oh-my-knowledge 0.33.0 → 0.34.0

package/dist/doctor/health/composer.js

@@ -130,7 +130,7 @@ function listSiblingSkills(skillRoot) {
 export function makeSkillHealthComposer(execFactory = defaultExecutorFactory) {
     return {
         id: SKILL_HEALTH_COMPOSER_ID,
-        kind: 'composer',
+        ruleKind: 'composer',
         severity: 'fatal',
         labelKey: 'cli.doctor.rule.skill_health_check',
         checkAll: (ctx) => composerCheckAll(ctx, execFactory),
@@ -293,4 +293,3 @@ function errorSummaryOutcome(ctx, phase, errMsg, extra = {}) {
         detail: { phase, error: errMsg, ...extra },
     };
 }
-//# sourceMappingURL=composer.js.map

package/dist/doctor/health/dimension-registry.d.ts

@@ -10,4 +10,3 @@ import type { HealthDimensionSpec } from './dimension-spec.js';
 export declare function registerHealthDimension(spec: HealthDimensionSpec): void;
 export declare function getRegisteredHealthDimensions(): HealthDimensionSpec[];
 export declare function __resetHealthDimensionsForTest(): void;
-//# sourceMappingURL=dimension-registry.d.ts.map

package/dist/doctor/health/dimension-registry.js

@@ -25,4 +25,3 @@ export function __resetHealthDimensionsForTest() {
     dimensions.length = 0;
     seenIds.clear();
 }
-//# sourceMappingURL=dimension-registry.js.map

package/dist/doctor/health/dimension-spec.d.ts

@@ -44,4 +44,3 @@ export interface HealthDimensionSpec {
      *  描述这一维度查什么、判定标准、example。 */
     promptSection: string;
 }
-//# sourceMappingURL=dimension-spec.d.ts.map

package/dist/doctor/health/dimension-spec.js

@@ -9,4 +9,3 @@
  * subtype 作为 finding 一级字段),后续可以加可选 hooks(parseFindings 等)。
  */
 export {};
-//# sourceMappingURL=dimension-spec.js.map

package/dist/doctor/health/load-custom-dimensions.d.ts

@@ -1,2 +1 @@
 export declare function loadAndRegisterCustomDimensions(filePath: string): number;
-//# sourceMappingURL=load-custom-dimensions.d.ts.map

package/dist/doctor/health/load-custom-dimensions.js

@@ -28,4 +28,3 @@ export function loadAndRegisterCustomDimensions(filePath) {
     }
     return count;
 }
-//# sourceMappingURL=load-custom-dimensions.js.map

package/dist/doctor/health/parser.d.ts

@@ -23,4 +23,3 @@ export interface HealthParseResult {
 /** 从 cleaned JSON 字串(extractJson 的输出)派发到各维度。 */
 export declare function parseHealthOutput(jsonText: string, dims: HealthDimensionSpec[]): HealthParseResult;
 export declare function deriveOverallHealth(byDimId: Map<string, HealthDimensionResult>): '良好' | '部分缺陷' | '严重缺陷';
-//# sourceMappingURL=parser.d.ts.map

package/dist/doctor/health/parser.js

@@ -225,4 +225,3 @@ export function deriveOverallHealth(byDimId) {
         return '部分缺陷';
     return '良好';
 }
-//# sourceMappingURL=parser.js.map

package/dist/doctor/health/prompt-builder.d.ts

@@ -1,3 +1,2 @@
 export { buildHealthPrompt, } from '../../shared/llm-prompts/skill-health.js';
 export type { PromptInput, } from '../../shared/llm-prompts/skill-health.js';
-//# sourceMappingURL=prompt-builder.d.ts.map

package/dist/doctor/health/prompt-builder.js

@@ -1,2 +1 @@
 export { buildHealthPrompt, } from '../../shared/llm-prompts/skill-health.js';
-//# sourceMappingURL=prompt-builder.js.map

package/dist/doctor/health/register.d.ts

@@ -10,4 +10,3 @@
  * composer 跑之前注册即可(eg. 在 omk 配置 / plugin 入口里)。
  */
 export {};
-//# sourceMappingURL=register.d.ts.map

