YACf,MAAM,EAAE,IAAI,EAAE,MAAe,EAAE,OAAO,EAAE,gBAAgB,CAAC,IAAI,EAAE,SAAS,CAAC,EAAE,CAAC;YAC5E,OAAO;QACT,CAAC;QACD,IAAI,MAAM,CAAC,YAAY,EAAE,CAAC;YACxB,SAAS,IAAI,CAAC,CAAC;YACf,MAAM;gBACJ,IAAI,EAAE,MAAe;gBACrB,OAAO,EAAE,gBAAgB,CACvB;oBACE,SAAS,EAAE;wBACT,EAAE,IAAI,EAAE,MAAM,CAAC,YAAY,EAAE,IAAI,EAAE,MAAM,CAAC,gBAAgB,IAAI,EAAE,EAAE;qBACnE;iBACF,EACD,SAAS,CACV;aACF,CAAC;YACF,OAAO;QACT,CAAC;QACD,MAAM;YACJ,IAAI,EAAE,MAAe;YACrB,OAAO,EAAE,gBAAgB,CAAC,EAAE,IAAI,EAAE,MAAM,CAAC,YAAY,IAAI,MAAM,EAAE,EAAE,SAAS,GAAG,CAAC,CAAC;SAClF,CAAC;IACJ,CAAC,CAAC;AACJ,CAAC"}

package/dist/src/core/mock-tools.d.ts

@@ -0,0 +1,24 @@
+/**
+ * packages/eval/src/core/mock-tools.ts - 标准 mock 工具集
+ *
+ * 提供一组确定性的内存工具，供 golden 用例构造真实的工作流：
+ *   echo      - 回显参数
+ *   readFile  - 读 case 预置的内存文件系统
+ *   writeFile - 写内存文件系统
+ *   listDir   - 列出前缀匹配的文件
+ *   search    - 跨文件全文检索
+ *   fail      - 恒定失败（测错误恢复轨迹）
+ *
+ * 工具按 case 隔离：每次 createMockTools 都返回独立 fs 副本。
+ */
+import type { Tool } from '@aipack-ai/agent';
+/** 标准工具名列表（用例 input.tools 缺省时的注册范围） */
+export declare const STANDARD_MOCK_TOOLS: readonly ["echo", "readFile", "writeFile", "listDir", "search", "fail"];
+export type StandardMockToolName = (typeof STANDARD_MOCK_TOOLS)[number];
+/**
+ * 创建一组隔离的 mock 工具。
+ *
+ * @param fs 预置文件系统（会被深拷贝，case 间互不影响）
+ * @param names 需要的工具名列表；缺省全部标准工具
+ */
+export declare function createMockTools(fs?: Record<string, string>, names?: string[]): Tool[];

package/dist/src/core/mock-tools.js

@@ -0,0 +1,161 @@
+/**
+ * packages/eval/src/core/mock-tools.ts - 标准 mock 工具集
+ *
+ * 提供一组确定性的内存工具，供 golden 用例构造真实的工作流：
+ *   echo      - 回显参数
+ *   readFile  - 读 case 预置的内存文件系统
+ *   writeFile - 写内存文件系统
+ *   listDir   - 列出前缀匹配的文件
+ *   search    - 跨文件全文检索
+ *   fail      - 恒定失败（测错误恢复轨迹）
+ *
+ * 工具按 case 隔离：每次 createMockTools 都返回独立 fs 副本。
+ */
+const STRING_SCHEMA = {
+    type: 'object',
+    properties: {
+        message: { type: 'string', description: '要回显的消息' },
+    },
+    required: ['message'],
+};
+const READ_FILE_SCHEMA = {
+    type: 'object',
+    properties: {
+        path: { type: 'string', description: '文件路径' },
+    },
+    required: ['path'],
+};
+const WRITE_FILE_SCHEMA = {
+    type: 'object',
+    properties: {
+        path: { type: 'string', description: '文件路径' },
+        content: { type: 'string', description: '写入内容' },
+    },
+    required: ['path', 'content'],
+};
+const LIST_DIR_SCHEMA = {
+    type: 'object',
+    properties: {
+        path: { type: 'string', description: '目录前缀' },
+    },
+};
+const SEARCH_SCHEMA = {
+    type: 'object',
+    properties: {
+        query: { type: 'string', description: '检索关键词' },
+    },
+    required: ['query'],
+};
+const FAIL_SCHEMA = {
+    type: 'object',
+    properties: {},
+};
+function textResult(text, details = null) {
+    const content = [{ type: 'text', text }];
+    return { content, details };
+}
+function errorResult(message) {
+    const content = [{ type: 'text', text: `error: ${message}` }];
+    return { content, details: { error: message } };
