oh-my-knowledge 0.32.0 → 0.34.0

package/dist/server/report-store.js

@@ -39,12 +39,12 @@ export function createFileStore(dir) {
         if (!data || typeof data !== 'object')
             return null;
         const record = data;
-        if (record.kind === 'evaluation') {
+        if (record.reportKind === 'evaluation') {
             if (!record.meta || !record.summary || !Array.isArray(record.results))
                 return null;
             return { ...record, id: typeof record.id === 'string' && record.id ? record.id : fallbackId };
         }
-        if (record.kind === 'batch-evaluation') {
+        if (record.reportKind === 'batch-evaluation') {
             if (!record.meta || !Array.isArray(record.items))
                 return null;
             return { ...record, id: typeof record.id === 'string' && record.id ? record.id : fallbackId };
@@ -57,14 +57,14 @@ export function createFileStore(dir) {
             && Array.isArray(record.results)) {
             return {
                 ...record,
-                kind: 'evaluation',
+                reportKind: 'evaluation',
                 id: typeof record.id === 'string' && record.id ? record.id : fallbackId,
             };
         }
         return null;
     }
     function isEvaluationReport(report) {
-        return report.kind === 'evaluation';
+        return report.reportKind === 'evaluation';
     }
     // Studio 每个 / 和 /skills/<name> 请求都调 list(),里面对每个 .json 同步 readFile +
     // JSON.parse。报告数上来后这是主性能瓶颈。缓存策略:fingerprint = dir mtime + 文件名
@@ -217,9 +217,9 @@ export async function queryJob(jobStore, id) {
 export async function queryRunList(reportStore) {
     return (await reportStore.list()).map((report) => ({
         id: report.id,
-        kind: report.kind,
+        reportKind: report.reportKind,
         meta: report.meta,
-        ...(report.kind === 'evaluation' ? { summary: report.summary } : { items: report.items }),
+        ...(report.reportKind === 'evaluation' ? { summary: report.summary } : { items: report.items }),
     }));
 }
 export async function queryRun(reportStore, id) {
@@ -247,4 +247,3 @@ export async function queryTrend(reportStore, variantName) {
         runs,
     };
 }
-//# sourceMappingURL=report-store.js.map

package/dist/server/skill-index.d.ts

@@ -11,4 +11,3 @@ export declare function _resetSkillIndexCache(): void;
 export declare function buildSkillIndex(reports: ReportDocument[], analysesDir: string, doctorsDir: string, observationsDir?: string): SkillIndex;
 /** 单 skill 详情查询(用于 /skills/<name> 详情页)— 复用 buildSkillIndex 的扫描结果。 */
 export declare function getSkillEntry(idx: SkillIndex, skillName: string): SkillIndexEntry | null;
-//# sourceMappingURL=skill-index.d.ts.map