package/dist/doctor/health/register.js

@@ -17,4 +17,3 @@ for (const spec of BUILTIN_HEALTH_DIMENSIONS) {
     registerHealthDimension(spec);
 }
 registerRule(makeSkillHealthComposer());
-//# sourceMappingURL=register.js.map

package/dist/doctor/index.d.ts

@@ -16,4 +16,3 @@ import type { Artifact, DoctorReport, DoctorRunOptions } from '../types/index.js
  */
 export declare function resolveDoctorTargets(target: string | null | undefined, cwd: string): Artifact[];
 export declare function runDoctor(opts: DoctorRunOptions): Promise<DoctorReport>;
-//# sourceMappingURL=index.d.ts.map

package/dist/doctor/index.js

@@ -229,7 +229,7 @@ export async function runDoctor(opts) {
             ? 'warnings_only'
             : 'passed';
     return {
-        kind: 'doctor',
+        reportKind: 'doctor',
         schemaVersion: DOCTOR_REPORT_SCHEMA_VERSION,
         id: nextReportId(),
         timestamp: new Date().toISOString(),
@@ -243,4 +243,3 @@ export async function runDoctor(opts) {
         ruleStats,
     };
 }
-//# sourceMappingURL=index.js.map

package/dist/doctor/messages.d.ts

@@ -18,4 +18,3 @@ export type DoctorMessageKey = 'cli.doctor.rule.skill_readable' | 'cli.doctor.ru
 export declare const DOCTOR_MESSAGES: Record<DoctorMessageKey, MessageEntry>;
 export declare function tDoctorMessage(key: DoctorMessageKey, lang?: Lang, params?: Record<string, string | number>): string;
 export {};
-//# sourceMappingURL=messages.d.ts.map

package/dist/doctor/messages.js

@@ -199,7 +199,7 @@ export const DOCTOR_MESSAGES = {
         en: '{count} samples missing prompt field',
     },
     'cli.doctor.samples_contract.hint': {
-        zh: '用例必须至少包含 prompt 字段。详见 docs/specs/sample-design-spec.md',
+        zh: '用例必须至少包含 prompt 字段。详见 docs/zh/specs/sample-design-spec.md',
         en: 'samples must contain at least a prompt field. See docs/specs/sample-design-spec.md',
     },
 };
@@ -214,4 +214,3 @@ export function tDoctorMessage(key, lang = DEFAULT_LANG, params) {
     }
     return text;
 }
-//# sourceMappingURL=messages.js.map

package/dist/doctor/preflight.d.ts

@@ -29,4 +29,3 @@ export declare function buildDoctorPreflightContext(input: {
     requires?: DependencyRequirements;
     skillDir: string;
 }): DoctorPreflightContext | null;
-//# sourceMappingURL=preflight.d.ts.map

package/dist/doctor/preflight.js

@@ -29,4 +29,3 @@ export function buildDoctorPreflightContext(input) {
         requires: input.requires,
     };
 }
-//# sourceMappingURL=preflight.js.map

package/dist/doctor/renderer.d.ts

@@ -14,4 +14,3 @@ import type { Lang } from '../types/shared.js';
 import type { DoctorReport } from '../types/index.js';
 export declare function renderDoctorReportText(report: DoctorReport, lang: Lang, write?: (s: string) => void): void;
 export declare function renderDoctorReportJson(report: DoctorReport): string;
-//# sourceMappingURL=renderer.d.ts.map

package/dist/doctor/renderer.js

@@ -119,4 +119,3 @@ export function renderDoctorReportText(report, lang, write = (s) => process.stde
 export function renderDoctorReportJson(report) {
     return JSON.stringify(report, null, 2);
 }
-//# sourceMappingURL=renderer.js.map

package/dist/doctor/rules.d.ts

@@ -28,4 +28,3 @@ export declare const BUILTIN_RULES: DoctorRule[];
 export declare function registerRule(rule: DoctorRuleLike): void;
 export declare function getRegisteredRules(): DoctorRuleLike[];
 export declare function __resetCustomRulesForTest(): void;
-//# sourceMappingURL=rules.d.ts.map

package/dist/doctor/rules.js

@@ -286,4 +286,3 @@ export function getRegisteredRules() {
 export function __resetCustomRulesForTest() {
     customRules.length = 0;
 }
-//# sourceMappingURL=rules.js.map

