oh-my-knowledge 0.52.3 → 0.54.0

package/dist/executors/{script.js → script/index.js}

@@ -1,7 +1,9 @@
-import { materializeForCliConfigDir } from '../eval-core/mocks-runtime.js';
-import { executorResultValidationError, normalizeExecResultToolIdentities, } from '../shared/executor-result.js';
-import { resolveScriptCommand } from './script-command.js';
-import { DEFAULT_TIMEOUT_MS, interruptedExecResult, spawnWithSigintPropagation, timeoutExecResult, } from './shared.js';
+import { materializeForCliConfigDir } from '../../eval-core/mocks-runtime.js';
+import { executorResultValidationError, normalizeExecResultToolIdentities, } from '../../shared/executor-result.js';
+import { resolveScriptCommand } from './command.js';
+import { DEFAULT_TIMEOUT_MS } from '../core/limits.js';
+import { interruptedExecResult, timeoutExecResult, } from '../core/runtime.js';
+import { spawnWithSigintPropagation } from '../core/subprocess.js';
 // script executor 由用户自定义,omk 无法保证它实现 skill 隔离。
 // 任何 allowedSkills(包括 [])下都 stderr 一次性 warn,不阻塞执行,
 // 让用户知道 strict-baseline / 显式 allowedSkills 在 script executor 下静默无效。

package/dist/grading/judge.d.ts

@@ -53,7 +53,7 @@ export declare function judgeId(config: JudgeConfig): string;
 export declare function computeJudgeAgreement(judgeScores: number[][]): JudgeAgreement;
 /**
  * Judge a single (output, rubric) pair with N judge models in parallel. Each judge
- * may use a different executor (e.g. claude:opus + openai-api:gpt-4o + gemini:pro). Each
+ * may use a different executor (e.g. claude:opus + openai-api:gpt-4o). Each
  * judge can also be repeated `judgeRepeat` times — final per-judge score is its mean.
  *
  * Returns: aggregate DimensionResult (score = mean across judges; this is the "consensus"

package/dist/grading/judge.js

@@ -342,7 +342,7 @@ export function computeJudgeAgreement(judgeScores) {
 }
 /**
  * Judge a single (output, rubric) pair with N judge models in parallel. Each judge
- * may use a different executor (e.g. claude:opus + openai-api:gpt-4o + gemini:pro). Each
+ * may use a different executor (e.g. claude:opus + openai-api:gpt-4o). Each
  * judge can also be repeated `judgeRepeat` times — final per-judge score is its mean.
  *
  * Returns: aggregate DimensionResult (score = mean across judges; this is the "consensus"

package/dist/observability/conversation-catalog.js

@@ -542,7 +542,7 @@ function trajectoryRevisionFromStat(sourceStat, status) {
     return `${sourceStat.size}:${sourceStat.mtimeMs}:${status}`;
 }
 function isTerminalTaskStatus(status) {
-    return status === 'completed' || status === 'aborted' || status === 'interrupted';
+    return status === 'completed' || status === 'failed' || status === 'aborted' || status === 'interrupted';
 }
 function secondsToMs(value) {
     const seconds = numberValue(value);

package/dist/observability/experience.js

@@ -1012,6 +1012,7 @@ function isExperienceTurnSummaryArray(value) {
         && isOptionalTimestamp(turn.startTimestamp)
         && isOptionalTimestamp(turn.endTimestamp)
         && (turn.status === 'completed'
+            || turn.status === 'failed'
             || turn.status === 'aborted'
             || turn.status === 'interrupted'
             || turn.status === 'open'
@@ -2621,6 +2622,9 @@ function timelineEventsFromTraceEvent(session, event, eventIndex) {
             memoryMode: event.memoryMode,
             historyMode: event.historyMode,
             contextWindowId: event.contextWindowId,
+            parentRunId: event.parentRunId,
+            delegationDepth: event.delegationDepth,
+            sourceOrigin: event.sourceOrigin,
             availableTools: event.availableTools,
             instructions: event.instructions,
             goal: event.goal,