package/dist/server/skill-index.js

@@ -17,6 +17,7 @@
  */
 import { existsSync, readdirSync, readFileSync, statSync } from 'node:fs';
 import { join, isAbsolute, basename, dirname } from 'node:path';
+import { confidenceOf } from '../observability/skill-health-analyzer.js';
 import { computeVerdict } from '../eval-core/verdict.js';
 import { detectInsights } from './skill-insights.js';
 import { DEFAULT_OBSERVATIONS_DIR, loadLatestObservationInboxReports } from '../observability/inbox.js';
@@ -68,7 +69,7 @@ function buildIndexFingerprint(reports, analysesDir, doctorsDir, observationsDir
     // 每一段的 right-hand-side 从旧的 "{dir-mtime}-{file-count}" 双标量升级成
     // safeDirJsonContentFingerprint 返回的 "{dir-mtime}|{file1}:{m}:{s},..."
     // content-aware 字符串。
-    const reportIds = reports.map((r) => `${r.id}:${r.meta?.timestamp ?? ''}:${r.kind === 'evaluation' ? r.meta.evolve?.skillName ?? '' : ''}`).join(',');
+    const reportIds = reports.map((r) => `${r.id}:${r.meta?.timestamp ?? ''}:${r.reportKind === 'evaluation' ? r.meta.evolve?.skillName ?? '' : ''}`).join(',');
     const doctorsFp = safeDirJsonContentFingerprint(doctorsDir);
     const analysesFp = safeDirJsonContentFingerprint(analysesDir);
     const observationsFp = safeDirJsonContentFingerprint(observationsDir);
@@ -222,7 +223,7 @@ function scanDoctorReports(dir) {
             continue;
         try {
             const data = JSON.parse(readFileSync(join(dir, file), 'utf-8'));
-            if (data?.kind !== 'doctor' || !Array.isArray(data.skills))
+            if (data?.reportKind !== 'doctor' || !Array.isArray(data.skills))
                 continue;
             const ts = data.timestamp;
             for (const sr of data.skills) {
@@ -259,7 +260,7 @@ export function buildSkillIndex(reports, analysesDir, doctorsDir, observationsDi
     // ── eval 聚合(历史 list)─────────────────────────────────
     const evalBy = {};
     for (const r of reports) {
-        if (r.kind !== 'evaluation')
+        if (r.reportKind !== 'evaluation')
             continue;
         const variants = r.meta.variants || [];
         for (const v of variants) {
@@ -295,6 +296,8 @@ export function buildSkillIndex(reports, analysesDir, doctorsDir, observationsDi
                         failureRate: h.toolFailureRate,
                         segmentCount: h.segmentCount,
                         gapRate: h.gap?.weightedGapRate ?? 0,
+                        // Legacy reports (pre-confidence) fall back to deriving from segmentCount.
+                        confidence: h.confidence ?? confidenceOf(h.segmentCount),
                     };
                     if (!observeBy[skill])
                         observeBy[skill] = [];
@@ -344,7 +347,7 @@ export function buildSkillIndex(reports, analysesDir, doctorsDir, observationsDi
     // list 页对每个 entry 跑 detectInsights 的 CPU 开销迁移到这里,只 miss 时算一次。
     const insightsBySkill = new Map();
     for (const ent of entries) {
-        const evalReport = ent.eval ? reports.find((r) => r.id === ent.eval.reportId && r.kind === 'evaluation') : undefined;
+        const evalReport = ent.eval ? reports.find((r) => r.id === ent.eval.reportId && r.reportKind === 'evaluation') : undefined;
         insightsBySkill.set(ent.skillName, detectInsights(ent, evalReport ?? null, {
             diagnostics: diagnosisBundle.bySkill[ent.skillName] ?? [],
         }));
@@ -378,4 +381,3 @@ export function buildSkillIndex(reports, analysesDir, doctorsDir, observationsDi
 export function getSkillEntry(idx, skillName) {
     return idx.entries.find((e) => e.skillName === skillName) ?? null;
 }
-//# sourceMappingURL=skill-index.js.map

package/dist/server/skill-insights.d.ts

@@ -29,4 +29,3 @@ export declare function flattenRecommendations(insights: Insight[]): InsightReco
 /** 按 audience 分桶 — UI 拆区用。 */
 export declare function groupInsightsByAudience(insights: Insight[]): Record<InsightAudience, Insight[]>;
 export type _SkillEvalSnapshotForInsight = SkillEvalSnapshot;
-//# sourceMappingURL=skill-insights.d.ts.map

