oh-my-knowledge 0.33.0 → 0.35.0

package/dist/diagnosis/observe-mapper.d.ts

@@ -18,7 +18,7 @@ export interface SkillChainAdvisorySource {
 export interface ObservationRuntimeCheckSource {
     skillName: string;
     id: string;
-    kind: 'hardRule' | 'workflowNode';
+    nodeKind: 'hardRule' | 'workflowNode';
     status: 'passed' | 'attention' | 'manual_review';
     title?: string;
     expectation?: string;
@@ -53,7 +53,7 @@ export interface ExperienceReviewerReportFindingSource {
 export interface SkillDerivedStandardSource {
     skillName: string;
     id: string;
-    kind: 'hard_rule_candidate' | 'workflow_candidate';
+    standardKind: 'hard_rule_candidate' | 'workflow_candidate';
     status: 'pending_review' | 'author_confirmed' | 'rejected' | 'stale';
     title: string;
     body?: string;
@@ -65,4 +65,3 @@ export interface SkillDerivedStandardSource {
 }
 export declare function buildObserveDiagnostics(input: ObserveDiagnosisInput): DiagnosisBundle;
 export declare function maxLifecycle(a: DiagnosisLifecycle, b: DiagnosisLifecycle): DiagnosisLifecycle;
-//# sourceMappingURL=observe-mapper.d.ts.map

package/dist/diagnosis/observe-mapper.js

@@ -71,13 +71,13 @@ function advisoryToDiagnosis(generatedAt, advisory) {
 function runtimeCheckToDiagnosis(generatedAt, check) {
     if (check.status === 'passed')
         return null;
-    const target = check.kind === 'hardRule' ? `runtime:hardRule:${check.id}` : `runtime:workflowNode:${check.id}`;
+    const target = check.nodeKind === 'hardRule' ? `runtime:hardRule:${check.id}` : `runtime:workflowNode:${check.id}`;
     return makeDiagnosis(generatedAt, {
         skillName: check.skillName,
         type: 'runtime_issue',
-        signal: check.kind === 'hardRule' ? 'runtime_hard_rule_review' : 'runtime_workflow_review',
+        signal: check.nodeKind === 'hardRule' ? 'runtime_hard_rule_review' : 'runtime_workflow_review',
         target,
-        title: check.title ?? (check.kind === 'hardRule' ? 'Hard rule runtime check needs review' : 'Workflow runtime check needs review'),
+        title: check.title ?? (check.nodeKind === 'hardRule' ? 'Hard rule runtime check needs review' : 'Workflow runtime check needs review'),
         summary: check.reason ?? check.expectation,
         severity: check.status === 'manual_review' ? 'medium' : 'low',
         audience: 'reviewer',
@@ -135,14 +135,14 @@ function derivedStandardToDiagnosis(generatedAt, standard) {
     return makeDiagnosis(generatedAt, {
         skillName: standard.skillName,
         type: 'standard_candidate',
-        signal: standard.kind,
-        target: `standard:${standard.kind}:${standard.id}`,
+        signal: standard.standardKind,
+        target: `standard:${standard.standardKind}:${standard.id}`,
         title: standard.title,
         summary: standard.body,
         severity: standard.status === 'author_confirmed' ? 'info' : 'low',
         audience: 'skill-author',
         lifecycle: lifecycleFromStandardStatus(standard.status),
-        sourceKind: standard.kind,
+        sourceKind: standard.standardKind,
         sourceId: standard.id,
         evidenceRefs: [],
         producer: standard.source === 'manual' ? 'manual' : 'llm_soft',
@@ -273,4 +273,3 @@ function severityRank(severity) {
 function hashParts(...parts) {
     return createHash('sha256').update(parts.join('\u0000')).digest('hex').slice(0, 16);
 }
-//# sourceMappingURL=observe-mapper.js.map

package/dist/diagnosis/observe-producer.d.ts

@@ -6,4 +6,3 @@ export interface BuildObserveDiagnosticsFromReportOptions {
     skillChains?: Record<string, ObservationSkillChain>;
 }
 export declare function buildObserveDiagnosticsFromReport(report: ObservationInboxReport, options?: BuildObserveDiagnosticsFromReportOptions): DiagnosisBundle;
-//# sourceMappingURL=observe-producer.d.ts.map