package/dist/observability/inbox.d.ts

@@ -1,4 +1,5 @@
-import type { BuildObservationInboxReportOptions, ObservationEvidence, ObservationInboxItem, ObservationInboxReport, ObservationMessageRef, ObservationMessageWindow, ObservationSessionTimeRange, ObservationSeverityReasonCode, ObservationSignalSubtype, ObservationSignalType, ObservationSkillRollup, ObservationSourceKind } from '../types/index.js';
+import type { BuildObservationInboxReportOptions, ObservationEvidence, ObservationInboxItem, ObservationInboxReport, ObservationMessageRef, ObservationMessageWindow, ObservationSessionTimeRange, ObservationSeverityReasonCode, ObservationSignalSubtype, ObservationSignalType, ObservationSkillRollup, ObservationSourceKind, TraceIngestionSummary } from '../types/index.js';
+import { type TraceSession } from './trace-adapter.js';
 import { type PersistedObservationExperienceReport } from './experience.js';
 export type { BuildObservationInboxReportOptions, ObservationEvidence, ObservationInboxItem, ObservationInboxReport, ObservationMessageRef, ObservationMessageWindow, ObservationSessionTimeRange, ObservationSeverityReasonCode, ObservationSignalSubtype, ObservationSignalType, ObservationSkillRollup, ObservationSourceKind, };
 export type PersistedObservationInboxReport = Omit<ObservationInboxReport, 'experience'> & {
@@ -19,6 +20,8 @@ export declare function severityReasonCodeFor(item: Pick<ObservationInboxItem, '
  * `buildObserveDiagnosticsFromReport(report)` 写入 `report.diagnostics`。
  */
 export declare function buildObservationInboxReport(tracePath: string, options?: BuildObservationInboxReportOptions): ObservationInboxReport;
+/** Build the existing inbox artifact from an in-memory source-neutral corpus. */
+export declare function buildObservationInboxReportFromTraceSessions(tracePath: string, sessions: TraceSession[], ingestion: TraceIngestionSummary, options?: BuildObservationInboxReportOptions): ObservationInboxReport;
 export declare function aggregateInboxItems(items: ObservationInboxItem[]): ObservationInboxItem[];
 export declare function saveObservationInboxReport(report: ObservationInboxReport, outDir?: string): string;
 export declare function compactObservationInboxReport(report: ObservationInboxReport): PersistedObservationInboxReport;

package/dist/observability/inbox.js

@@ -5,7 +5,7 @@ import { OMK_HOME } from '../eval-core/default-dirs.js';
 import { isReportFileName, randomRunToken, reportFilePath } from '../eval-core/artifact-file-names.js';
 import { migrateLegacyReportFiles } from '../eval-core/report-file-migration.js';
 import { extractGapSignalsFromTrace } from '../analysis/gap-analyzer.js';
-import { loadTraceSessions, tracesToResultEntries, skillSegmentTimestampObserved, } from './trace-adapter.js';
+import { loadTraceSessions, segmentTraceBySkill, tracesToResultEntries, skillSegmentTimestampObserved, } from './trace-adapter.js';
 import { normalizeTraceTimestamp } from './trace-ir.js';
 import { isSearchToolCall, toolCallQuery } from '../shared/tool-search.js';
 import { isToolCallFailure, isToolCallSuccess } from '../shared/tool-call-status.js';
@@ -410,7 +410,12 @@ function skillSessionCountKey(segment) {
  * `buildObserveDiagnosticsFromReport(report)` 写入 `report.diagnostics`。
  */
 export function buildObservationInboxReport(tracePath, options = {}) {
-    const { sessions, segments, ingestion } = tracesToResultEntries(tracePath);
+    const { sessions, ingestion } = tracesToResultEntries(tracePath);
+    return buildObservationInboxReportFromTraceSessions(tracePath, sessions, ingestion, options);
+}
+/** Build the existing inbox artifact from an in-memory source-neutral corpus. */
+export function buildObservationInboxReportFromTraceSessions(tracePath, sessions, ingestion, options = {}) {
+    const segments = sessions.flatMap(segmentTraceBySkill);
     const skillSegments = segments.filter((segment) => segment.skillName !== 'general');
     const generatedAt = new Date().toISOString();
     const sessionTimeRanges = buildSessionTimeRanges(sessions);

package/dist/observability/trace-ir.d.ts

@@ -60,8 +60,8 @@ export interface TraceUsageEvent extends TraceEventBase {
     model?: string;
     inputTokens: number;
     outputTokens: number;
-    cacheReadTokens: number;
-    cacheCreationTokens: number;
+    cacheReadTokens?: number;
+    cacheCreationTokens?: number;
     reasoningTokens?: number;
 }
 export interface TraceModelActivityEvent extends TraceEventBase {
@@ -74,7 +74,7 @@ export interface TraceModelActivityEvent extends TraceEventBase {
 }
 export interface TraceLifecycleEvent extends TraceEventBase {
     eventKind: 'lifecycle';
-    phase: 'session_started' | 'session_ended' | 'turn_started' | 'turn_completed' | 'turn_aborted' | 'turn_interrupted';
+    phase: 'session_started' | 'session_ended' | 'turn_started' | 'turn_completed' | 'turn_failed' | 'turn_aborted' | 'turn_interrupted' | 'turn_ended_unknown' | 'step_started' | 'step_completed';
     reason?: string;
     durationMs?: number;
 }
@@ -105,6 +105,9 @@ export interface TraceRuntimeContextEvent extends TraceEventBase {
     memoryMode?: string;
     historyMode?: string;
     contextWindowId?: string;
+    parentRunId?: string;
+    delegationDepth?: number;
+    sourceOrigin?: string;
     availableTools?: string[];
     instructions?: string;
     goal?: string;