package/dist/server/skill-insights.js

@@ -64,7 +64,7 @@ function getFailedSamples(report, variant) {
  * 各 detector 内部一般也用 `>= 2` 作为 cluster 阈值(详情参考各 detector 注释)。
  *
  * 采样不足提醒:单 skill < 2 个非诱错 sample 时不会触发任何 cluster-based insight,
- * 这是设计行为(没数据不下结论),用户层面应该看到"用例数 < 5,建议补样本"提示
+ * 这是设计行为(没数据不下结论),用户层面应该看到"用例数 < 5,建议补用例"提示
  * (在 detail page 空态有,list 页面 todo)。
  */
 function severityOfCount(n) {
@@ -74,6 +74,23 @@ function severityOfCount(n) {
         return 'medium';
     return 'low';
 }
+/** Cap an observe-derived severity by sample-size confidence: an underpowered snapshot
+ *  (very few segments) must not raise a hard high-severity insight — the gap / failure
+ *  rate is real but the sample is too small to act on as a high-priority signal. */
+function capByObserveConfidence(severity, observe) {
+    if (observe?.confidence === 'underpowered')
+        return severity === 'high' ? 'low' : severity;
+    return severity;
+}
+/** Evidence caveat appended to underpowered observe insights (null when N is sufficient). */
+function underpoweredCaveat(observe) {
+    if (observe?.confidence !== 'underpowered')
+        return null;
+    return {
+        perspective: 'observe', status: 'silent',
+        message: `样本量仅 ${observe.segmentCount} 段(underpowered) — 以下为低置信参考信号,需累积更多真实使用 trace 才能下结论`,
+    };
+}
 function pickIllustrations(samples, n = 2) {
     return samples.slice(0, n).map((s) => ({
         sampleId: s.sampleId,
@@ -100,7 +117,7 @@ function detectEnvironmentBlocked(evalReport, variantName) {
         id: 'environment-blocked-mocks',
         category: 'environment-blocked-mocks',
         audience: 'sample-author',
-        title: `${blocked.length} 条评测样本因 mock 没盖到 LLM 想调的工具而 fail`,
+        title: `${blocked.length} 条评测用例因 mock 没盖到 LLM 想调的工具而 fail`,
         description: 'LLM 行为正确但工具调用被 mock-strict 拦截 — 这是 sample 设计问题,不是 skill 问题。要么 mock 列表漏写,要么 command_glob/file_path 写得太严没匹配到 LLM 真实调用。',
         severity: severityOfCount(blocked.length),
         affectedCount: blocked.length,
@@ -175,7 +192,7 @@ function detectSkillDocGap(doctor, evalReport, observe, variantName) {
     if (failedDocSamples.length > 0) {
         evidence.push({
             perspective: 'eval-functional', status: 'flagged',
-            message: `${failedDocSamples.length} 条评测样本失败,LLM 没找到该读哪段文档(${failedDocSamples.map((f) => f.sampleId).join(', ')})`,
+            message: `${failedDocSamples.length} 条评测用例失败,LLM 没找到该读哪段文档(${failedDocSamples.map((f) => f.sampleId).join(', ')})`,
             ref: failedDocSamples[0].sampleId,
             illustrations: pickIllustrations(failedDocSamples, 2),
         });
@@ -237,7 +254,7 @@ function detectSkillDocGap(doctor, evalReport, observe, variantName) {
         category: 'skill-doc-gap',
         audience: failedDocSamples.length > 0 ? 'skill-author' : 'sample-author',
         title: failedDocSamples.length > 0
-            ? `LLM 漏读 skill 引用的文件,导致 ${failedDocSamples.length} 条样本失败`
+            ? `LLM 漏读 skill 引用的文件,导致 ${failedDocSamples.length} 条用例失败`
             : 'skill 引用了文件但没声明在 sample.environment 里',
         description: 'skill 文档提到某些前置文件,但 sample 没标"已就绪"。LLM 看 prompt 时会试图先探测/Read 这些文件,在 mocks-strict 环境下被拦,流程在第一步就停。',
         severity,
@@ -341,7 +358,7 @@ NAMESPACE=$(<your-cli> auth status -o json | jq -r '.username')
         id: `failure-mode-skill:${topMode}`,
         category: 'failure-mode-skill',
         audience: 'skill-author',
-        title: `${topSamples.length} 条样本踩同一个写法坑:${topMode}`,
+        title: `${topSamples.length} 条用例踩同一个写法坑:${topMode}`,
         description: MODE_USERFACING_DESC[topMode] ?? `多 sample 共享 failureMode 「${topMode}」,skill 文档对此没抑制`,
         severity: severityOfCount(topSamples.length),
         affectedCount: topSamples.length,
@@ -379,13 +396,15 @@ function detectProductionInstability(doctor, observe) {
         return null;
     const depRule = doctor?.results.find((r) => r.ruleId === 'dependencies_present'
         && (r.status === 'warn' || r.status === 'fail'));
+    const severity = capByObserveConfidence('high', observe);
+    const caveat = underpoweredCaveat(observe);
     return {
         id: 'production-instability',
         category: 'production-instability',
         audience: 'skill-author',