package/dist/eval-core/bootstrap.d.ts

@@ -77,4 +77,3 @@ export declare function bootstrapDiffCI(scoresA: number[], scoresB: number[], al
  * @param metricFn Function reducing a resampled array to a scalar.
  */
 export declare function bootstrapWithMetric(scores: number[], metricFn: (resampled: number[]) => number, alpha?: number, samples?: number, seed?: number): BootstrapCI;
-//# sourceMappingURL=bootstrap.d.ts.map

package/dist/eval-core/bootstrap.js

@@ -179,4 +179,3 @@ export function bootstrapWithMetric(scores, metricFn, alpha = DEFAULT_BOOTSTRAP_
 function round4(x) {
     return Math.round(x * 10000) / 10000;
 }
-//# sourceMappingURL=bootstrap.js.map

package/dist/eval-core/cache.d.ts

@@ -28,4 +28,3 @@ mocksStrict?: boolean,
  *  输出/工具调用/分数都可能不同,跨 effort 共享 cache 会让报告 meta 标的 effort
  *  跟实际生成的 effort 不一致,违反"两份报告比分数前先比 cliVersion / effort"语义。 */
 effort?: string): string;
-//# sourceMappingURL=cache.d.ts.map

package/dist/eval-core/cache.js

@@ -135,4 +135,3 @@ effort) {
         .slice(0, 16);
     return `v5:${hash}`;
 }
-//# sourceMappingURL=cache.js.map

package/dist/eval-core/comparability.d.ts

@@ -8,4 +8,3 @@ export interface ComparabilityWarning {
 export declare function reportComparabilityWarnings(report: Report): ComparabilityWarning[];
 export declare function crossReportComparabilityWarnings(before: Report, after: Report): ComparabilityWarning[];
 export declare function formatComparabilityWarnings(warnings: ComparabilityWarning[], lang?: Lang): string;
-//# sourceMappingURL=comparability.d.ts.map

package/dist/eval-core/comparability.js

@@ -43,9 +43,9 @@ function packageReasons(pkg, label) {
 }
 function runtimeUnverifiableReasons(runtime) {
     const reasons = [];
-    if (runtime.kind === 'api')
+    if (runtime.runtimeKind === 'api')
         return reasons;
-    if (runtime.kind === 'agent-sdk') {
+    if (runtime.runtimeKind === 'agent-sdk') {
         reasons.push(...packageReasons(runtime.sdk, 'sdk'));
     }
     const binary = runtime.binary;
@@ -70,7 +70,7 @@ function runtimeUnverifiableReasons(runtime) {
     else if (binary.source === 'unknown') {
         reasons.push('binary source unknown');
     }
-    if (runtime.kind === 'unknown')
+    if (runtime.runtimeKind === 'unknown')
         reasons.push('runtime kind unknown');
     return [...new Set(reasons)];
 }
@@ -293,4 +293,3 @@ export function formatComparabilityWarnings(warnings, lang = 'zh') {
         : 'Comparability warnings: the following differences affect strict comparison / construct validity';
     return [header, ...warnings.map((warning) => `  - ${lang === 'zh' ? warning.zh : warning.en}`)].join('\n');
 }
-//# sourceMappingURL=comparability.js.map

package/dist/eval-core/dependency-checker.d.ts

@@ -34,4 +34,3 @@ export declare function checkDependencies(deps: DependencyRequirements, cwd: str
  * 声明,落到 defaultCwd。
  */
 export declare function preflightDependencies(skillContents: string[], samples: Sample[], cwd: string, explicitRequires?: DependencyRequirements, artifacts?: Artifact[]): Promise<DependencyCheckResult>;
-//# sourceMappingURL=dependency-checker.d.ts.map

package/dist/eval-core/dependency-checker.js

@@ -159,7 +159,8 @@ function buildPreflightEnv(artifacts) {
         return process.env;
     const extraPaths = [];
     for (const a of artifacts) {
-        if (!a.locator)
+        // git artifact 的 locator 是 git spec、非在盘目录,对它取 dirname 会得到伪路径;跳过(与 extractSkillDir 一致)。
+        if (!a.locator || a.source === 'git')
             continue;
         const dir = dirname(a.locator);
         const nodeBin = join(dir, 'node_modules', '.bin');
@@ -339,4 +340,3 @@ export async function preflightDependencies(skillContents, samples, cwd, explici
     }
     return result;
 }
-//# sourceMappingURL=dependency-checker.js.map