package/dist/diagnosis/observe-producer.js

@@ -112,7 +112,7 @@ function chainRuntimeChecks(chain) {
         .map((check) => ({
         skillName: chain.skillName,
         id: check.id,
-        kind: check.kind,
+        nodeKind: check.nodeKind,
         status: check.status,
         title: check.title,
         expectation: check.expectation,
@@ -123,8 +123,8 @@ function chainRuntimeChecks(chain) {
 }
 function evidenceRefsFromRuntimeCheck(check, skillName) {
     return (check.evidenceSnippets ?? []).map((snippet, index) => ({
-        id: `runtime:${skillName}:${check.kind}:${check.id}:${index}`,
-        kind: check.kind,
+        id: `runtime:${skillName}:${check.nodeKind}:${check.id}:${index}`,
+        kind: check.nodeKind,
         label: check.title,
         snippet,
     }));
@@ -196,4 +196,3 @@ function evidenceRef(ref) {
         snippet: ref.snippet,
     };
 }
-//# sourceMappingURL=observe-producer.js.map

package/dist/diagnosis/studio-projection.d.ts

@@ -4,4 +4,3 @@ export declare function buildStudioDiagnosisSummary(bundle?: DiagnosisBundle): S
 export declare function countDiagnostics(bundle?: DiagnosisBundle): number;
 export declare function mergeDiagnosisBundles(bundles: DiagnosisBundle[], generatedAt?: string): DiagnosisBundle;
 export declare function activeStudioDiagnostics(bundle?: DiagnosisBundle): Diagnosis[];
-//# sourceMappingURL=studio-projection.d.ts.map

package/dist/diagnosis/studio-projection.js

@@ -81,4 +81,3 @@ function severityRank(severity) {
         return 2;
     return 1;
 }
-//# sourceMappingURL=studio-projection.js.map

package/dist/diagnosis/types.d.ts

@@ -1,4 +1,3 @@
 export type { Diagnosis, DiagnosisAudience, DiagnosisBundle, DiagnosisEvidenceRef, DiagnosisLifecycle, DiagnosisOccurrence, DiagnosisPatch, DiagnosisScope, DiagnosisSeverity, DiagnosisSource, DiagnosisSourceCoverage, DiagnosisType, StudioDiagnosisSummary, } from '../types/diagnosis.js';
 import type { DiagnosisLifecycle } from '../types/diagnosis.js';
 export declare function isActiveDiagnosisLifecycle(lifecycle: DiagnosisLifecycle): boolean;
-//# sourceMappingURL=types.d.ts.map

package/dist/diagnosis/types.js

@@ -17,4 +17,3 @@ const ACTIVE_DIAGNOSIS_LIFECYCLES = new Set([
 export function isActiveDiagnosisLifecycle(lifecycle) {
     return ACTIVE_DIAGNOSIS_LIFECYCLES.has(lifecycle);
 }
-//# sourceMappingURL=types.js.map

package/dist/doctor/fixer.d.ts

@@ -9,4 +9,3 @@ export interface FixOptions {
     resolveSkillPath?: (reportPath: string) => string;
 }
 export declare function runDoctorFix(opts: FixOptions): Promise<boolean>;
-//# sourceMappingURL=fixer.d.ts.map

package/dist/doctor/fixer.js

@@ -323,4 +323,3 @@ export async function runDoctorFix(opts) {
     }
     return changed;
 }
-//# sourceMappingURL=fixer.js.map

package/dist/doctor/health/builtin-dimensions.d.ts

@@ -8,4 +8,3 @@
  */
 import type { HealthDimensionSpec } from './dimension-spec.js';
 export declare const BUILTIN_HEALTH_DIMENSIONS: HealthDimensionSpec[];
-//# sourceMappingURL=builtin-dimensions.d.ts.map

package/dist/doctor/health/builtin-dimensions.js

@@ -88,4 +88,3 @@ export const BUILTIN_HEALTH_DIMENSIONS = [
 注意:大多数 skill 即使没有示例,LLM 也能从指令描述中推断出正确用法。只有格式确实复杂且无法从描述中推断时才报。`,
     },
 ];
-//# sourceMappingURL=builtin-dimensions.js.map

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

@@ -16,4 +16,3 @@ import type { ComposerRule } from '../../types/doctor.js';
 export declare const SKILL_HEALTH_COMPOSER_ID = "skill_health";
 export type ExecutorFactory = (name: string) => ExecutorFn;
 export declare function makeSkillHealthComposer(execFactory?: ExecutorFactory): ComposerRule;
-//# sourceMappingURL=composer.d.ts.map