+}
+/** 标准工具名列表（用例 input.tools 缺省时的注册范围） */
+export const STANDARD_MOCK_TOOLS = [
+    'echo',
+    'readFile',
+    'writeFile',
+    'listDir',
+    'search',
+    'fail',
+];
+/**
+ * 创建一组隔离的 mock 工具。
+ *
+ * @param fs 预置文件系统（会被深拷贝，case 间互不影响）
+ * @param names 需要的工具名列表；缺省全部标准工具
+ */
+export function createMockTools(fs = {}, names) {
+    const files = new Map(Object.entries(structuredClone(fs)));
+    const all = {
+        echo: {
+            name: 'echo',
+            description: '回显输入消息（测试用）',
+            parameters: STRING_SCHEMA,
+            execute: async (_id, args) => {
+                const message = args?.message;
+                return textResult(String(message ?? ''));
+            },
+        },
+        readFile: {
+            name: 'readFile',
+            description: '读取内存文件系统中的文件内容',
+            parameters: READ_FILE_SCHEMA,
+            execute: async (_id, args) => {
+                const { path } = (args ?? {});
+                if (typeof path !== 'string')
+                    return errorResult('path 必须为字符串');
+                const content = files.get(path);
+                if (content === undefined)
+                    return errorResult(`文件不存在: ${path}`);
+                return textResult(content, { path, bytes: content.length });
+            },
+        },
+        writeFile: {
+            name: 'writeFile',
+            description: '写入内存文件系统（新建或覆盖）',
+            parameters: WRITE_FILE_SCHEMA,
+            execute: async (_id, args) => {
+                const { path, content } = (args ?? {});
+                if (typeof path !== 'string' || typeof content !== 'string') {
+                    return errorResult('path 与 content 必须为字符串');
+                }
+                files.set(path, content);
+                return textResult(`ok: wrote ${content.length} bytes to ${path}`, {
+                    path,
+                    bytes: content.length,
+                });
+            },
+        },
+        listDir: {
+            name: 'listDir',
+            description: '列出内存文件系统中指定前缀下的文件',
+            parameters: LIST_DIR_SCHEMA,
+            execute: async (_id, args) => {
+                const { path } = (args ?? {});
+                const prefix = typeof path === 'string' ? path : '';
+                const entries = [...files.keys()]
+                    .filter((k) => k.startsWith(prefix))
+                    .sort();
+                return textResult(entries.join('\n') || '(empty)', { entries });
+            },
+        },
+        search: {
+            name: 'search',
+            description: '跨内存文件全文检索，返回 file:line:text 命中行',
+            parameters: SEARCH_SCHEMA,
+            execute: async (_id, args) => {
+                const { query } = (args ?? {});
+                if (typeof query !== 'string' || query.length === 0) {
+                    return errorResult('query 必须为非空字符串');
+                }
+                const hits = [];
+                for (const [file, content] of files) {