package/dist/eval-core/evaluation-execution.d.ts

@@ -63,4 +63,3 @@ export declare function preflight(executor: ExecutorFn, model: string, timeoutMs
  *   pipeline must build the map covering every judge before calling this).
  */
 export declare function preflightAllJudges(judgeModels: import('../types/index.js').JudgeConfig[], judgeExecutors: Record<string, ExecutorFn>, timeoutMs?: number): Promise<void>;
-//# sourceMappingURL=evaluation-execution.d.ts.map

package/dist/eval-core/evaluation-execution.js

@@ -333,4 +333,3 @@ export async function preflightAllJudges(judgeModels, judgeExecutors, timeoutMs)
         await preflight(exec, jc.model, timeoutMs);
     }
 }
-//# sourceMappingURL=evaluation-execution.js.map

package/dist/eval-core/evaluation-job.d.ts

@@ -52,4 +52,3 @@ export declare function createFailedJob({ job, error, errorCategory, finishedAt,
     errorCategory?: EvaluationErrorCategory;
     finishedAt?: string;
 }): EvaluationJob;
-//# sourceMappingURL=evaluation-job.d.ts.map

package/dist/eval-core/evaluation-job.js

@@ -116,4 +116,3 @@ export function createFailedJob({ job, error, errorCategory = classifyEvaluation
         errorCategory,
     };
 }
-//# sourceMappingURL=evaluation-job.js.map

package/dist/eval-core/evaluation-reporting.d.ts

@@ -34,4 +34,3 @@ export interface PersistableReport {
 export declare function persistReport(report: PersistableReport, outputDir: string | null): string | null;
 export declare function generateRunId(variants: string[]): string;
 export {};
-//# sourceMappingURL=evaluation-reporting.d.ts.map

package/dist/eval-core/evaluation-reporting.js

@@ -92,9 +92,11 @@ function buildExecutorRuntimesByVariant({ variants, model, executorName, tasks,
         if (runtimes[variant])
             continue;
         const artifact = artifacts.find((a) => a.name === variant);
-        const fallbackSkillDir = artifact?.kind === 'baseline'
+        // git artifact 无在盘 skillDir(content 经 SDK 注入),与 baseline 一样取 null —— 与主路径
+        // extractSkillDir 的判定一致,避免 git variant 的 runtime fingerprint 被 cwd 的 node_modules 污染。
+        const fallbackSkillDir = artifact?.kind === 'baseline' || artifact?.source === 'git'
             ? null
-            : artifact?.locator && artifact.source !== 'git'
+            : artifact?.locator
                 ? dirname(artifact.locator)
                 : request?.skillDir;
         runtimes[variant] = getExecutorRuntimeFingerprint(executorName, model, {
@@ -175,7 +177,7 @@ export function aggregateReport({ runId, variants, model, judgeModel, noJudge, e
         debiasModeList.push('length');
     const totalCostReported = Object.values(summary).every((variant) => variant.execCostReported !== false && variant.judgeCostReported !== false);
     return {
-        kind: 'evaluation',
+        reportKind: 'evaluation',
         id: runId,
         meta: {
             variants,
@@ -289,4 +291,3 @@ export function generateRunId(variants) {
         .join('-vs-');
     return `${variantPart}-${date}-${time}`;
 }
-//# sourceMappingURL=evaluation-reporting.js.map

package/dist/eval-core/execution-strategy.d.ts

@@ -8,4 +8,3 @@ export declare function resolveArtifactExecutionStrategy(artifact: Artifact): Ex
 export declare function resolveExperimentType(artifact: Artifact): ExperimentType;
 export declare function buildVariantConfig(artifact: Artifact): VariantConfig;
 export declare function resolveExecutionStrategy(task: Task, model: string, timeoutMs?: number, verbose?: boolean, effort?: 'low' | 'medium' | 'high' | 'xhigh' | 'max', samplesBaseDir?: string): ExecutionPlan;
-//# sourceMappingURL=execution-strategy.d.ts.map

