oh-my-knowledge 0.30.0 → 0.31.0

package/dist/src/observability/experience.js

@@ -1,9 +1,26 @@
 import { createHash } from 'node:crypto';
+import { existsSync, readFileSync } from 'node:fs';
+import { homedir } from 'node:os';
+import { join } from 'node:path';
 import { buildExperienceProblemPatterns, mergeExperienceProblemPatterns, } from './problem-patterns.js';
-import { observationMetricAnnotationVerdict } from './review-state.js';
+import { observationMetricAnnotationVerdict, observationReviewStateKey } from './review-state.js';
 import { extractCommandEnvelopeText, stripCommandEnvelopeText } from './trace-attribution.js';
-import { hasAssistantDeliverySignalText, hasUserHardRuleText, isScheduledTaskPromptText } from './text-signals.js';
+import { hasAssistantDeliverableArtifactText, hasAssistantDeliverySignalText, hasUserHardRuleText, isAssistantProgressUpdateText, isAssistantProtocolReplyText, isRuntimeProtocolPromptText, isSyntheticUserMessageText, isToolResultFailureText, isUserInteractionMetricText } from './text-signals.js';
 import { durationMsBetween } from '../shared/time.js';
+import { parseSkillFrontmatter, validateSkillHardRules, validateSkillWorkflows } from '../shared/hard-rules.js';
+export function aggregateExperienceChecklistItemStatus(statuses) {
+    if (statuses.includes('degraded'))
+        return 'degraded';
+    if (statuses.includes('failed'))
+        return 'failed';
+    if (statuses.includes('unknown'))
+        return 'unknown';
+    if (statuses.includes('not_declared'))
+        return 'not_declared';
+    if (statuses.includes('passed'))
+        return 'passed';
+    return 'not_applicable';
+}
 const TEXT_BOUNDARY_CHARS = new Set([
     ' ', '\n', '\r', '\t',
     ',', '，', ';', '；', '、', '.', '。', '!', '！', '?', '？', ':', '：',
@@ -75,6 +92,22 @@ const NEGATIVE_FEEDBACK_TERMS = [
     '菜啊',
 ];
 const BOUNDED_NEGATIVE_FEEDBACK_TERMS = ['失败', '垃圾', '菜'];
+const ADDITIONAL_NEGATIVE_FEEDBACK_TERMS = [
+    '不对',
+    '错了',
+    '有误',
+    '不可以',
+    '不是这个',
+    '不是我要的',
+    '跑偏',
+    '绕路',
+    '你没懂',
+    '没懂',
+    '重做',
+    '重新做',
+    '重来',
+    '不符合预期',
+];
 const POSITIVE_FEEDBACK_TERMS = [
     '非常好',
     '很好',
@@ -101,18 +134,22 @@ const POSITIVE_FEEDBACK_TERMS = [
     'excellent',
     'awesome',
 ];
-const USER_INTERRUPTION_RE = /\[Request interrupted by user(?: for tool use)?\]|interrupted by user|用户中断/i;
+const USER_INTERRUPTION_RE = /\[Request interrupted by user(?: for tool use)?\]|interrupted by user|用户中断|停止任务|停一下|先别|别动|等一下|等下|等等|取消(?:任务|执行)?|先暂停|暂停一下/i;
 const TIMELINE_PREVIEW_EVENT_LIMIT = 240;
 const ZERO_INDICATORS = {
     userMessageCount: 0,
     userFollowUpCount: 0,
     userCorrectionCount: 0,
     userInterruptionCount: 0,
+    sessionInterruptedCount: 0,
     negativeFeedbackCount: 0,
     positiveFeedbackCount: 0,
     userGoalShiftCount: 0,
     hardRuleTextHitCount: 0,
     assistantDeliverySignalCount: 0,
+    deliverableArtifactSignalCount: 0,
+    selfCorrectionCount: 0,
+    repeatedExecutionCount: 0,
     toolCallCount: 0,
     toolFailureCount: 0,
     highObservationCount: 0,
@@ -123,6 +160,7 @@ const ZERO_INDICATORS = {
 export function buildObservationExperienceReport(input) {
     const sessionsBySourceTrace = new Map(input.sessions.map((session) => [session.sourcePath, session]));
     const sessionGroupsById = groupSessionsByLogicalId(input.sessions);
+    const sessionGroupsByKey = groupSessionsByExperienceKey(input.sessions);
     const goalSlices = [];
     const invocations = [];
     for (const segment of input.segments) {
@@ -130,19 +168,22 @@ export function buildObservationExperienceReport(input) {
             continue;
         const session = sessionsBySourceTrace.get(segment.sourceTrace ?? '') ?? sessionGroupsById.get(segment.sessionId)?.[0];