package/dist/observability/turn-index.js

@@ -1,8 +1,10 @@
 import { createHash } from 'node:crypto';
 const TERMINAL_LIFECYCLE_LABELS = new Set([
     'turn_completed',
+    'turn_failed',
     'turn_aborted',
     'turn_interrupted',
+    'turn_ended_unknown',
 ]);
 /** Reconstruct every observable task without consulting Skill attribution. */
 export function reconstructExperienceTurns(timeline) {
@@ -124,6 +126,8 @@ function turnStatus(events) {
     const labels = new Set(events.map((event) => event.label));
     if (labels.has('turn_completed'))
         return 'completed';
+    if (labels.has('turn_failed'))
+        return 'failed';
     if (labels.has('turn_aborted'))
         return 'aborted';
     if (labels.has('turn_interrupted'))

package/dist/renderer/conversation-renderer.js

@@ -54,8 +54,8 @@ export function renderConversationIndexPage(model, lang = DEFAULT_LANG) {
     const rows = conversations.map((conversation) => renderConversationRow(conversation, lang)).join('');
     const content = model.conversations.length > 0
         ? `<div class="conversation-list" role="list">${rows}</div>`
-        : `<div class="empty-state"><strong>${zh ? '还没有可浏览的 Codex 对话' : 'No Codex conversations yet'}</strong><span>${zh ? 'Studio 会直接读取 Codex 的本机会话索引，不需要先运行 observe。' : 'Studio reads the local Codex conversation index directly; observe is not required.'}</span></div>`;
-    return layout(zh ? 'Codex 对话' : 'Codex conversations', `
+        : `<div class="empty-state"><strong>${zh ? '还没有可浏览的对话' : 'No conversations yet'}</strong><span>${zh ? '从受支持的 runtime 摄取任务轨迹后，会话会显示在这里。' : 'Sessions appear here after a supported runtime supplies task trajectories.'}</span></div>`;
+    return layout(zh ? '对话' : 'Conversations', `
     <main class="conversation-page conversation-index-app" data-activity-revision="${e(activity.revision)}">
       <header class="conversation-app-head">
         <a class="conversation-app-brand" href="/${langQuery}">
@@ -67,7 +67,7 @@ export function renderConversationIndexPage(model, lang = DEFAULT_LANG) {
         </nav>
       </header>
       <div class="conversation-app-body">
-        <section class="conversation-browser" aria-label="${zh ? 'Codex 对话' : 'Codex conversations'}">
+        <section class="conversation-browser" aria-label="${zh ? '对话' : 'Conversations'}">
           <header class="conversation-toolbar">
             <div class="conversation-browser-title">
               <h1>${zh ? '对话' : 'Conversations'}</h1>
@@ -117,7 +117,7 @@ export function renderConversationDetailPage(conversation, lang = DEFAULT_LANG)
       <header class="conversation-page-head conversation-detail-head">
         <div>
           <a class="back-link" href="/conversations${langSuffix}">${zh ? '返回对话总览' : 'Back to conversations'}</a>
-          <p class="conversation-eyebrow">CODEX · ${e(shortThreadId(conversation.sourceThreadId))}</p>
+          <p class="conversation-eyebrow">${e(sourceLabel(conversation))} · ${e(shortThreadId(conversation.sourceThreadId))}</p>
           <h1>${renderSafeInlineMarkdown(conversation.title)}</h1>
           <div class="detail-meta">
             ${conversation.model ? `<span>${e(conversation.model)}</span>` : ''}
@@ -157,8 +157,8 @@ function renderConversationRow(conversation, lang) {
         ? '<span class="index-pending">—</span>'
         : `<span><b>${conversation.turnCount}</b>${zh ? '任务' : 'tasks'}</span>`;
     const activity = openTask
-        ? `<strong class="running-label"><i aria-hidden="true"></i>${zh ? '进行中' : 'Running'}</strong><span>${e(updated)} · ${e(conversation.model ?? 'Codex')}</span>`
-        : `<strong>${e(updated)}</strong><span>${e(conversation.model ?? 'Codex')}</span>`;
+        ? `<strong class="running-label"><i aria-hidden="true"></i>${zh ? '进行中' : 'Running'}</strong><span>${e(updated)} · ${e(conversation.model ?? sourceLabel(conversation))}</span>`
+        : `<strong>${e(updated)}</strong><span>${e(conversation.model ?? sourceLabel(conversation))}</span>`;
     const liveTaskLink = openTaskHref
         ? `<a class="live-task-link" href="${e(openTaskHref)}" aria-label="${zh ? '查看进行中任务的实时轨迹' : 'View the live trajectory for the running task'}"><i aria-hidden="true"></i>${zh ? '查看实时轨迹' : 'View live'}</a>`
         : '';
@@ -345,10 +345,24 @@ function compactPath(value) {
 function shortThreadId(value) {
     return value.length > 18 ? `${value.slice(0, 8)}…${value.slice(-6)}` : value;
 }
+function sourceLabel(conversation) {
+    if (conversation.sourceKind === 'dsh')
+        return 'DeepSeek Harness';
+    if (conversation.sourceKind === 'codex')
+        return 'Codex';
+    if (conversation.sourceKind === 'claude')
+        return 'Claude';
+    if (conversation.sourceKind === 'openclaw')
+        return 'OpenClaw';
+    if (conversation.sourceKind === 'markdown_log')
+        return 'Markdown';
+    return 'Runtime';
+}
 function statusLabel(status, lang) {
     const zh = lang === 'zh';
     const labels = {
         completed: ['已完成', 'Completed'],
+        failed: ['已失败', 'Failed'],
         aborted: ['已中止', 'Aborted'],
         interrupted: ['已打断', 'Interrupted'],
         open: ['进行中', 'Open'],