package/dist/eval-core/execution-strategy.js

@@ -53,7 +53,7 @@ export function buildVariantConfig(artifact) {
     // 因此缺失时直接 throw,强迫调用者修配置,不走默认值掩盖。
     if (!artifact.experimentRole) {
         throw new Error(`artifact "${artifact.name}" 缺少 experimentRole:应该由 CLI --control/--treatment、`
-            + `--config eval.yaml 或 batch workflow 显式注入。参见 docs/specs/terminology-spec.md 三-4。`);
+            + `--config eval.yaml 或 batch workflow 显式注入。参见 docs/zh/specs/terminology-spec.md 三-4。`);
     }
     return {
         variant: artifact.name,
@@ -72,7 +72,11 @@ export function buildVariantConfig(artifact) {
     };
 }
 function extractSkillDir(artifact) {
-    if (!artifact.locator)
+    // git artifact 的 content 经 SDK 注入、根本不落盘,其 locator 是 git spec(`review` / `skills/review`)
+    // 而非磁盘目录。若对它取 dirname,会得到相对进程 cwd 的伪 skillDir(`.` / `skills`),进而让 buildExecEnv
+    // 把 `<cwd>/node_modules/.bin` 误并进 PATH、污染 runtime fingerprint —— 与 baseline(skillDir=null)
+    // 不对称,造成 Δ 污染。git 与 baseline 一样无在盘 skillDir,返回 null。
+    if (!artifact.locator || artifact.source === 'git')
         return null;
     return dirname(artifact.locator);
 }
@@ -162,4 +166,3 @@ export function resolveExecutionStrategy(task, model, timeoutMs, verbose, effort
             };
     }
 }
-//# sourceMappingURL=execution-strategy.js.map

package/dist/eval-core/fact-checker.d.ts

@@ -22,4 +22,3 @@ export declare function extractPathClaims(output: string): string[];
  * Check facts in agent output by verifying file paths exist in cwd.
  */
 export declare function checkFacts(output: string, cwd: string): FactCheckResult;
-//# sourceMappingURL=fact-checker.d.ts.map

package/dist/eval-core/fact-checker.js

@@ -63,4 +63,3 @@ export function checkFacts(output, cwd) {
     const verifiedRate = totalCount > 0 ? Number((verifiedCount / totalCount).toFixed(2)) : 1;
     return { claims, verifiedCount, totalCount, verifiedRate };
 }
-//# sourceMappingURL=fact-checker.js.map

package/dist/eval-core/layer-gates.d.ts

@@ -14,4 +14,3 @@ export interface LayerGateResult {
  * 抽为纯函数便于单测；commands/eval-runner.ts 负责 IO（读 config、打印、退出码）。
  */
 export declare function evaluateLayerGates(summary: Record<string, VariantSummary>, threshold: number): LayerGateResult;
-//# sourceMappingURL=layer-gates.d.ts.map

package/dist/eval-core/layer-gates.js

@@ -39,4 +39,3 @@ export function evaluateLayerGates(summary, threshold) {
     }
     return { allPass, lines };
 }
-//# sourceMappingURL=layer-gates.js.map

package/dist/eval-core/mocks-runtime.d.ts

@@ -97,4 +97,3 @@ declare function readMockHookTemplate(): string;
 export declare const _readMockHookTemplateForTest: typeof readMockHookTemplate;
 export declare function _testMakeTempConfigDir(): string;
 export {};
-//# sourceMappingURL=mocks-runtime.d.ts.map

package/dist/eval-core/mocks-runtime.js

@@ -556,4 +556,3 @@ export function _testMakeTempConfigDir() {
     mkdirSync(dir, { recursive: true });
     return dir;
 }
-//# sourceMappingURL=mocks-runtime.js.map

package/dist/eval-core/schema.d.ts

@@ -27,4 +27,3 @@ export declare function buildVariantResult(execResult: ExecResult, gradeResult:
  */
 export declare function buildVariantSummary(entries: VariantResult[]): VariantSummary;
 export {};
-//# sourceMappingURL=schema.d.ts.map

package/dist/eval-core/schema.js

@@ -306,4 +306,3 @@ function buildEnsembleAggregate(ok) {
         judgeAgreement: { ...agreement, sampleCount: alignedMatrix[0].length },
     };
 }
-//# sourceMappingURL=schema.js.map

package/dist/eval-core/statistics.d.ts

@@ -83,4 +83,3 @@ export declare function tTest(a: number[], b: number[]): {
     df: number;
     significant: boolean;
 };
-//# sourceMappingURL=statistics.d.ts.map