         const sourceTrace = segment.sourceTrace ?? session?.sourcePath ?? '';
+        const sessionGroupKey = session ? experienceSessionGroupKey(session) : `trace:${sourceTrace || segment.sessionId}`;
         const relatedItems = relatedObservationItems(segment, input.items);
         const bounds = session ? segmentRecordBounds(session, segment) : { start: 0, end: 0 };
         const timeline = session ? buildTimeline(session, bounds.start, bounds.end) : [];
+        const metricScopeId = hashParts('session', segment.skillName, sessionGroupKey);
         const userRefs = timeline.filter((event) => event.kind === 'user_message');
-        const indicators = indicatorsForSegment(segment, relatedItems, timeline, input.reviewState);
+        const indicators = indicatorsForSegment(segment, relatedItems, timeline, metricScopeId, input.reviewState);
         const observationRefs = relatedItems.map(observationEvidenceRef);
         const evidenceChain = evidenceChainForTimeline(timeline, observationRefs);
-        const ruleFindings = ruleFindingsForEvidence(indicators, timeline, observationRefs, evidenceChain, input.reviewState);
+        const ruleFindings = ruleFindingsForEvidence(indicators, timeline, observationRefs, evidenceChain, metricScopeId, input.reviewState);
         const assistiveInference = assistiveInferenceForEvidence(indicators, evidenceChain, ruleFindings);
         const problemPatterns = buildExperienceProblemPatterns({
             skillName: segment.skillName,
             sessionId: segment.sessionId,
             timeline,
+            metricScopeId,
             reviewState: input.reviewState,
         });
         const hasGoalShift = userRefs.some((ref) => hasUserGoalShiftSignal(ref.snippet ?? ''));
@@ -167,6 +208,7 @@ export function buildObservationExperienceReport(input) {
             id: invocationId,
             skillName: segment.skillName,
             sessionId: segment.sessionId,
+            sessionGroupKey,
             sourceTrace,
             sourceKind: segment.sourceKind ?? sourceKindForPath(sourceTrace),
             entrypoint: session ? session.entrypoint ?? inferEntrypointFromRecords(session) : undefined,
@@ -198,7 +240,7 @@ export function buildObservationExperienceReport(input) {
             timeline,
         });
     }
-    const sessions = summarizeExperienceSessions(invocations, sessionGroupsById);
+    const sessions = summarizeExperienceSessions(invocations, sessionGroupsByKey, input.generatedAt, input.reviewState);
     const skills = summarizeExperienceSkills(sessions, invocations);
     return {
         kind: 'observe-experience',
@@ -226,6 +268,13 @@ function relatedObservationItems(segment, items) {
 function logicalSessionId(session) {
     return session.sessionGroupId ?? session.sessionId;
 }
+function experienceSessionGroupKey(session) {
+    const logicalId = logicalSessionId(session);
+    if (session.traceRole && session.traceRole !== 'standalone' && session.sessionGroupPath) {
+        return `group:${session.sessionGroupPath}\u0000${logicalId}`;
+    }
+    return `trace:${session.sourcePath}`;
+}
 function groupSessionsByLogicalId(sessions) {
     const groups = new Map();
     for (const session of sessions) {
@@ -239,6 +288,19 @@ function groupSessionsByLogicalId(sessions) {
     }
     return groups;
 }
+function groupSessionsByExperienceKey(sessions) {
+    const groups = new Map();
+    for (const session of sessions) {
+        const key = experienceSessionGroupKey(session);
+        const group = groups.get(key) ?? [];
+        group.push(session);
+        groups.set(key, group);
+    }
+    for (const [key, group] of groups.entries()) {