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

@@ -26,10 +26,10 @@ export function resolveDoctorTargets(target, cwd) {
         const skillsDir = join(cwd, 'skills');
         if (existsSync(skillsDir) && statSync(skillsDir).isDirectory()) {
             const variants = discoverVariants(skillsDir).filter((v) => v !== 'baseline');
-            return variants.length > 0 ? resolveArtifacts(skillsDir, variants, { strictBaseline: false }) : [];
+            return variants.length > 0 ? resolveArtifacts(skillsDir, variants, { strictBaseline: false, materialize: false }) : [];
         }
         const variants = discoverVariants(cwd).filter((v) => v !== 'baseline');
-        return variants.length > 0 ? resolveArtifacts(cwd, variants, { strictBaseline: false }) : [];
+        return variants.length > 0 ? resolveArtifacts(cwd, variants, { strictBaseline: false, materialize: false }) : [];
     }
     const absTarget = resolve(target);
     if (!existsSync(absTarget)) {
@@ -38,7 +38,7 @@ export function resolveDoctorTargets(target, cwd) {
     const stat = statSync(absTarget);
     if (stat.isFile() && absTarget.endsWith('.md')) {
         // 单文件:用 file-path 模式(走 resolveArtifacts 的 "包含 /" 分支)
-        return resolveArtifacts(dirname(absTarget), [absTarget], { strictBaseline: false });
+        return resolveArtifacts(dirname(absTarget), [absTarget], { strictBaseline: false, materialize: false });
     }
     if (stat.isDirectory()) {
         // 目录自身就是 directory-skill (含 SKILL.md): 按单个 skill 解析,
@@ -48,10 +48,10 @@ export function resolveDoctorTargets(target, cwd) {
         // skillRoot 丢失,assets/foo.md 这类相对依赖会锚到 doctor 的 cwd 而不是
         // skill 目录, 误报缺文件。
         if (existsSync(join(absTarget, 'SKILL.md'))) {
-            return resolveArtifacts(dirname(absTarget), [basename(absTarget)], { strictBaseline: false });
+            return resolveArtifacts(dirname(absTarget), [basename(absTarget)], { strictBaseline: false, materialize: false });
         }
         const variants = discoverVariants(absTarget).filter((v) => v !== 'baseline');
-        return variants.length > 0 ? resolveArtifacts(absTarget, variants, { strictBaseline: false }) : [];
+        return variants.length > 0 ? resolveArtifacts(absTarget, variants, { strictBaseline: false, materialize: false }) : [];
     }
     throw new Error(`doctor target must be a .md file or directory: ${target}`);
 }
@@ -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

@@ -2,8 +2,8 @@
  * Executor result cache.
  *
  * Caches successful executor results to disk to avoid redundant API calls.
- * Cache key v5 = sha256(model + system + prompt + cwd + allowedSkills + executor +
- *                       runtime + mocks + mocksStrict + effort).
+ * Cache key v6 = sha256(model + system + prompt + cwd + allowedSkills + executor +
+ *                       runtime + mocks + mocksStrict + effort + artifactContentHash).
  * Loaded into memory on init, flushed to disk on save().
  *
  * Prefix bumps intentionally invalidate old entries when construct-validity
@@ -15,6 +15,11 @@
  *       报告 meta 标的 effort 跟实际跑的 effort 不一致 — 测量可比性污染);
  *       同时 cache.set 不再砍 turns / toolCalls — 工具类断言 + diagnostic 要看 trace,
  *       砍掉的话 cached rerun 会让工具断言为空、diagnostic 没真实证据,跟 cold run 不一致
+ * - v6: artifact 内容指纹(contentHash)。`system` 只含 SKILL.md 正文,但本地 dir-skill 的
+ *       references/ 资产是真实运行时输入(cwd=skillRoot,agent 可读),改资产只动 contentHash、
+ *       不动 system → 旧 key 会命中旧输出、贴到新 artifactHashes 上,形成静默测量污染。把
+ *       contentHash 纳入 key,资产变即重跑。git skill 的 contentHash 只随 SKILL.md 变(其资产不
+ *       暴露给 executor、本就不该触发重跑),口径自洽
  */
 import type { ExecutorCache } from '../types/index.js';
 export declare function createCache(cacheDir: string): ExecutorCache;
@@ -27,5 +32,7 @@ mocksStrict?: boolean,
 /** Executor effort 也进 key:effort 'low'/'medium'/'high' 改变 LLM 思考预算,
  *  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
+effort?: string, 
+/** artifact 内容指纹(整树 / 单文件哈)。本地 dir-skill 改 references/ 资产只动此值、不动 system,
+ *  不进 key 会让改资产后命中旧输出 → 静默污染。空(baseline / 无 skill)等价无指纹。 */
+artifactContentHash?: string): string;