package/dist/eval-core/statistics.js

@@ -207,4 +207,3 @@ export function tTest(a, b) {
         significant,
     };
 }
-//# sourceMappingURL=statistics.js.map

package/dist/eval-core/task-planner.d.ts

@@ -9,4 +9,3 @@ import type { Artifact, Sample, SampleEnvironment, Task } from '../types/index.j
 export declare function renderEnvironmentSection(env: SampleEnvironment | undefined): string | null;
 export declare function buildTasks(samples: Sample[], variants: string[], skills: Record<string, string | null>): Task[];
 export declare function buildTasksFromArtifacts(samples: Sample[], artifacts: Artifact[]): Task[];
-//# sourceMappingURL=task-planner.d.ts.map

package/dist/eval-core/task-planner.js

@@ -69,4 +69,3 @@ export function buildTasksFromArtifacts(samples, artifacts) {
     }
     return tasks;
 }
-//# sourceMappingURL=task-planner.js.map

package/dist/eval-core/verdict.d.ts

@@ -92,4 +92,3 @@ export declare function computeVerdict(report: Report, options?: VerdictOptions)
 export declare function formatVerdictText(result: VerdictResult, options?: {
     verbose?: boolean;
 }): string;
-//# sourceMappingURL=verdict.d.ts.map

package/dist/eval-core/verdict.js

@@ -369,4 +369,3 @@ export function formatVerdictText(result, options = {}) {
     }
     return lines.join('\n');
 }
-//# sourceMappingURL=verdict.js.map

package/dist/eval-workflows/batch-evaluation-workflow.d.ts

@@ -1,8 +1,12 @@
-import type { Artifact, BatchEvaluationReport, EvaluationReport, JobStore, ProgressCallback } from '../types/index.js';
+import type { BatchEvaluationReport, EvaluationReport, JobStore, ProgressCallback, VariantSpec } from '../types/index.js';
 interface RunSingleEvaluationOptions {
     samplesPath: string;
     skillDir: string;
-    artifacts: Artifact[];
+    /** 每个 batch entry 的实验结构(baseline control vs 当前 skill treatment),由
+     *  runEvaluation 的 spec-based 解析统一绑定 role / 隔离。 */
+    variantSpecs: VariantSpec[];
+    /** strict-baseline default,透传给 per-skill runEvaluation(baseline → allowedSkills=[])。 */
+    strictBaseline?: boolean;
     model: string;
     judgeModel: string;
     outputDir: string | null;
@@ -42,6 +46,18 @@ interface CompletedBatchSkillRun {
     report: EvaluationReport;
     filePath: string | null;
 }
+/** Batch 每个 entry 的实验结构固定:baseline control vs 当前 skill treatment。
+ *  构造 VariantSpec 交给 runEvaluation 的 spec-based 解析按身份绑定 role / 隔离——与单跑
+ *  路径同一处绑定逻辑,batch 不再自管 artifact 拼装(此前两处手抄 resolveArtifacts().map()
+ *  导致 #183 角色误绑各存一份)。
+ *  两个 variant 的 allowedSkills 都从 eval.yaml variants[].allowedSkills 取:treatment 按 skill
+ *  名 entry.name 查、baseline 按保留名 `baseline` 查,挂到对应 spec 上由 prepareEvaluationRun
+ *  统一绑定。baseline 的显式声明(白名单或 `[]`)必须保留——eval.yaml variant.allowedSkills
+ *  优先于 strictBaseline 默认,漏挂会让 `--batch --config` 的 baseline 隔离配置静默失效。 */
+export declare function buildBatchVariantSpecs(entry: {
+    name: string;
+    skillPath: string;
+}, variantAllowedSkills?: Record<string, string[]>): VariantSpec[];
 export declare function buildBatchEvaluationReport({ batchRunId, skillDir, skillEntries, skillResults, model, judgeModel, noJudge, executorName, judgeExecutorName, project, owner, tags, concurrency, timeoutMs, totalCostUSD, repeat, judgeModels, noCache, bootstrap, bootstrapSamples, lengthDebias, budget, }: {
     batchRunId: string;
     skillDir: string;
@@ -127,4 +143,3 @@ export declare function executeBatchEvaluationRuns({ skillDir, skillEntries, mod
     filePath: string | null;
 }>;
 export {};
-//# sourceMappingURL=batch-evaluation-workflow.d.ts.map