+        groups.set(key, group.sort(compareSessionsForTimeline));
+    }
+    return groups;
+}
 function compareSessionsForTimeline(a, b) {
     const roleRank = (session) => session.traceRole === 'main' ? 0 : session.traceRole === 'standalone' ? 1 : 2;
     const rank = roleRank(a) - roleRank(b);
@@ -331,6 +393,30 @@ function isAssistantDeliveryEvent(event) {
     const text = event.fullText ?? event.snippet ?? '';
     return hasAssistantDeliverySignalText(text);
 }
+function isAssistantDeliverableArtifactEvent(event) {
+    if (event.kind !== 'assistant_message')
+        return false;
+    const text = event.fullText ?? event.snippet ?? '';
+    return hasAssistantDeliverableArtifactText(text);
+}
+function isAssistantProgressUpdateEvent(event) {
+    if (event.kind !== 'assistant_message')
+        return false;
+    const text = event.fullText ?? event.snippet ?? '';
+    return isAssistantProgressUpdateText(text);
+}
+function hasSelfCorrectionSignal(event) {
+    if (event.kind !== 'assistant_message')
+        return false;
+    const text = event.fullText ?? event.snippet ?? '';
+    return /刚才.*(?:不对|错了|有误)|发现.*(?:不对|错了|问题|遗漏)|重新(?:检查|分析|执行|生成|整理|补(?:充)?(?:结论|答案|回答))|改用|换成|修正|我再(?:检查|重新|看)|跑偏|偏(?:题|了|向)|没(?:有)?(?:完整|完全)?(?:回答|覆盖)|漏(?:了|掉)|不完整|我刚追问了一版|按原问题(?:重新)?(?:补|回|答)|\b(?:recheck|retry|rerun|mistake|wrong)\b/i.test(text);
+}
+function hasRepeatedExecutionSignal(event) {
+    if (event.kind !== 'assistant_message' && event.kind !== 'tool_use')
+        return false;
+    const text = `${event.label ?? ''} ${event.toolName ?? ''} ${event.fullText ?? event.snippet ?? ''}`;
+    return /重复(?:执行|尝试|读取|搜索|调用)|再次(?:执行|读取|搜索|调用)|重新(?:执行|读取|搜索|调用|跑|运行)|再(?:执行|读取|搜索|调用|跑)一遍|重试|\b(?:retry|rerun)\b/i.test(text);
+}
 function buildTimeline(session, start, end) {
     return buildTimelineWindow(session, start, end).slice(0, TIMELINE_PREVIEW_EVENT_LIMIT);
 }
@@ -366,7 +452,36 @@ function timelineEventsFromRecord(session, record, messageIndex) {
         const assistant = rec;
         return assistantEvents(assistant, base, messageIndex);
     }
-    return [];
+    return sessionEventFromRecord(rec, base, messageIndex);
+}
+function sessionEventFromRecord(record, base, messageIndex) {
+    const type = typeof record.type === 'string' ? record.type : '';
+    const rawText = safeRecordText(record);
+    if (!isSessionBoundaryType(type) && !hasAssistantTurnFailedText(rawText))
+        return [];
+    return [timelineEvent({
+            ...base,
+            kind: 'runtime_context',
+            role: 'other',
+            order: messageIndex * 10,
+            snippet: snippet(rawText || type, 700),
+            fullText: fullText(rawText || type),
+            label: type || 'session event',
+        })];
+}
+function isSessionBoundaryType(type) {
+    return /^session[._-](?:started|ended)$/i.test(type);
+}
+function safeRecordText(record) {
+    try {
+        return JSON.stringify(record);
+    }
+    catch {
+        return String(record ?? '');
+    }
+}
+function hasAssistantTurnFailedText(value) {
+    return /\[assistant turn failed\]|assistant turn failed|assistant_turn_failed|turn failed/i.test(value);
 }
 function userEvents(record, base, messageIndex) {
     const content = record.message.content;
@@ -387,16 +502,17 @@ function userEvents(record, base, messageIndex) {
         else if (part.type === 'tool_result') {
             const full = fullText(part.content);
             const text = snippet(part.content, 900);
+            const failed = part.is_error === true || isToolResultFailureText(part.content);
             events.push(timelineEvent({
                 ...base,
                 kind: 'tool_result',
                 role: 'tool',
                 order: messageIndex * 10 + 5 + resultIndex,