         title: `生产环境跑这个 skill 时,工具失败率 ${(observe.failureRate * 100).toFixed(0)}%`,
         description: '真实用户用这个 skill,LLM 调出去的工具经常报错。可能是凭证/网络/上游 SLA 问题,也可能是 skill 教错了用什么工具/参数。',
-        severity: 'high',
+        severity,
         affectedCount: Math.round(observe.segmentCount * observe.failureRate),
         stageRefs: {
             observeRefs: ['high-failure-rate'],
@@ -408,11 +427,12 @@ function detectProductionInstability(doctor, observe) {
                         ? 'doctor 没警告依赖,失败可能在 CLI/凭证/上游 API 层(skill 自身可能没错)'
                         : '没跑 doctor,无法对照',
                 },
+            ...(caveat ? [caveat] : []),
         ],
         recommendations: [
             {
                 action: '排查产线工具失败原因:CLI 安装 / 凭证有效期 / 网络 / 上游 API SLA',
-                priority: 'high',
+                priority: severity,
             },
             ...(depRule ? [{
                     action: '依据 doctor 警告补依赖声明 / 改 skill 引用',
@@ -454,7 +474,7 @@ function detectCoverageGap(evalReport, observe, variantName) {
     if (evalUncovered.length > 0) {
         evidence.push({
             perspective: 'eval-functional', status: 'flagged',
-            message: `评测样本也漏覆盖 ${evalUncovered.length} 个文件:${evalUncovered.slice(0, 3).join(', ')}${evalUncovered.length > 3 ? '...' : ''}`,
+            message: `评测用例也漏覆盖 ${evalUncovered.length} 个文件:${evalUncovered.slice(0, 3).join(', ')}${evalUncovered.length > 3 ? '...' : ''}`,
         });
     }
     else {
@@ -463,8 +483,10 @@ function detectCoverageGap(evalReport, observe, variantName) {
             message: '评测覆盖率 OK,gap 主要来自生产用法跟 sample 设计有差距',
         });
     }
-    const severity = observe.gapRate >= 0.4 ? 'high'
-        : observe.gapRate >= 0.2 ? 'medium' : 'low';
+    const caveat = underpoweredCaveat(observe);
+    if (caveat)
+        evidence.push(caveat);
+    const severity = capByObserveConfidence(observe.gapRate >= 0.4 ? 'high' : observe.gapRate >= 0.2 ? 'medium' : 'low', observe);
     return {
         id: 'coverage-gap',
         category: 'coverage-gap',
@@ -515,14 +537,20 @@ function detectSkillTooLong(doctor, observe) {
             perspective: 'observe', status: 'flagged',
             message: `生产 gap ${(observe.gapRate * 100).toFixed(0)}% — 长 skill 让 LLM 找不到段落,跟 doctor 警告印证`,
         });
+        const caveat = underpoweredCaveat(observe);
+        if (caveat)
+            evidence.push(caveat);
     }
+    // doctor warn 是主信号;只有当 observe gap 高且样本量足够时才升到 medium。underpowered 的高 gap
+    // 不足以把严重度抬上去 —— 跟 detectProductionInstability / detectCoverageGap 的 confidence 口径一致。
+    const observeBump = observe != null && observe.gapRate >= 0.3 && observe.confidence !== 'underpowered';
     return {
         id: 'skill-too-long',
         category: 'skill-too-long',
         audience: 'skill-author',
         title: 'skill 文档太长,LLM 容易漏读',
         description: 'skill 内容超过推荐上限。长 skill 在 LLM 上下文里会被压缩注意力,LLM 跟着前几段走,后面的关键约束容易漏读。',
-        severity: observe && observe.gapRate >= 0.3 ? 'medium' : 'low',
+        severity: observeBump ? 'medium' : 'low',
         affectedCount: 1,
         stageRefs: {
             doctorRuleIds: ['skill_readable'],
@@ -585,7 +613,7 @@ function detectOmkDoctorBlindspot(doctor, evalReport, variantName) {
         evidence: [
             {
                 perspective: 'eval-functional', status: 'flagged',
-                message: `${candidates.length} 条样本因 ${observedModes.join(' / ')} 失败,doctor 全 pass`,
+                message: `${candidates.length} 条用例因 ${observedModes.join(' / ')} 失败,doctor 全 pass`,
                 ref: candidates[0].sampleId,
             },
             {
@@ -606,7 +634,7 @@ function detectOmkDoctorBlindspot(doctor, evalReport, variantName) {
                     snippet: `// omk 仓库新增 composer rule(omk 维护者负责,非 skill 开发者职责)
 export const skillAntiPatternComposer: ComposerRule = {
   id: 'skill_anti_patterns',
-  kind: 'composer',
+  ruleKind: 'composer',
   severity: 'warn',
   labelKey: 'doctor.skill_anti_patterns.label',
   async checkAll(ctx): Promise<ComposerOutcome[]> {
@@ -785,4 +813,3 @@ export function groupInsightsByAudience(insights) {
         out[ins.audience].push(ins);
     return out;
 }
-//# sourceMappingURL=skill-insights.js.map