package/dist/eval-workflows/batch-evaluation-workflow.js

@@ -1,9 +1,34 @@
-import { dirname, resolve } from 'node:path';
+import { dirname } from 'node:path';
 import { DEFAULT_OUTPUT_DIR, generateRunId, getCliVersion, getGitInfo, persistReport } from '../eval-core/evaluation-reporting.js';
 import { buildEvaluationRequest, createEvaluationRun, createSucceededJob, finalizeEvaluationRun } from '../eval-core/evaluation-job.js';
 import { getExecutorRuntimeFingerprint } from '../executors/runtime-fingerprint.js';
 import { createFileJobStore, DEFAULT_JOBS_DIR } from '../server/job-store.js';
-import { resolveArtifacts } from '../inputs/skill-loader.js';
+/** Batch 每个 entry 的实验结构固定:baseline control vs 当前 skill treatment。
+ *  构造 VariantSpec 交给 runEvaluation 的 spec-based 解析按身份绑定 role / 隔离——与单跑
+ *  路径同一处绑定逻辑,batch 不再自管 artifact 拼装(此前两处手抄 resolveArtifacts().map()
+ *  导致 #183 角色误绑各存一份)。
+ *  两个 variant 的 allowedSkills 都从 eval.yaml variants[].allowedSkills 取:treatment 按 skill
+ *  名 entry.name 查、baseline 按保留名 `baseline` 查,挂到对应 spec 上由 prepareEvaluationRun
+ *  统一绑定。baseline 的显式声明(白名单或 `[]`)必须保留——eval.yaml variant.allowedSkills
+ *  优先于 strictBaseline 默认,漏挂会让 `--batch --config` 的 baseline 隔离配置静默失效。 */
+export function buildBatchVariantSpecs(entry, variantAllowedSkills) {
+    const baselineAllowed = variantAllowedSkills?.baseline;
+    const treatmentAllowed = variantAllowedSkills?.[entry.name];
+    return [
+        {
+            name: 'baseline',
+            role: 'control',
+            expr: 'baseline',
+            ...(baselineAllowed !== undefined && { allowedSkills: baselineAllowed }),
+        },
+        {
+            name: entry.name,
+            role: 'treatment',
+            expr: entry.skillPath,
+            ...(treatmentAllowed !== undefined && { allowedSkills: treatmentAllowed }),
+        },
+    ];
+}
 function commonRuntime(runtimes) {
     const values = Object.values(runtimes);
     if (values.length === 0)
@@ -110,7 +135,7 @@ export function buildBatchEvaluationReport({ batchRunId, skillDir, skillEntries,
         ...(noJudge ? {} : { runtime: getExecutorRuntimeFingerprint(jc.executor, jc.model, { skillDir }) }),
     }));
     const report = {
-        kind: 'batch-evaluation',
+        reportKind: 'batch-evaluation',
         id: batchRunId,
         mode: 'skill',
         meta: {
@@ -146,21 +171,13 @@ export async function executeBatchEvaluationRuns({ skillDir, skillEntries, model
     for (let i = 0; i < skillEntries.length; i++) {
         const entry = skillEntries[i];
         onSkillProgress?.({ phase: 'start', skill: entry.name, current: i + 1, total: skillEntries.length });
-        // Batch mode 的实验结构固定为 baseline control vs 当前 skill treatment。
-        const perSkillAllowedSkills = variantAllowedSkills?.[entry.name] !== undefined
-            ? { ...variantAllowedSkills, [entry.skillPath]: variantAllowedSkills[entry.name] }
-            : variantAllowedSkills;
-        const skillArtifacts = resolveArtifacts(resolve(skillDir), ['baseline', entry.skillPath], { strictBaseline, variantAllowedSkills: perSkillAllowedSkills }).map((artifact) => {
-            if (artifact.name === entry.skillPath) {
-                return { ...artifact, name: entry.name, experimentRole: 'treatment' };
-            }
-            return { ...artifact, experimentRole: 'control' };
-        });
+        const variantSpecs = buildBatchVariantSpecs(entry, variantAllowedSkills);
         const childRunId = `${batchRunId}-${String(i + 1).padStart(2, '0')}-${safeRunIdPart(entry.name)}`;
         const { report, filePath } = await runSingleEvaluation({
             samplesPath: entry.samplesPath,
             skillDir,
-            artifacts: skillArtifacts,
+            variantSpecs,
+            strictBaseline,
             model,
             judgeModel,
             outputDir,
@@ -226,4 +243,3 @@ export async function executeBatchEvaluationRuns({ skillDir, skillEntries, model
         await resolvedJobStore.save(job.jobId, job);
     return { report: batchReport, filePath };
 }
-//# sourceMappingURL=batch-evaluation-workflow.js.map