                 toolUseId: part.tool_use_id,
-                isError: part.is_error === true,
+                isError: failed,
                 snippet: text,
                 fullText: full,
-                label: part.is_error === true ? 'tool result error' : 'tool result',
+                label: failed ? 'tool result error' : 'tool result',
             }));
             resultIndex += 1;
         }
@@ -425,7 +541,7 @@ function userTextEventsFromText(record, rawText, base, messageIndex, offset) {
         events.push(timelineEvent({
             ...base,
             kind,
-            role: kind === 'user_message' ? 'user' : 'tool',
+            role: kind === 'user_message' ? 'user' : kind === 'synthetic_user_event' ? 'other' : 'tool',
             order: messageIndex * 10 + offset + (commandEnvelope ? 1 : 0),
             snippet: text,
             fullText: full,
@@ -460,14 +576,15 @@ function assistantEvents(record, base, messageIndex) {
     const rawText = textParts.join('\n');
     const text = snippet(rawText, 700);
     if (text) {
+        const protocolReply = isAssistantProtocolReplyText(rawText);
         events.unshift(timelineEvent({
             ...base,
-            kind: 'assistant_message',
+            kind: protocolReply ? 'runtime_context' : 'assistant_message',
             role: 'assistant',
             order: messageIndex * 10,
             snippet: text,
             fullText: fullText(rawText),
-            label: 'assistant message',
+            label: protocolReply ? 'assistant protocol reply' : 'assistant message',
         }));
     }
     return events;
@@ -483,6 +600,8 @@ function userTextEventKind(record, text) {
         return 'skill_context';
     if (isRuntimeContextRecord(record, text))
         return 'runtime_context';
+    if (isSyntheticUserMessageText(text))
+        return 'synthetic_user_event';
     return 'user_message';
 }
 function userTextEventLabel(kind) {
@@ -490,6 +609,8 @@ function userTextEventLabel(kind) {
         return 'skill context';
     if (kind === 'runtime_context')
         return 'runtime context';
+    if (kind === 'synthetic_user_event')
+        return 'synthetic user event';
     return 'user message';
 }
 function observationEvidenceRef(item) {
@@ -526,13 +647,13 @@ function evidenceRefFromTimeline(event) {
 }
 function evidenceChainForTimeline(timeline, observationRefs) {
     const events = uniqueTimelineEvents(timeline).sort(compareTimelineEvents);
-    const userEvents = events.filter((event) => event.kind === 'user_message');
+    const userEvents = events.filter((event) => event.kind === 'user_message' && !isSyntheticUserMessageText(event.snippet ?? ''));
     const runtimeEvents = events.filter((event) => event.kind === 'runtime_context');
     const skillEvents = events.filter((event) => event.kind === 'skill_context');
     const assistantEvents = events.filter((event) => event.kind === 'assistant_message');
     const toolUseEvents = events.filter((event) => event.kind === 'tool_use');
     const toolResultEvents = events.filter((event) => event.kind === 'tool_result');
-    const toolFailureEvents = toolResultEvents.filter((event) => event.isError === true);
+    const toolFailureEvents = toolResultEvents.filter(isToolFailureEvent);
     return {
         userMessageCount: userEvents.length,
         runtimeContextCount: runtimeEvents.length,
@@ -550,10 +671,10 @@ function evidenceChainForTimeline(timeline, observationRefs) {
         lastAssistantMessage: assistantEvents.at(-1) ? evidenceRefFromTimeline(assistantEvents.at(-1)) : undefined,
     };
 }
-function ruleFindingsForEvidence(indicators, timeline, observationRefs, evidenceChain, reviewState) {