package/dist/eval-core/cache.js

@@ -2,8 +2,8 @@
  * Executor result cache.
  *
  * Caches successful executor results to disk to avoid redundant API calls.
- * Cache key v5 = sha256(model + system + prompt + cwd + allowedSkills + executor +
- *                       runtime + mocks + mocksStrict + effort).
+ * Cache key v6 = sha256(model + system + prompt + cwd + allowedSkills + executor +
+ *                       runtime + mocks + mocksStrict + effort + artifactContentHash).
  * Loaded into memory on init, flushed to disk on save().
  *
  * Prefix bumps intentionally invalidate old entries when construct-validity
@@ -15,6 +15,11 @@
  *       报告 meta 标的 effort 跟实际跑的 effort 不一致 — 测量可比性污染);
  *       同时 cache.set 不再砍 turns / toolCalls — 工具类断言 + diagnostic 要看 trace,
  *       砍掉的话 cached rerun 会让工具断言为空、diagnostic 没真实证据,跟 cold run 不一致
+ * - v6: artifact 内容指纹(contentHash)。`system` 只含 SKILL.md 正文,但本地 dir-skill 的
+ *       references/ 资产是真实运行时输入(cwd=skillRoot,agent 可读),改资产只动 contentHash、
+ *       不动 system → 旧 key 会命中旧输出、贴到新 artifactHashes 上,形成静默测量污染。把
+ *       contentHash 纳入 key,资产变即重跑。git skill 的 contentHash 只随 SKILL.md 变(其资产不
+ *       暴露给 executor、本就不该触发重跑),口径自洽
  */
 import { readFileSync, writeFileSync, mkdirSync, existsSync } from 'node:fs';
 import { join } from 'node:path';
@@ -114,7 +119,10 @@ mocksStrict,
 /** Executor effort 也进 key:effort 'low'/'medium'/'high' 改变 LLM 思考预算,
  *  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) {
+effort, 
+/** artifact 内容指纹(整树 / 单文件哈)。本地 dir-skill 改 references/ 资产只动此值、不动 system,
+ *  不进 key 会让改资产后命中旧输出 → 静默污染。空(baseline / 无 skill)等价无指纹。 */
+artifactContentHash) {
     // allowedSkills 序列化:undefined → "" / [] → "[]" / [...] → 排序后 JSON。
     // 排序保证 ["a","b"] 和 ["b","a"] 命中同一缓存(语义等价)。
     const isoStr = allowedSkills === undefined
@@ -130,9 +138,8 @@ effort) {
     // executor + runtime + effort 进 cache key:同 model 名走不同 executor 或同 executor
     // 换 binary/SDK 版本时输出可能不同,旧 cache 不可复用。
     const hash = createHash('sha256')
-        .update(`${model || ''}\n${system || ''}\n${prompt || ''}\n${cwd || ''}\n${isoStr}\n${executor || ''}\n${runtimeFingerprint || ''}\n${mockStr}\n${strictStr}\n${effortStr}`)
+        .update(`${model || ''}\n${system || ''}\n${prompt || ''}\n${cwd || ''}\n${isoStr}\n${executor || ''}\n${runtimeFingerprint || ''}\n${mockStr}\n${strictStr}\n${effortStr}\n${artifactContentHash || ''}`)
         .digest('hex')
         .slice(0, 16);
-    return `v5:${hash}`;
+    return `v6:${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

@@ -159,9 +159,12 @@ function buildPreflightEnv(artifacts) {
         return process.env;
     const extraPaths = [];
     for (const a of artifacts) {
-        if (!a.locator)
+        // 与 extractSkillDir 一致:dir-skill 优先隔离副本 execRoot(副本无 node_modules、.bin 检查恒不命中);
+        // 否则本地文件-skill 取 .md 所在目录。git 文件-skill 的 locator 是 git spec、非在盘目录,无 execRoot
+        // 时跳过(取 dirname 会得伪路径)。
+        const dir = a.execRoot ?? (a.locator && a.source !== 'git' ? dirname(a.locator) : null);
+        if (!dir)
             continue;
-        const dir = dirname(a.locator);
         const nodeBin = join(dir, 'node_modules', '.bin');
         if (existsSync(nodeBin))
             extraPaths.push(nodeBin);
@@ -339,4 +342,3 @@ 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