package/dist/eval-workflows/evaluation-pipeline/preflight-warnings.d.ts

@@ -19,4 +19,3 @@ export declare function buildPowerWarnings(sampleCount: number, repeat: number,
 export declare function emitPowerWarnings(sampleCount: number, repeat: number, lang: Lang): void;
 export declare function buildIsolationWarnings(artifacts: Artifact[], strictBaseline: boolean | undefined): string[];
 export declare function emitIsolationWarnings(artifacts: Artifact[], strictBaseline: boolean | undefined): void;
-//# sourceMappingURL=preflight-warnings.d.ts.map

package/dist/eval-workflows/evaluation-pipeline/preflight-warnings.js

@@ -64,4 +64,3 @@ export function emitIsolationWarnings(artifacts, strictBaseline) {
         process.stderr.write(`${w}\n`);
     }
 }
-//# sourceMappingURL=preflight-warnings.js.map

package/dist/eval-workflows/evaluation-pipeline/report-finalize.d.ts

@@ -29,4 +29,3 @@ export declare function finalizeEvaluationReport({ report, results, artifacts, v
     samples: Sample[];
 }): Report;
 export {};
-//# sourceMappingURL=report-finalize.d.ts.map

package/dist/eval-workflows/evaluation-pipeline/report-finalize.js

@@ -54,4 +54,3 @@ export function finalizeEvaluationReport({ report, results, artifacts, variantNa
     }
     return report;
 }
-//# sourceMappingURL=report-finalize.js.map

package/dist/eval-workflows/evaluation-pipeline/run-state.d.ts

@@ -57,4 +57,3 @@ export declare function finalizeSuccessfulRun(state: EvaluationRunState): {
 };
 export declare function persistSuccessfulJob(state: EvaluationRunState, job: EvaluationJob): Promise<void>;
 export declare function persistFailedJob(state: EvaluationRunState, err: unknown): Promise<void>;
-//# sourceMappingURL=run-state.d.ts.map

package/dist/eval-workflows/evaluation-pipeline/run-state.js

@@ -85,4 +85,3 @@ export async function persistFailedJob(state, err) {
         await state.resolvedJobStore.save(state.jobId, failedJob);
     }
 }
-//# sourceMappingURL=run-state.js.map

package/dist/eval-workflows/evaluation-pipeline/test-set-hash.d.ts

@@ -18,4 +18,3 @@
  */
 export declare function computeTestSetHash(samplesPath: string, sourceFiles?: string[]): string | null;
 export declare function _computeTestSetHashForTest(samplesPath: string, sourceFiles?: string[]): string | null;
-//# sourceMappingURL=test-set-hash.d.ts.map

package/dist/eval-workflows/evaluation-pipeline/test-set-hash.js

@@ -43,4 +43,3 @@ export function computeTestSetHash(samplesPath, sourceFiles) {
 export function _computeTestSetHashForTest(samplesPath, sourceFiles) {
     return computeTestSetHash(samplesPath, sourceFiles);
 }
-//# sourceMappingURL=test-set-hash.js.map

package/dist/eval-workflows/evaluation-pipeline.d.ts

@@ -94,4 +94,3 @@ export declare function executeEvaluationPipeline({ samplesPath, samplesBaseDir,
     report: Report;
     filePath: string | null;
 }>;
-//# sourceMappingURL=evaluation-pipeline.d.ts.map

package/dist/eval-workflows/evaluation-pipeline.js

@@ -176,4 +176,3 @@ requires: _requires, layeredStats = false, repeat, batch, judgeRepeat, judgeMode
         await stopAllServers();
     }
 }
-//# sourceMappingURL=evaluation-pipeline.js.map