+function ruleFindingsForEvidence(indicators, timeline, observationRefs, evidenceChain, metricScopeId, reviewState) {
     const events = uniqueTimelineEvents(timeline);
     const userEvents = events.filter((event) => event.kind === 'user_message');
-    const metricUserEvents = userEvents.filter((event) => !isScheduledTaskPromptText(event.snippet ?? ''));
+    const metricUserEvents = userEvents.filter((event) => isUserInteractionMetricText(event.snippet ?? ''));
     const refs = (matches) => matches.slice(0, 5).map(evidenceRefFromTimeline);
     const findings = [];
     const push = (code, level, count, evidenceRefs = []) => {
@@ -562,16 +683,17 @@ function ruleFindingsForEvidence(indicators, timeline, observationRefs, evidence
         findings.push({ code, level, count, evidenceRefs: uniqueEvidenceRefs(evidenceRefs).slice(0, 5) });
     };
     push('high_observation_seen', 'attention', indicators.highObservationCount, observationRefs);
-    push('user_correction_seen', 'attention', indicators.userCorrectionCount, refs(metricUserEvents.filter((event) => metricIsActive(event, 'user_correction', hasUserCorrectionSignal(event.snippet ?? ''), reviewState))));
-    push('user_interruption_seen', 'attention', indicators.userInterruptionCount, refs(metricUserEvents.filter((event) => metricIsActive(event, 'user_interruption', USER_INTERRUPTION_RE.test(event.snippet ?? ''), reviewState))));
+    push('user_correction_seen', 'attention', indicators.userCorrectionCount, refs(metricUserEvents.filter((event) => metricIsActive(event, 'user_correction', hasUserCorrectionSignal(event.snippet ?? ''), reviewState, metricScopeId))));
+    push('user_interruption_seen', 'attention', indicators.userInterruptionCount, refs(metricUserEvents.filter((event) => metricIsActive(event, 'user_interruption', USER_INTERRUPTION_RE.test(event.snippet ?? ''), reviewState, metricScopeId))));
+    push('session_interrupted_seen', 'attention', indicators.sessionInterruptedCount, refs(sessionInterruptedEvents(events)));
     push('negative_feedback_seen', 'attention', indicators.negativeFeedbackCount, refs(metricUserEvents.filter((event) => metricIsActive(event, 'negative_feedback', hasNegativeFeedbackSignal(event.snippet ?? ''), reviewState))));
-    push('tool_failure_seen', 'sample', indicators.toolFailureCount, refs(events.filter((event) => event.kind === 'tool_result' && event.isError === true)));
+    push('tool_failure_seen', 'sample', indicators.toolFailureCount, refs(events.filter(isToolFailureEvent)));
     push('medium_observation_seen', 'sample', indicators.mediumObservationCount, observationRefs);
     push('hedging_seen', 'sample', indicators.hedgingCount, observationRefs);
     push('explicit_marker_seen', 'sample', indicators.explicitMarkerCount, observationRefs);
-    push('hard_rule_seen', 'sample', indicators.hardRuleTextHitCount, refs(metricUserEvents.filter((event) => metricIsActive(event, 'hard_rule', hasUserHardRuleText(event.snippet ?? ''), reviewState))));
+    push('hard_rule_seen', 'sample', indicators.hardRuleTextHitCount, refs(metricUserEvents.filter((event) => metricIsActive(event, 'hard_rule', hasUserHardRuleText(event.snippet ?? ''), reviewState, metricScopeId))));
     push('positive_feedback_seen', 'normal', indicators.positiveFeedbackCount, refs(metricUserEvents.filter((event) => metricIsActive(event, 'positive_feedback', hasPositiveFeedbackSignal(event.snippet ?? ''), reviewState))));
-    push('user_goal_shift_seen', 'normal', indicators.userGoalShiftCount, refs(metricUserEvents.filter((event) => metricIsActive(event, 'user_goal_shift', hasUserGoalShiftSignal(event.snippet ?? ''), reviewState))));
+    push('user_goal_shift_seen', 'normal', indicators.userGoalShiftCount, refs(metricUserEvents.filter((event) => metricIsActive(event, 'user_goal_shift', hasUserGoalShiftSignal(event.snippet ?? ''), reviewState, metricScopeId))));