package/dist/renderer/html-renderer.js

@@ -57,7 +57,20 @@ function renderDebiasModeTag(modes, lang) {
 function pluralizeEn(count, singular, plural = `${singular}s`) {
     return `${count} ${count === 1 ? singular : plural}`;
 }
-function gatherRuntimeScopes(meta) {
+function executorDisplayName(executor, lang) {
+    if (executor === 'dsh-host') {
+        return lang === 'zh' ? 'DeepSeek Harness（宿主模式）' : 'DeepSeek Harness (host mode)';
+    }
+    return executor || 'unknown';
+}
+function executorModelDisplayName(executor, model, lang) {
+    const executorName = executorDisplayName(executor, lang);
+    const modelName = model || 'unknown';
+    if (executor === 'dsh-host')
+        return `${executorName} · ${modelName}`;
+    return `${executorName}:${modelName}`;
+}
+function gatherRuntimeScopes(meta, lang) {
     const scopes = [];
     if (meta.executorRuntimes && Object.keys(meta.executorRuntimes).length > 0) {
         Object.entries(meta.executorRuntimes)
@@ -75,14 +88,19 @@ function gatherRuntimeScopes(meta) {
             .slice()
             .sort((a, b) => `${a.executor}:${a.model}`.localeCompare(`${b.executor}:${b.model}`))
             .forEach((entry) => {
-            if (entry.runtime)
-                scopes.push({ role: 'judge', scope: `${entry.executor}:${entry.model}`, runtime: entry.runtime });
+            if (entry.runtime) {
+                scopes.push({
+                    role: 'judge',
+                    scope: executorModelDisplayName(entry.executor, entry.model, lang),
+                    runtime: entry.runtime,
+                });
+            }
         });
     }
     if (meta.diagnostic?.enabled && meta.diagnostic.runtime) {
         scopes.push({
             role: 'diagnostic',
-            scope: `${meta.diagnostic.executor}:${meta.diagnostic.model}`,
+            scope: executorModelDisplayName(meta.diagnostic.executor, meta.diagnostic.model, lang),
             runtime: meta.diagnostic.runtime,
         });
     }
@@ -105,6 +123,7 @@ function runtimeTooltip(runtime) {
         `cost=${runtime.capabilities.costUSD}`,
         `trace=${runtime.capabilities.trace}`,
         `skillIsolation=${runtime.capabilities.skillIsolation}`,
+        ...(runtime.auditability ? [`auditability=${runtime.auditability.status}`] : []),
         ...(runtime.binary?.contentHash
             ? [`contentHash=${runtime.binary.contentHash}`]
             : []),
@@ -115,7 +134,7 @@ function runtimeTooltip(runtime) {
 // fingerprint 重复 3 遍，扫读成本高。新版按 (fingerprint, versionText)
 // 分组，scope 合并到 tag 内 "适用于 ..." 后缀。
 function renderRuntimeFingerprintTags(meta, lang) {
-    const scopes = gatherRuntimeScopes(meta);
+    const scopes = gatherRuntimeScopes(meta, lang);
     if (scopes.length === 0)
         return '';
     const groups = new Map();
@@ -476,11 +495,11 @@ export function renderRunDetail(report, lang = DEFAULT_LANG, skillContext) {
         if (list.length === 0)
             return `<span class="meta-tag">${t('judge', lang)}: —</span>`;
         if (list.length === 1)
-            return `<span class="meta-tag">${t('judge', lang)}: ${e(`${list[0].executor}:${list[0].model}`)}</span>`;
-        return `<span class="meta-tag" title="${t('ensembleDesc', lang)}">${t('judgeModelsLabel', lang)}: ${list.map((j) => e(`${j.executor}:${j.model}`)).join(' · ')}</span>`;
+            return `<span class="meta-tag">${t('judge', lang)}: ${e(executorModelDisplayName(list[0].executor, list[0].model, lang))}</span>`;
+        return `<span class="meta-tag" title="${t('ensembleDesc', lang)}">${t('judgeModelsLabel', lang)}: ${list.map((j) => e(executorModelDisplayName(j.executor, j.model, lang))).join(' · ')}</span>`;
     })()}
     ${m.judgeRepeat && m.judgeRepeat > 1 ? `<span class="meta-tag" title="${t('judgeStddevDesc', lang)}">${t('judgeRepeatLabel', lang)}: ${m.judgeRepeat}</span>` : ''}
-    <span class="meta-tag">${t('executor', lang)}: ${e(m.executor || 'unknown')}</span>
+    <span class="meta-tag">${t('executor', lang)}: ${e(executorDisplayName(m.executor, lang))}</span>
     ${m.effort ? `<span class="meta-tag" title="${e(lang === 'zh' ? 'executor LLM 的扩展思考预算(--effort)。跨 effort 报告不可严格比较' : 'reasoning effort for executor LLM (--effort); reports across different efforts are not strictly comparable')}">effort: ${e(m.effort)}</span>` : ''}
     <span class="meta-tag"${execCostReported ? '' : ` title="${e(lang === 'zh' ? 'executor 不报 USD 成本(如 codex CLI),无法估算' : 'executor does not report USD cost (e.g. codex CLI); not measurable')}"`}>${t('cost', lang)}: ${fmtCost(totalExecCost, execCostReported)}</span>
     <span class="meta-tag"${totalCostReported ? '' : ` title="${e(costCompletenessTooltip(lang))}"`}>${totalCostLabel}: ${fmtKnownCost(m.totalCostUSD, totalCostReported)}</span>${processCostTag ? `
@@ -639,10 +658,10 @@ export function renderBatchEvaluationDetail(report, lang = DEFAULT_LANG) {
         if (list.length === 0)
             return `<span class="meta-tag">${t('judge', lang)}: —</span>`;
         if (list.length === 1)
-            return `<span class="meta-tag">${t('judge', lang)}: ${e(`${list[0].executor}:${list[0].model}`)}</span>`;
-        return `<span class="meta-tag" title="${t('ensembleDesc', lang)}">${t('judgeModelsLabel', lang)}: ${list.map((j) => e(`${j.executor}:${j.model}`)).join(' · ')}</span>`;
+            return `<span class="meta-tag">${t('judge', lang)}: ${e(executorModelDisplayName(list[0].executor, list[0].model, lang))}</span>`;
+        return `<span class="meta-tag" title="${t('ensembleDesc', lang)}">${t('judgeModelsLabel', lang)}: ${list.map((j) => e(executorModelDisplayName(j.executor, j.model, lang))).join(' · ')}</span>`;
     })()}
-      <span class="meta-tag">${t('executor', lang)}: ${e(m.executor || 'unknown')}</span>
+      <span class="meta-tag">${t('executor', lang)}: ${e(executorDisplayName(m.executor, lang))}</span>
       <span class="meta-tag"${totalCostReported ? '' : ` title="${e(costCompletenessTooltip(lang))}"`}>${t('totalCost', lang)}: ${fmtKnownCost(m.totalCostUSD, totalCostReported)}</span>
     </div>
     ${(() => {