package/dist/shared/hard-rules.d.ts

@@ -39,4 +39,3 @@ export declare function extractSkillHardRules(content: string): SkillHardRule[];
 export declare function validateSkillWorkflows(content: string): SkillWorkflowsValidationResult;
 export declare function extractSkillWorkflows(content: string): SkillWorkflow[];
 export declare function extractMarkdownStepWorkflows(content: string): SkillWorkflow[];
-//# sourceMappingURL=hard-rules.d.ts.map

package/dist/shared/hard-rules.js

@@ -239,4 +239,3 @@ function parseStepHeading(title) {
         return undefined;
     return { title: (match[4] || title).trim() };
 }
-//# sourceMappingURL=hard-rules.js.map

package/dist/shared/llm-prompts/index.d.ts

@@ -11,4 +11,3 @@ export declare function readPromptDocument(options: {
     id: string;
     version: string;
 }): LlmPromptDocument;
-//# sourceMappingURL=index.d.ts.map

package/dist/shared/llm-prompts/index.js

@@ -19,4 +19,3 @@ export function readPromptDocument(options) {
         hash: hashPromptText(body),
     };
 }
-//# sourceMappingURL=index.js.map

package/dist/shared/llm-prompts/skill-health.d.ts

@@ -19,4 +19,3 @@ export interface PromptInput {
     siblingSkills: string[];
 }
 export declare function buildHealthPrompt(input: PromptInput, dims: HealthDimensionSpec[]): string;
-//# sourceMappingURL=skill-health.d.ts.map

package/dist/shared/llm-prompts/skill-health.js

@@ -167,4 +167,3 @@ function renderBlock4JsonSchema(skillName, dims) {
 
 读 SKILL.md(已 inline)→ 模拟执行 → 出 JSON。`;
 }
-//# sourceMappingURL=skill-health.js.map

package/dist/shared/time.d.ts

@@ -1,2 +1 @@
 export declare function durationMsBetween(start?: string, end?: string): number | undefined;
-//# sourceMappingURL=time.d.ts.map

package/dist/shared/time.js

@@ -7,4 +7,3 @@ export function durationMsBetween(start, end) {
         return undefined;
     return endMs - startMs;
 }
-//# sourceMappingURL=time.js.map

package/dist/shared/tool-search.d.ts

@@ -6,4 +6,3 @@ export declare function toolCallQuery(tc: ToolCallInfo): {
 export declare function bashSearchOrProbeCommand(command: string): boolean;
 export declare function isSearchToolCall(tc: ToolCallInfo): boolean;
 export declare function isFailedSearchToolCall(tc: ToolCallInfo): boolean;
-//# sourceMappingURL=tool-search.d.ts.map

package/dist/shared/tool-search.js

@@ -101,4 +101,3 @@ export function isFailedSearchToolCall(tc) {
     }
     return false;
 }
-//# sourceMappingURL=tool-search.js.map

package/dist/types/dependencies.d.ts

@@ -20,4 +20,3 @@ export interface DependencyCheckResult {
     ok: boolean;
     missing: DependencyIssue[];
 }
-//# sourceMappingURL=dependencies.d.ts.map

package/dist/types/dependencies.js

@@ -1,2 +1 @@
 export {};
-//# sourceMappingURL=dependencies.js.map

package/dist/types/diagnosis.d.ts

@@ -85,4 +85,3 @@ export interface StudioDiagnosisSummary {
     bySkill: Record<string, number>;
     byAudience: Record<string, number>;
 }
-//# sourceMappingURL=diagnosis.d.ts.map