     push('runtime_context_excluded', 'normal', evidenceChain.runtimeContextCount, evidenceChain.firstRuntimeContext ? [evidenceChain.firstRuntimeContext] : []);
     push('skill_context_excluded', 'normal', evidenceChain.skillContextCount, evidenceChain.firstSkillContext ? [evidenceChain.firstSkillContext] : []);
     if (findings.filter((finding) => finding.level !== 'normal').length === 0) {
@@ -642,48 +764,73 @@ function assistiveInferenceForEvidence(indicators, evidenceChain, ruleFindings)
         evidenceRefs,
     };
 }
-function indicatorsForSegment(segment, relatedItems, timeline, reviewState) {
+function indicatorsForSegment(segment, relatedItems, timeline, metricScopeId, reviewState) {
     const userRefs = timeline.filter((event) => event.kind === 'user_message');
-    const humanUserRefs = userRefs.filter((ref) => Boolean(ref.snippet));
-    const interactionUserRefs = humanUserRefs.filter((ref) => !isScheduledTaskPromptText(ref.snippet ?? ''));
+    const humanUserRefs = userRefs.filter((ref) => Boolean(ref.snippet) && !isSyntheticUserMessageText(ref.snippet ?? ''));
+    const interactionUserRefs = humanUserRefs.filter((ref) => isUserInteractionMetricText(ref.snippet ?? ''));
     return {
         userMessageCount: humanUserRefs.length,
-        userFollowUpCount: interactionUserRefs.reduce((sum, ref, index) => sum + (metricIsActive(ref, 'user_follow_up', index > 0, reviewState) ? 1 : 0), 0),
-        userCorrectionCount: interactionUserRefs.reduce((sum, ref) => sum + metricCount(ref, 'user_correction', findUserCorrectionMatches(ref.snippet ?? '').length, reviewState), 0),
-        userInterruptionCount: interactionUserRefs.reduce((sum, ref) => sum + (metricIsActive(ref, 'user_interruption', USER_INTERRUPTION_RE.test(ref.snippet ?? ''), reviewState) ? 1 : 0), 0),
+        userFollowUpCount: interactionUserRefs.reduce((sum, ref, index) => sum + (metricIsActive(ref, 'user_follow_up', index > 0 && !hasUserGoalShiftSignal(ref.snippet ?? ''), reviewState, metricScopeId) ? 1 : 0), 0),
+        userCorrectionCount: interactionUserRefs.reduce((sum, ref) => sum + metricCount(ref, 'user_correction', findUserCorrectionMatches(ref.snippet ?? '').length, reviewState, metricScopeId), 0),
+        userInterruptionCount: interactionUserRefs.reduce((sum, ref) => sum + (metricIsActive(ref, 'user_interruption', USER_INTERRUPTION_RE.test(ref.snippet ?? ''), reviewState, metricScopeId) ? 1 : 0), 0),
+        sessionInterruptedCount: sessionInterruptedEvents(timeline).length,
         negativeFeedbackCount: interactionUserRefs.reduce((sum, ref) => sum + metricCount(ref, 'negative_feedback', findNegativeFeedbackMatches(ref.snippet ?? '').length, reviewState), 0),
         positiveFeedbackCount: interactionUserRefs.reduce((sum, ref) => sum + metricCount(ref, 'positive_feedback', findPositiveFeedbackMatches(ref.snippet ?? '').length, reviewState), 0),
-        userGoalShiftCount: interactionUserRefs.reduce((sum, ref) => sum + metricCount(ref, 'user_goal_shift', findUserGoalShiftMatches(ref.snippet ?? '').length, reviewState), 0),
-        hardRuleTextHitCount: interactionUserRefs.reduce((sum, ref) => sum + (metricIsActive(ref, 'hard_rule', hasUserHardRuleText(ref.snippet ?? ''), reviewState) ? 1 : 0), 0),
+        userGoalShiftCount: interactionUserRefs.reduce((sum, ref) => sum + metricCount(ref, 'user_goal_shift', findUserGoalShiftMatches(ref.snippet ?? '').length, reviewState, metricScopeId), 0),
+        hardRuleTextHitCount: interactionUserRefs.reduce((sum, ref) => sum + (metricIsActive(ref, 'hard_rule', hasUserHardRuleText(ref.snippet ?? ''), reviewState, metricScopeId) ? 1 : 0), 0),