package/dist/renderer/knowledge-debugger-renderer.js

@@ -1101,6 +1101,7 @@ export function renderKnowledgeDebuggerPage(model, lang = DEFAULT_LANG, options
         syncing: zh ? '同步中' : 'Syncing',
         reconnecting: zh ? '重连中' : 'Reconnecting',
         failed: zh ? '实时更新失败' : 'Live update failed',
+        taskFailed: zh ? '任务已失败' : 'Task failed',
         following: zh ? '跟随中' : 'Following',
         resume: zh ? '跟随最新' : 'Follow latest',
         pending: zh ? '查看更新' : 'View update',
@@ -1456,8 +1457,12 @@ function lifecycleEventLabel(label, lang) {
         session_ended: { zh: '会话结束', en: 'Session ended' },
         turn_started: { zh: '本轮开始', en: 'Turn started' },
         turn_completed: { zh: '本轮完成', en: 'Turn completed' },
+        turn_failed: { zh: '本轮失败', en: 'Turn failed' },
         turn_aborted: { zh: '本轮中止', en: 'Turn aborted' },
         turn_interrupted: { zh: '本轮被打断', en: 'Turn interrupted' },
+        turn_ended_unknown: { zh: '本轮结束状态未知', en: 'Turn ended with unknown status' },
+        step_started: { zh: '步骤开始', en: 'Step started' },
+        step_completed: { zh: '步骤完成', en: 'Step completed' },
     };
     return labels[label ?? '']?.[lang] ?? (label || (lang === 'zh' ? '运行状态变化' : 'Lifecycle event'));
 }
@@ -1466,7 +1471,7 @@ function lifecycleMilestoneTone(label) {
         return 'start';
     if (label === 'session_ended' || label === 'turn_completed')
         return 'end';
-    if (label === 'turn_aborted' || label === 'turn_interrupted')
+    if (label === 'turn_failed' || label === 'turn_aborted' || label === 'turn_interrupted')
         return 'warning';
     return 'neutral';
 }