@@ -85,7 +85,10 @@ export async function executeTasks({ tasks, executor, executorName, model, noJud
         // don't share cache entries, and include runtime fingerprint so a binary/SDK bump
         // cannot replay old-runtime outputs under new-runtime report metadata.
         // 同时把 mocks + mocksStrict 进 key:改 mock 配置必须重跑,不能命中老 cache。
-        const key = cacheKey(model, executionPlan.cacheSystem, executionPlan.input.prompt, executionPlan.input.cwd, task.artifact.allowedSkills, effectiveExecutorName, executorRuntime.fingerprint, executionPlan.input.mocks, executionPlan.input.mocksStrict, effort);
+        const key = cacheKey(model, executionPlan.cacheSystem, executionPlan.input.prompt, executionPlan.input.cwd, task.artifact.allowedSkills, effectiveExecutorName, executorRuntime.fingerprint, executionPlan.input.mocks, executionPlan.input.mocksStrict, effort, 
+        // artifact 内容指纹进 key:本地 dir-skill 改 references/ 资产只动 contentHash,system 不变,
+        // 不进 key 会命中旧输出贴到新 artifactHashes(静默污染)。
+        task.artifact.contentHash);
         const cached = cache?.get(key);
         const execStart = Date.now();
         if (cached) {
@@ -333,4 +336,3 @@ 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

@@ -92,11 +92,14 @@ function buildExecutorRuntimesByVariant({ variants, model, executorName, tasks,
         if (runtimes[variant])
             continue;
         const artifact = artifacts.find((a) => a.name === variant);
-        const fallbackSkillDir = artifact?.kind === 'baseline'
-            ? null
-            : artifact?.locator && artifact.source !== 'git'
-                ? dirname(artifact.locator)
-                : request?.skillDir;
+        // 与主路径 extractSkillDir 一致:dir-skill 优先隔离副本 execRoot(副本无 node_modules、PATH 不污染);
+        // 否则 baseline / git 文件-skill 取 null,本地文件-skill 取 .md 所在目录。
+        const fallbackSkillDir = artifact?.execRoot
+            ?? (artifact?.kind === 'baseline' || artifact?.source === 'git'
+                ? null
+                : artifact?.locator
+                    ? dirname(artifact.locator)
+                    : request?.skillDir);
         runtimes[variant] = getExecutorRuntimeFingerprint(executorName, model, {
             skillDir: fallbackSkillDir,
         });
@@ -149,7 +152,11 @@ export function aggregateReport({ runId, variants, model, judgeModel, noJudge, e
             }
         }
     }
-    const artifactHashes = Object.fromEntries(artifacts.map((artifact) => [artifact.name, artifact.content ? hashString(artifact.content) : 'no-skill']));
+    // 整树内容指纹:解析期已由 resolveArtifacts 用 hashArtifactSource 算好挂在 artifact.contentHash 上
+    // (目录-skill 覆盖整棵可分发树含 references/ 资产、文件-skill 为单文件字节),与 install 受管记录的
+    // contentHash 落在同一空间——证据可绑定的前提,也修掉「只哈 SKILL.md 正文、改资产指纹不变」的资产瞎。
+    // baseline / 无 skill 记 'no-skill'。
+    const artifactHashes = Object.fromEntries(artifacts.map((artifact) => [artifact.name, artifact.contentHash ?? 'no-skill']));
     const sampleHashes = Object.fromEntries(samples.map((s) => [s.sample_id, hashSample(s)]));
     const judgeRepeat = request?.judgeRepeat && request.judgeRepeat > 1 ? request.judgeRepeat : undefined;
     const runtimeOptions = { skillDir: request?.skillDir };
@@ -175,7 +182,7 @@ 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,
@@ -189,6 +196,10 @@ export function aggregateReport({ runId, variants, model, judgeModel, noJudge, e
             timestamp: new Date().toISOString(),
             cliVersion: getCliVersion(),
             nodeVersion: process.version,
+            // schemaVersion 3 起,所有 dir-skill(本地 + git)都经隔离副本物化、整棵可分发树哈,与 install
+            // 受管记录 contentHash 同空间(evidence 全绑)。2 是过渡纪元(本地 dir-skill 树哈、git dir-skill
+            // 仅 SKILL.md 字节、不绑);git dir-skill 的 v2 与 v3 不可比。作判别位:消费方对缺位/旧报告不错配比对。
+            schemaVersion: 3,
             artifactHashes,
             sampleHashes,
             ...(noJudge ? {} : { judgePromptHash: getJudgePromptHash(lengthDebiasOn) }),
@@ -289,4 +300,3 @@ 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