         assistantDeliverySignalCount: timeline.filter(isAssistantDeliveryEvent).length,
+        deliverableArtifactSignalCount: timeline.reduce((sum, ref) => sum + (metricIsActive(ref, 'deliverable_artifact', isAssistantDeliverableArtifactEvent(ref), reviewState) ? 1 : 0), 0),
+        selfCorrectionCount: timeline.reduce((sum, ref) => sum + (metricIsActive(ref, 'self_correction', hasSelfCorrectionSignal(ref), reviewState) ? 1 : 0), 0),
+        repeatedExecutionCount: timeline.reduce((sum, ref) => sum + (metricIsActive(ref, 'repeated_execution', hasRepeatedExecutionSignal(ref), reviewState) ? 1 : 0), 0),
         toolCallCount: segment.metrics.numToolCalls,
-        toolFailureCount: segment.metrics.numToolFailures,
+        toolFailureCount: Math.max(segment.metrics.numToolFailures, timeline.filter(isToolFailureEvent).length),
         highObservationCount: relatedItems.filter((item) => item.severity === 'high').length,
         mediumObservationCount: relatedItems.filter((item) => item.severity === 'medium').length,
         hedgingCount: relatedItems.filter((item) => item.signalType === 'hedging').reduce((sum, item) => sum + item.occurrences, 0),
         explicitMarkerCount: relatedItems.filter((item) => item.signalType === 'explicit_marker').reduce((sum, item) => sum + item.occurrences, 0),
     };
 }
-function metricIsActive(ref, metricKey, ruleDetected, reviewState) {
-    const verdict = observationMetricAnnotationVerdict(reviewState, ref, metricKey);
+function isToolFailureEvent(event) {
+    return event.kind === 'tool_result' && (event.isError === true || isToolResultFailureText(event.fullText ?? event.snippet ?? ''));
+}
+function sessionInterruptedEvents(timeline) {
+    const events = uniqueTimelineEvents(timeline).sort(compareTimelineEvents);
+    const interrupted = [];
+    for (let index = 0; index < events.length; index += 1) {
+        const event = events[index];
+        const text = `${event.label ?? ''}\n${event.snippet ?? ''}\n${event.fullText ?? ''}`;
+        if (hasAssistantTurnFailedText(text)) {
+            interrupted.push(event);
+            continue;
+        }
+        if (/^session[._-]ended$/i.test(event.label ?? '')) {
+            const next = events.slice(index + 1).find((candidate) => /^session[._-]started$/i.test(candidate.label ?? ''));
+            if (next)
+                interrupted.push(event);
+        }
+    }
+    return uniqueTimelineEvents(interrupted);
+}
+function metricIsActive(ref, metricKey, ruleDetected, reviewState, metricScopeId) {
+    const verdict = observationMetricAnnotationVerdict(reviewState, { ...ref, metricScopeId }, metricKey);
     if (verdict === 'confirmed')
         return true;
     if (verdict === 'rejected')
         return false;
     return ruleDetected;
 }
-function metricCount(ref, metricKey, ruleCount, reviewState) {
-    const verdict = observationMetricAnnotationVerdict(reviewState, ref, metricKey);
+function metricCount(ref, metricKey, ruleCount, reviewState, metricScopeId) {
+    const verdict = observationMetricAnnotationVerdict(reviewState, { ...ref, metricScopeId }, metricKey);
     if (verdict === 'confirmed')
         return Math.max(1, ruleCount);
     if (verdict === 'rejected')
         return 0;
     return ruleCount;
 }
-function summarizeExperienceSessions(invocations, sessionGroupsById) {
+function summarizeExperienceSessions(invocations, sessionGroupsByKey, generatedAt, reviewState) {
     const byKey = new Map();
     for (const invocation of invocations) {
-        const key = `${invocation.skillName}\u0000${invocation.sessionId}`;
+        const key = `${invocation.skillName}\u0000${invocation.sessionGroupKey}`;
         const group = byKey.get(key) ?? [];
         group.push(invocation);