package/dist/renderer/observation-inbox-renderer.js

@@ -188,8 +188,8 @@ export function renderObservationInboxPage(model, lang = DEFAULT_LANG) {
     </div>`;
     };
     const renderSourceBadge = (item) => {
-        const label = item.sourceKind === 'openclaw' ? 'OpenClaw' : item.sourceKind === 'codex' ? 'Codex' : item.sourceKind === 'markdown_log' ? 'Markdown log' : item.sourceKind === 'claude' ? 'Claude' : 'Unknown';
-        const color = item.sourceKind === 'openclaw' ? '#7c3aed' : item.sourceKind === 'codex' ? '#1677ff' : item.sourceKind === 'markdown_log' ? 'var(--green)' : item.sourceKind === 'claude' ? 'var(--accent)' : 'var(--text-muted)';
+        const label = item.sourceKind === 'dsh' ? 'DeepSeek Harness' : item.sourceKind === 'openclaw' ? 'OpenClaw' : item.sourceKind === 'codex' ? 'Codex' : item.sourceKind === 'markdown_log' ? 'Markdown log' : item.sourceKind === 'claude' ? 'Claude' : 'Unknown';
+        const color = item.sourceKind === 'dsh' ? '#0f766e' : item.sourceKind === 'openclaw' ? '#7c3aed' : item.sourceKind === 'codex' ? '#1677ff' : item.sourceKind === 'markdown_log' ? 'var(--green)' : item.sourceKind === 'claude' ? 'var(--accent)' : 'var(--text-muted)';
         return `<span title="调用日志来源：${e(label)}" style="display:inline-flex;margin-top:4px;padding:2px 6px;border-radius:999px;background:var(--bg-muted);color:${color};font-size:11px;font-weight:650">${e(label)}</span>`;
     };
     const confidenceHeaderHelp = '判断把握：OMK 对“这条 过程发现 是否需要处理/是否高风险/需关注”的规则判断有多确定。归属把握：OMK 把这条 过程发现 归到当前 skill 名下有多确定，例如明确调用 skill 通常更高。';

package/dist/renderer/trajectory-live.d.ts

@@ -4,6 +4,7 @@ export interface TrajectoryLiveLabels {
     syncing: string;
     reconnecting: string;
     failed: string;
+    taskFailed: string;
     following: string;
     resume: string;
     pending: string;

package/dist/renderer/trajectory-live.js

@@ -48,7 +48,9 @@ export function createTrajectoryLiveController(options) {
         const state = terminalState ?? (!followLatest && pendingRevision
             ? 'pending'
             : (followLatest ? 'following' : 'paused'));
-        const terminalLabel = terminalState ? labels[terminalState] : undefined;
+        const terminalLabel = terminalState
+            ? terminalState === 'failed' ? labels.taskFailed : labels[terminalState]
+            : undefined;
         followButton.dataset.state = state;
         followButton.dataset.following = String(followLatest);
         followButton.setAttribute('aria-pressed', String(followLatest));
@@ -212,6 +214,7 @@ export function createTrajectoryLiveController(options) {
             try {
                 const update = JSON.parse(event.data);
                 const explicitTerminal = update.status === 'completed'
+                    || update.status === 'failed'
                     || update.status === 'aborted'
                     || update.status === 'interrupted';
                 const streamTerminal = update.liveObservable === false || explicitTerminal;
@@ -226,7 +229,7 @@ export function createTrajectoryLiveController(options) {
                 if (!update.revision || update.revision === shell.dataset.liveRevision) {
                     const quiet = update.status === 'unknown';
                     const connectionLabel = terminalState
-                        ? labels[terminalState]
+                        ? terminalState === 'failed' ? labels.taskFailed : labels[terminalState]
                         : quiet ? labels.reconnecting : labels.live;
                     setConnectionState(terminalState ?? (quiet ? 'reconnecting' : 'live'), connectionLabel);
                     updateFollowControl();

package/dist/shared/trace-source-kind.js

@@ -1,6 +1,7 @@
 const TRACE_SOURCE_KINDS = new Set([
     'claude',
     'codex',
+    'dsh',
     'openclaw',
     'markdown_log',
     'unknown',