package/dist/types/diagnosis.js

@@ -1,2 +1 @@
 export {};
-//# sourceMappingURL=diagnosis.js.map

package/dist/types/doctor.d.ts

@@ -10,7 +10,7 @@ export type DoctorSkillStatus = 'pass' | 'warn' | 'fail';
 export type DoctorOutcome = 'passed' | 'warnings_only' | 'failed';
 /** Bumped whenever DoctorReport schema changes in a way CI consumers should
  *  be able to detect. CI can pin/check this when parsing the JSON. */
-export declare const DOCTOR_REPORT_SCHEMA_VERSION = "1.0.0";
+export declare const DOCTOR_REPORT_SCHEMA_VERSION = "2.0.0";
 export interface DoctorRuleResult {
     ruleId: string;
     severity: DoctorSeverity;
@@ -41,11 +41,11 @@ export interface ComposerOutcome extends DoctorRuleCheckOutcome {
     severity?: DoctorSeverity;
 }
 /** 复合 rule:一次 checkAll() 产出多条 result,适合"单次 LLM 调用 → N 个维度结果"
- *  这类场景。doctor engine 检测 kind === 'composer' 时把数组结果展开,每条共享
+ *  这类场景。doctor engine 检测 ruleKind === 'composer' 时把数组结果展开,每条共享
  *  同一个 groupId(= composer.id)以便 renderer 分组。 */
 export interface ComposerRule {
     id: string;
-    kind: 'composer';
+    ruleKind: 'composer';
     /** 默认 severity(子 outcome 未指定时用)。 */
     severity: DoctorSeverity;
     /** 默认 labelKey(子 outcome 未指定时用,通常是 composer 整体的标签)。 */
@@ -94,7 +94,7 @@ export interface DoctorSkillReport {
     status: DoctorSkillStatus;
 }
 export interface DoctorReport {
-    kind: 'doctor';
+    reportKind: 'doctor';
     /** Schema version the JSON consumer can pin/check. Bumped on any
      *  user-visible change to this report's shape. See DOCTOR_REPORT_SCHEMA_VERSION. */
     schemaVersion: string;
@@ -154,4 +154,3 @@ export interface DoctorRunOptions {
     runHealthCheck?: boolean;
     effort?: 'low' | 'medium' | 'high' | 'xhigh' | 'max';
 }
-//# sourceMappingURL=doctor.d.ts.map

package/dist/types/doctor.js

@@ -1,7 +1,6 @@
 /** Bumped whenever DoctorReport schema changes in a way CI consumers should
  *  be able to detect. CI can pin/check this when parsing the JSON. */
-export const DOCTOR_REPORT_SCHEMA_VERSION = '1.0.0';
+export const DOCTOR_REPORT_SCHEMA_VERSION = '2.0.0';
 export function isComposerRule(r) {
-    return r.kind === 'composer';
+    return r.ruleKind === 'composer';
 }
-//# sourceMappingURL=doctor.js.map

package/dist/types/eval.d.ts

@@ -164,6 +164,8 @@ export interface VariantSpec {
     name: string;
     role: ExperimentRole;
     expr: string;
+    cwd?: string;
+    allowedSkills?: string[];
 }
 export interface EvalConfigVariant {
     name: string;
@@ -374,4 +376,3 @@ export interface Task {
     cwd: string | null;
     _sample: Sample;
 }
-//# sourceMappingURL=eval.d.ts.map

package/dist/types/eval.js

@@ -1,2 +1 @@
 export {};
-//# sourceMappingURL=eval.js.map

package/dist/types/executor.d.ts

@@ -142,10 +142,9 @@ export interface ExecutorRuntimeBinary {
 export interface ExecutorRuntimeFingerprint {
     executor: string;
     model: string;
-    kind: ExecutorRuntimeKind;
+    runtimeKind: ExecutorRuntimeKind;
     fingerprint: string;
     binary?: ExecutorRuntimeBinary;
     sdk?: ExecutorRuntimePackage;
     capabilities: ExecutorRuntimeCapabilities;
 }
-//# sourceMappingURL=executor.d.ts.map

package/dist/types/executor.js

@@ -1,2 +1 @@
 export {};
-//# sourceMappingURL=executor.js.map

package/dist/types/index.d.ts

@@ -9,4 +9,4 @@ export * from './diagnosis.js';
 export * from './dependencies.js';
 export * from './skill-index.js';
 export * from './observability.js';
-//# sourceMappingURL=index.d.ts.map
+export * from './managed.js';

package/dist/types/index.js

@@ -9,4 +9,4 @@ export * from './diagnosis.js';
 export * from './dependencies.js';
 export * from './skill-index.js';
 export * from './observability.js';
-//# sourceMappingURL=index.js.map
+export * from './managed.js';

package/dist/types/judge.d.ts

@@ -192,4 +192,3 @@ export interface GradeResult {
      *  → judgeCostUSD 累计是 lower-bound 而非真值。缺位 / true 当真值。 */
     judgeCostReportedByExecutor?: boolean;
 }
-//# sourceMappingURL=judge.d.ts.map