+                    content.split('\n').forEach((line, i) => {
+                        if (line.includes(query))
+                            hits.push(`${file}:${i + 1}:${line.trim()}`);
+                    });
+                }
+                return textResult(hits.join('\n') || '(no matches)', { hits: hits.length });
+            },
+        },
+        fail: {
+            name: 'fail',
+            description: '恒定返回错误的工具（测错误恢复）',
+            parameters: FAIL_SCHEMA,
+            execute: async () => errorResult('intentional failure'),
+        },
+    };
+    if (!names || names.length === 0)
+        return [...STANDARD_MOCK_TOOLS].map((n) => all[n]);
+    return names.map((n) => all[n]).filter((t) => t !== undefined);
+}
+//# sourceMappingURL=mock-tools.js.map

package/dist/src/core/mock-tools.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"mock-tools.js","sourceRoot":"","sources":["../../../src/core/mock-tools.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAIH,MAAM,aAAa,GAAG;IACpB,IAAI,EAAE,QAAQ;IACd,UAAU,EAAE;QACV,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,QAAQ,EAAE;KACnD;IACD,QAAQ,EAAE,CAAC,SAAS,CAAC;CACtB,CAAC;AAEF,MAAM,gBAAgB,GAAG;IACvB,IAAI,EAAE,QAAQ;IACd,UAAU,EAAE;QACV,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,MAAM,EAAE;KAC9C;IACD,QAAQ,EAAE,CAAC,MAAM,CAAC;CACnB,CAAC;AAEF,MAAM,iBAAiB,GAAG;IACxB,IAAI,EAAE,QAAQ;IACd,UAAU,EAAE;QACV,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,MAAM,EAAE;QAC7C,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,MAAM,EAAE;KACjD;IACD,QAAQ,EAAE,CAAC,MAAM,EAAE,SAAS,CAAC;CAC9B,CAAC;AAEF,MAAM,eAAe,GAAG;IACtB,IAAI,EAAE,QAAQ;IACd,UAAU,EAAE;QACV,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,MAAM,EAAE;KAC9C;CACF,CAAC;AAEF,MAAM,aAAa,GAAG;IACpB,IAAI,EAAE,QAAQ;IACd,UAAU,EAAE;QACV,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,OAAO,EAAE;KAChD;IACD,QAAQ,EAAE,CAAC,OAAO,CAAC;CACpB,CAAC;AAEF,MAAM,WAAW,GAAG;IAClB,IAAI,EAAE,QAAQ;IACd,UAAU,EAAE,EAAE;CACf,CAAC;AAEF,SAAS,UAAU,CAAC,IAAY,EAAE,UAAmB,IAAI;IACvD,MAAM,OAAO,GAAmB,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC;IACzD,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,CAAC;AAC9B,CAAC;AAED,SAAS,WAAW,CAAC,OAAe;IAClC,MAAM,OAAO,GAAmB,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,OAAO,EAAE,EAAE,CAAC,CAAC;IAC9E,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE,CAAC;AAClD,CAAC;AAED,uCAAuC;AACvC,MAAM,CAAC,MAAM,mBAAmB,GAAG;IACjC,MAAM;IACN,UAAU;IACV,WAAW;IACX,SAAS;IACT,QAAQ;IACR,MAAM;CACE,CAAC;AAIX;;;;;GAKG;AACH,MAAM,UAAU,eAAe,CAC7B,KAA6B,EAAE,EAC/B,KAAgB;IAEhB,MAAM,KAAK,GAAwB,IAAI,GAAG,CAAC,MAAM,CAAC,OAAO,CAAC,eAAe,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;IAEhF,MAAM,GAAG,GAAyB;QAChC,IAAI,EAAE;YACJ,IAAI,EAAE,MAAM;YACZ,WAAW,EAAE,aAAa;YAC1B,UAAU,EAAE,aAAa;YACzB,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,IAAI,EAAE,EAAE;gBAC3B,MAAM,OAAO,GAAI,IAA8B,EAAE,OAAO,CAAC;gBACzD,OAAO,UAAU,CAAC,MAAM,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,CAAC;YAC3C,CAAC;SACF;QACD,QAAQ,EAAE;YACR,IAAI,EAAE,UAAU;YAChB,WAAW,EAAE,gBAAgB;YAC7B,UAAU,EAAE,gBAAgB;YAC5B,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,IAAI,EAAE,EAAE;gBAC3B,MAAM,EAAE,IAAI,EAAE,GAAG,CAAC,IAAI,IAAI,EAAE,CAAsB,CAAC;gBACnD,IAAI,OAAO,IAAI,KAAK,QAAQ;oBAAE,OAAO,WAAW,CAAC,aAAa,CAAC,CAAC;gBAChE,MAAM,OAAO,GAAG,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;gBAChC,IAAI,OAAO,KAAK,SAAS;oBAAE,OAAO,WAAW,CAAC,UAAU,IAAI,EAAE,CAAC,CAAC;gBAChE,OAAO,UAAU,CAAC,OAAO,EAAE,EAAE,IAAI,EAAE,KAAK,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;YAC9D,CAAC;SACF;QACD,SAAS,EAAE;YACT,IAAI,EAAE,WAAW;YACjB,WAAW,EAAE,iBAAiB;YAC9B,UAAU,EAAE,iBAAiB;YAC7B,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,IAAI,EAAE,EAAE;gBAC3B,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,GAAG,CAAC,IAAI,IAAI,EAAE,CAGpC,CAAC;gBACF,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,OAAO,OAAO,KAAK,QAAQ,EAAE,CAAC;oBAC5D,OAAO,WAAW,CAAC,uBAAuB,CAAC,CAAC;gBAC9C,CAAC;gBACD,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;gBACzB,OAAO,UAAU,CAAC,aAAa,OAAO,CAAC,MAAM,aAAa,IAAI,EAAE,EAAE;oBAChE,IAAI;oBACJ,KAAK,EAAE,OAAO,CAAC,MAAM;iBACtB,CAAC,CAAC;YACL,CAAC;SACF;QACD,OAAO,EAAE;YACP,IAAI,EAAE,SAAS;YACf,WAAW,EAAE,mBAAmB;YAChC,UAAU,EAAE,eAAe;YAC3B,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,IAAI,EAAE,EAAE;gBAC3B,MAAM,EAAE,IAAI,EAAE,GAAG,CAAC,IAAI,IAAI,EAAE,CAAsB,CAAC;gBACnD,MAAM,MAAM,GAAG,OAAO,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC;gBACpD,MAAM,OAAO,GAAG,CAAC,GAAG,KAAK,CAAC,IAAI,EAAE,CAAC;qBAC9B,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC;qBACnC,IAAI,EAAE,CAAC;gBACV,OAAO,UAAU,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,SAAS,EAAE,EAAE,OAAO,EAAE,CAAC,CAAC;YAClE,CAAC;SACF;QACD,MAAM,EAAE;YACN,IAAI,EAAE,QAAQ;YACd,WAAW,EAAE,iCAAiC;YAC9C,UAAU,EAAE,aAAa;YACzB,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,IAAI,EAAE,EAAE;gBAC3B,MAAM,EAAE,KAAK,EAAE,GAAG,CAAC,IAAI,IAAI,EAAE,CAAuB,CAAC;gBACrD,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;oBACpD,OAAO,WAAW,CAAC,gBAAgB,CAAC,CAAC;gBACvC,CAAC;gBACD,MAAM,IAAI,GAAa,EAAE,CAAC;gBAC1B,KAAK,MAAM,CAAC,IAAI,EAAE,OAAO,CAAC,IAAI,KAAK,EAAE,CAAC;oBACpC,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,CAAC,EAAE,EAAE;wBACtC,IAAI,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC;4BAAE,IAAI,CAAC,IAAI,CAAC,GAAG,IAAI,IAAI,CAAC,GAAG,CAAC,IAAI,IAAI,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC;oBACzE,CAAC,CAAC,CAAC;gBACL,CAAC;gBACD,OAAO,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,cAAc,EAAE,EAAE,IAAI,EAAE,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC;YAC9E,CAAC;SACF;QACD,IAAI,EAAE;YACJ,IAAI,EAAE,MAAM;YACZ,WAAW,EAAE,kBAAkB;YAC/B,UAAU,EAAE,WAAW;YACvB,OAAO,EAAE,KAAK,IAAI,EAAE,CAAC,WAAW,CAAC,qBAAqB,CAAC;SACxD;KACF,CAAC;IAEF,IAAI,CAAC,KAAK,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,GAAG,mBAAmB,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;IACrF,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAa,EAAE,CAAC,CAAC,KAAK,SAAS,CAAC,CAAC;AAC5E,CAAC"}