package/dist/types/executor.d.ts

@@ -88,7 +88,7 @@ export interface ExecutorInput {
      *  - claude-cli:物化为临时 settings.json + on-disk hook 脚本,跑完清理
      *  - script(自定义脚本):同样物化临时 settings,通过 env(OMK_MOCK_SETTINGS_FILE /
      *    OMK_MOCK_MCP_CONFIG_FILE / OMK_MOCKS_FILE)暴露给脚本;脚本负责消费该协议
-     *  - codex / codex-sdk / gemini / *-api:不支持,executor capability gate 会拒绝,
+     *  - codex / codex-sdk / *-api:不支持,executor capability gate 会拒绝,
      *    绝不静默忽略后把 mock_hit 记成模型失败 */
     mocks?: import('./eval.js').Mock[];
     /** 解析 mock.return_file 的相对路径锚点(默认 sample 文件所在目录)。 */
@@ -119,7 +119,13 @@ export interface ExecutorInput {
      */
     effort?: 'low' | 'medium' | 'high' | 'xhigh' | 'max';
 }
-export type ExecutorFn = (input: ExecutorInput) => Promise<ExecResult>;
+export type ExecutorRuntimeFingerprintResolver = (model: string, options?: {
+    skillDir?: string | null;
+}) => ExecutorRuntimeFingerprint;
+export type ExecutorFn = ((input: ExecutorInput) => Promise<ExecResult>) & {
+    /** Same-process hosts can report the runtime that actually owns execution. */
+    readonly runtimeFingerprint?: ExecutorRuntimeFingerprintResolver;
+};
 export interface ExecutorCache {
     get(key: string): ExecResult | null;
     set(key: string, value: ExecResult): void;
@@ -176,5 +182,10 @@ export interface ExecutorRuntimeFingerprint {
     fingerprint: string;
     binary?: ExecutorRuntimeBinary;
     sdk?: ExecutorRuntimePackage;
+    /** Whether the recorded fields cover the complete effective runtime composition. */
+    auditability?: {
+        status: 'complete' | 'partial';
+        reasons?: string[];
+    };
     capabilities: ExecutorRuntimeCapabilities;
 }

package/dist/types/judge.d.ts

@@ -1,9 +1,9 @@
 import type { ExecutorRuntimeFingerprint } from './executor.js';
 /** Single judge configuration: which executor to call and which model alias to pass. */
 export interface JudgeConfig {
-    /** Executor name (claude / openai / gemini / anthropic-api / openai-api / shell command). */
+    /** Executor name (claude / codex / anthropic-api / openai-api / shell command). */
     executor: string;
-    /** Model alias passed to the executor (e.g. "opus", "haiku", "gpt-4o", "gemini-2.0-pro"). */
+    /** Model alias passed to the executor (e.g. "opus", "haiku", "gpt-4o"). */
     model: string;
 }
 /** Persisted judge entry on Report.meta.judgeModels: judge config + runtime fingerprint of

package/dist/types/observability.d.ts

@@ -381,7 +381,7 @@ export type DebugKnowledgeAccessKind = 'injected' | 'read' | 'returned';
 export type TaskReplayStepKind = 'user_request' | 'user_message' | 'user_correction' | 'runtime_context' | 'skill_context' | 'tool_exchange' | 'unmatched_tool_result' | 'assistant_message' | 'model_activity' | 'lifecycle' | 'observation' | 'system_event';
 export type TaskReplayIntegrityCode = 'task_boundary_unavailable' | 'timeline_truncated' | 'malformed_records' | 'ignored_values' | 'unknown_events' | 'unmatched_tool_calls' | 'unmatched_tool_results' | 'missing_timestamps';
 export type TaskWindowBasis = 'turn_id' | 'turn_lifecycle' | 'user_message' | 'unresolved';
-export type ExperienceTurnStatus = 'completed' | 'aborted' | 'interrupted' | 'open' | 'unknown';
+export type ExperienceTurnStatus = 'completed' | 'failed' | 'aborted' | 'interrupted' | 'open' | 'unknown';
 /**
  * One user-visible task inside a source thread. `turnId` is the stable,
  * source-neutral identity used by Studio routes. `sourceTurnId` preserves the

package/dist/types/trace.d.ts

@@ -1,2 +1,2 @@
 /** Stable source identity shared by execution evidence and Trace IR. */
-export type TraceSourceKind = 'claude' | 'codex' | 'openclaw' | 'markdown_log' | 'unknown';
+export type TraceSourceKind = 'claude' | 'codex' | 'dsh' | 'openclaw' | 'markdown_log' | 'unknown';