package/dist/types/judge.js

@@ -1,4 +1,3 @@
 export const FAILURE_MODES = [
     '工作流跳步', '硬编码值', '幻觉输出', '工具误用', '环境拦截', '误读约束', '其他',
 ];
-//# sourceMappingURL=judge.js.map

package/dist/types/managed.d.ts

@@ -0,0 +1,85 @@
+import type { ArtifactKind } from './eval.js';
+/**
+ * 受管 artifact 记录(managed record)——证据门控管理的最小持久单元。
+ *
+ * 设计原则(见 docs/specs/evidence-gated-management.md §5/§6):**只落盘事实与指针**。
+ * verdict / 可比性 / underpowered 是从 Report 推导出来的,不是这里的持久列;生命周期标签
+ * (installed / measurable / stale)也是读时推导(见 `deriveManagedState`),不持久 ——
+ * 持久一个会随源文件漂移而过期的标签等于让记录"撒谎"。需要"状态随时间"时持久化
+ * append-only 事件(installedAt、未来的 promote/reject/rollback 决定),而非可变标签。
+ *
+ * 判别字命名:外层 `recordKind` / 决定的 `decisionKind` 都是限定名;唯一的裸 `kind` 是
+ * `ManagedArtifactRecord.kind: ArtifactKind`——这正是裸 kind 该有的语义(知识 artifact 类型)。
+ */
+/** 一次分发落点的事实(把 skill 拷进某个 agent 工具 skill 目录)。 */
+export interface ManagedDistributionTarget {
+    /** 目标标签,如 'Claude Code' / 'Codex/AGENTS' / 'custom'。 */
+    label: string;
+    /** 实际写入的绝对路径({skillsDir}/{name} 或 {skillsDir}/{name}.md)。 */
+    path: string;
+    /** 拷到此目标的内容 hash(= 安装时源 hash)。 */
+    contentHash: string;
+    copiedAt: string;
+}
+/** 指向一份 Report 的引用——不是 verdict 本体。install 时为空,eval/promote 追加。 */
+export interface ManagedEvidenceRef {
+    reportId: string;
+    /** 该 report 测的是哪份内容(artifact contentHash)。读时只把与记录当前 contentHash 匹配的
+     *  evidence 算作当前有效证据——重装到新内容后旧证据保留供回滚,但不让新内容显得已测。 */
+    contentHash: string;
+    recordedAt: string;
+}
+export type ManagedDecisionKind = 'promote' | 'reject' | 'rollback';
+/** 一次人工管理决定。install 时为空,promote/reject/rollback 追加。 */
+export interface ManagedDecision {
+    decisionKind: ManagedDecisionKind;
+    actor: string;
+    decidedAt: string;
+    reason?: string;
+}
+export interface ManagedArtifactSource {
+    /** 源类型(限定判别字,非裸 kind)。`file`=本地路径;`git`=当前仓库的某个 ref。 */
+    sourceKind: 'file' | 'git';
+    /** 源身份,按 sourceKind 分义:
+     *   - file:本地重哈根(目录-skill 为根目录、文件-skill 为 .md),`hashArtifactSource(locator, isDirectorySkill)` 直接 round-trip;
+     *   - git:`git:<ref>:<name>`(非临时物化路径),drift 由 resolver 重物化重哈 —— mutable ref 给真实漂移、SHA 给不可变。 */
+    locator: string;
+    /** git 来源的 ref(file 源无)。 */
+    ref?: string;
+    /** 目录-skill(SKILL.md + assets)还是裸 .md 文件-skill。 */
+    isDirectorySkill: boolean;
+}
+/** 一条记录一个文件 `.omk/managed/<id>.json`,自带 recordKind + schemaVersion 便于单独迁移。 */
+export interface ManagedArtifactRecord {
+    recordKind: 'managed-artifact';
+    /** v2 起 source 带 sourceKind(多源化)。v1(仅 #211)无,按去兼容直接判脏丢弃、不迁移。 */
+    schemaVersion: 2;
+    /** 稳定身份 = hash(kind, name);源路径是可变属性、不进 id——挪动源文件不孤儿化记录。 */
+    id: string;
+    name: string;
+    kind: ArtifactKind;
+    source: ManagedArtifactSource;
+    /** 安装时的源内容 hash——drift baseline。 */
+    contentHash: string;
+    /** 首次纳管时间("under management since");per-target 时效看 distribution[].copiedAt。 */
+    installedAt: string;
+    distribution: ManagedDistributionTarget[];
+    evidence: ManagedEvidenceRef[];
+    decisions: ManagedDecision[];
+}
+/** 读时推导的生命周期标签——不持久(见文件头说明)。 */
+export type ManagedLifecycleLabel = 'discovered' | 'installed' | 'measurable' | 'stale';
+export interface DerivedManagedState {
+    label: ManagedLifecycleLabel;
+    /** 当前源 hash 与记录的 contentHash 不一致(或源已不在)。 */
+    drifted: boolean;
+    hasEvidence: boolean;
+}
+/** 推导所需、但不在记录里的当前事实(源文件现状、doctor/samples 状态)。 */
+export interface DeriveManagedStateInput {
+    record: ManagedArtifactRecord;
+    /** 源文件当前内容 hash;undefined = 源已不在。 */
+    currentContentHash?: string;
+    /** 是否已有 samples / doctor 通过(用于 measurable);install 不传。 */
+    hasSamplesOrDoctorPass?: boolean;
+}