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "oh-my-knowledge",
-  "version": "0.52.3",
+  "version": "0.54.0",
   "packageManager": "yarn@4.16.0",
   "description": "OMK — Observe. Measure. Know. Evidence-backed knowledge changes for AI applications.",
   "type": "module",
@@ -13,6 +13,11 @@
   "files": [
     "dist/"
   ],
+  "dsh": {
+    "bundle": {
+      "patch": "./dist/dsh-plugin/cordis.patch.yml"
+    }
+  },
   "scripts": {
     "clean": "rm -rf dist dist-scripts node_modules/.cache/omk",
     "build": "run-s build:src build:scripts build:assets",
@@ -28,6 +33,7 @@
     "lint": "eslint 'src/**/*.ts' 'test/**/*.ts' --cache --cache-location node_modules/.cache/eslint/ --max-warnings 0",
     "lint-staged": "lint-staged",
     "test": "vitest run",
+    "test:profile": "yarn build && node dist-scripts/test-profile.js",
     "ci": "run-s lint typecheck build build:docs:check test",
     "prepublishOnly": "run-s clean build",
     "prepare": "husky"
@@ -79,6 +85,10 @@
     "release-evidence",
     "llm-observability",
     "prompt-engineering",
+    "deepseek-harness",
+    "dsh-plugin",
+    "dsh",
+    "cordis",
     "codex",
     "codex-cli",
     "openai",
@@ -98,12 +108,10 @@
   "author": "lizhiyao",
   "license": "MIT",
   "dependencies": {
-    "@anthropic-ai/claude-agent-sdk": "^0.3.143",
-    "@anthropic-ai/sdk": "^0.117.1",
+    "@anthropic-ai/sdk": "^0.120.0",
     "@inquirer/prompts": "^8.4.3",
     "@modelcontextprotocol/sdk": "^1.30.0",
     "@oclif/core": "^4",
-    "@openai/codex-sdk": "0.147.0",
     "ajv": "^8.18.0",
     "chart.js": "^4.5.1",
     "es-module-lexer": "^2.0.0",
@@ -111,6 +119,18 @@
     "simple-statistics": "^7.8.9",
     "zod": "^4.4.3"
   },
+  "peerDependencies": {
+    "@anthropic-ai/claude-agent-sdk": "^0.3.143",
+    "@openai/codex-sdk": "^0.149.0"
+  },
+  "peerDependenciesMeta": {
+    "@anthropic-ai/claude-agent-sdk": {
+      "optional": true
+    },
+    "@openai/codex-sdk": {
+      "optional": true
+    }
+  },
   "publishConfig": {
     "registry": "https://registry.npmjs.org"
   },
@@ -124,6 +144,6 @@
     "typescript": "^6.0.2",
     "typescript-eslint": "^8.58.0",
     "vitepress": "^1.6.4",
-    "vitest": "4.1.10"
+    "vitest": "4.1.11"
   }
 }

package/dist/executors/claude-protocol.d.ts

@@ -1,28 +0,0 @@
-import type { ExecResult } from '../types/index.js';
-import type { ClaudeSdkBaseMessage, ClaudeSdkResultMessage } from './shared.js';
-export interface ClaudeSdkMeasurements {
-    durationMs: number;
-    durationApiMs: number;
-    inputTokens: number;
-    outputTokens: number;
-    cacheReadTokens: number;
-    cacheCreationTokens: number;
-    costUSD: number;
-    numTurns: number;
-}
-export declare function normalizeClaudeSdkMeasurements(result: ClaudeSdkResultMessage): ClaudeSdkMeasurements | {
-    error: string;
-};
-export interface ClaudeStreamParseResult {
-    messages: ClaudeSdkBaseMessage[];
-    malformedLineCount: number;
-}
-export declare function parseClaudeStreamJson(stdout: string): ClaudeStreamParseResult;
-export declare function buildClaudeResult(options: {
-    messages: ClaudeSdkBaseMessage[];
-    wallClockDurationMs: number;
-    source: 'claude stream-json' | 'claude-sdk';
-    malformedLineCount?: number;
-    forcedError?: string;
-    messageTimestamps?: number[];
-}): ExecResult;

package/dist/executors/claude-sdk-trace.d.ts

@@ -1,9 +0,0 @@
-import type { ToolCallInfo, TurnInfo } from '../types/index.js';
-import type { ClaudeSdkBaseMessage } from './shared.js';
-export declare function isClaudeSdkResultMessage(message: ClaudeSdkBaseMessage): boolean;
-export declare function extractAgentTrace(messages: ClaudeSdkBaseMessage[], timestamps?: number[]): {
-    turns: TurnInfo[];
-    toolCalls: ToolCallInfo[];
-    fullNumTurns: number;
-    numSubAgents: number;
-};

package/dist/executors/codex-protocol.d.ts

@@ -1,24 +0,0 @@
-import type { ExecResult } from '../types/index.js';
-import type { CodexEvent } from './shared.js';
-export declare function extractCodexUsage(events: CodexEvent[]): {
-    input: number;
-    cached: number;
-    output: number;
-};
-/**
- * Match @openai/codex-sdk's `finalResponse`: the latest completed
- * agent_message is the answer. Earlier messages remain available in `turns`.
- */
-export declare function extractCodexFinalOutput(events: CodexEvent[]): string;
-export declare function extractCodexProtocolError(events: CodexEvent[]): string | undefined;
-export declare function validateCodexProtocol(events: CodexEvent[]): string | undefined;
-export declare function extractCodexStopReason(events: CodexEvent[]): string;
-export declare function sumCodexElapsed(resultEvents: CodexEvent[], wallClock: number): number;
-export interface BuildCodexResultOptions {
-    events: CodexEvent[];
-    wallClockDurationMs: number;
-    source: 'codex --json' | 'codex-sdk';
-    malformedLineCount?: number;
-    forcedError?: string;
-}
-export declare function buildCodexResult({ events, wallClockDurationMs, source, malformedLineCount, forcedError, }: BuildCodexResultOptions): ExecResult;

package/dist/executors/gemini.d.ts

@@ -1,2 +0,0 @@
-import type { ExecResult, ExecutorInput } from '../types/index.js';
-export declare function geminiExecutor({ model, system, prompt, timeoutMs }: ExecutorInput): Promise<ExecResult>;