package/dist/types/managed.js

@@ -0,0 +1 @@
+export {};

package/dist/types/observability.d.ts

@@ -43,7 +43,7 @@ export interface ObservationReviewStateEntry {
     snippet?: string;
 }
 export interface ObservationReviewState {
-    kind: 'observe-review-state';
+    reportKind: 'observe-review-state';
     schemaVersion: 1;
     updatedAt: string;
     entries: Record<string, ObservationReviewStateEntry>;
@@ -184,7 +184,7 @@ export interface ObservationSessionTimeRange {
     durationMs?: number;
 }
 export interface ObservationInboxReport {
-    kind: 'observe-inbox';
+    reportKind: 'observe-inbox';
     schemaVersion: 1;
     meta: {
         tracePath: string;
@@ -735,7 +735,7 @@ export interface ExperienceSkillSummary {
     relatedObservationIds: string[];
 }
 export interface ObservationExperienceReport {
-    kind: 'observe-experience';
+    reportKind: 'observe-experience';
     schemaVersion: 1;
     scope: 'evidence-only';
     generatedAt: string;
@@ -753,7 +753,7 @@ export interface ObservationExperienceReport {
 }
 export type ObservationRuntimeCheckStatus = 'passed' | 'attention' | 'manual_review';
 export interface ObservationRuntimeCheck {
-    kind: 'hardRule' | 'workflowNode';
+    nodeKind: 'hardRule' | 'workflowNode';
     id: string;
     title: string;
     expectation: string;
@@ -770,7 +770,7 @@ export interface SkillRuntimeEvidencePackRef extends Pick<ExperienceEvidenceRef,
 }
 export interface SkillRuntimeEvidencePackNode {
     nodeId: string;
-    kind: 'workflowNode' | 'hardRule';
+    nodeKind: 'workflowNode' | 'hardRule';
     title: string;
     expectation: string;
     deterministicStatus: ObservationRuntimeCheckStatus;
@@ -862,4 +862,3 @@ export interface ObservationSkillChain {
         evidencePack?: SkillRuntimeEvidencePack;
     };
 }
-//# sourceMappingURL=observability.d.ts.map

package/dist/types/observability.js

@@ -11,4 +11,3 @@
  *   - 跨层共享类型集中在 types/，单一来源
  */
 export {};
-//# sourceMappingURL=observability.js.map

package/dist/types/report.d.ts

@@ -341,7 +341,7 @@ export interface SampleSnapshot {
     tripwire?: boolean;
 }
 export interface EvaluationReport {
-    kind: 'evaluation';
+    reportKind: 'evaluation';
     id: string;
     meta: ReportMeta;
     summary: Record<string, VariantSummary>;
@@ -394,7 +394,7 @@ export interface BatchEvaluationItem {
     variance?: VarianceData;
 }
 export interface BatchEvaluationReport {
-    kind: 'batch-evaluation';
+    reportKind: 'batch-evaluation';
     id: string;
     mode: 'skill';
     meta: BatchEvaluationMeta;
@@ -557,4 +557,3 @@ export interface SaturationData {
         reason: string;
     }>;
 }
-//# sourceMappingURL=report.d.ts.map