         byKey.set(key, group);
@@ -695,7 +842,7 @@ function summarizeExperienceSessions(invocations, sessionGroupsById) {
         const reviewBasisCodes = basisCodesForIndicators(indicators);
         const reviewPriorityScore = scoreForIndicators(indicators);
         const timeline = uniqueTimelineEvents(group.flatMap((invocation) => invocation.timeline)).sort(compareTimelineEvents);
-        const sessionGroup = sessionGroupsById.get(first.sessionId) ?? [];
+        const sessionGroup = sessionGroupsByKey.get(first.sessionGroupKey) ?? [];
         const sourceSessionStartTimestamp = minString(sessionGroup.map((session) => session.startTimestamp));
         const sourceSessionEndTimestamp = maxString(sessionGroup.map((session) => session.endTimestamp));
         const timelineTree = sessionGroup.length > 0 ? buildSessionTimelineTree(first.sessionId, sessionGroup) : undefined;
@@ -727,11 +874,13 @@ function summarizeExperienceSessions(invocations, sessionGroupsById) {
             : fullSessionTimeline.filter((event) => typeof event.messageIndex === 'number' && event.messageIndex > previewEndRecordIndex).length;
         const observationRefs = uniqueEvidenceRefs(group.flatMap((invocation) => invocation.evidenceRefs.filter((ref) => ref.kind === 'observation')));
         const evidenceChain = evidenceChainForTimeline(timeline, observationRefs);
-        const ruleFindings = ruleFindingsForEvidence(indicators, timeline, observationRefs, evidenceChain);
+        const metricScopeId = hashParts('session', first.skillName, first.sessionGroupKey);
+        const ruleFindings = ruleFindingsForEvidence(indicators, timeline, observationRefs, evidenceChain, metricScopeId);
         const assistiveInference = assistiveInferenceForEvidence(indicators, evidenceChain, ruleFindings);
         const problemPatterns = mergeExperienceProblemPatterns(group.flatMap((invocation) => invocation.problemPatterns));
-        return {
-            id: hashParts('session', first.skillName, first.sessionId),
+        const storyInvocations = invocations.filter((invocation) => invocation.sessionGroupKey === first.sessionGroupKey);
+        const baseSession = {
+            id: metricScopeId,
             skillName: first.skillName,
             sessionId: first.sessionId,
             sourceTrace: first.sourceTrace,
@@ -777,12 +926,1007 @@ function summarizeExperienceSessions(invocations, sessionGroupsById) {
             rawSkillRefs: unique(group.map((invocation) => invocation.attribution.rawSkillRef).filter((value) => Boolean(value))).sort(),
             commandNames: unique(group.map((invocation) => invocation.attribution.commandName).filter((value) => Boolean(value))).sort(),
         };
+        const sessionStory = buildSessionStory(baseSession, storyInvocations);
+        const sessionWithStory = {
+            ...baseSession,
+            sessionStory,
+        };
+        const reviewerReport = buildReviewerReport(sessionWithStory, group, generatedAt, reviewState, storyInvocations, sessionStory);
+        return {
+            ...sessionWithStory,
+            reviewPriority: priorityForReviewerFindings(sessionWithStory, reviewerReport.findings),
+            reviewerReport,
+        };
     }).sort((a, b) => {
+        const priorityDiff = experiencePriorityRank(b.reviewPriority) - experiencePriorityRank(a.reviewPriority);
+        if (priorityDiff !== 0)
+            return priorityDiff;
         if (b.reviewPriorityScore !== a.reviewPriorityScore)
             return b.reviewPriorityScore - a.reviewPriorityScore;
         return b.endTimestamp.localeCompare(a.endTimestamp);
     });
 }
+function experiencePriorityRank(priority) {
+    if (priority === 'review_first')
+        return 2;
+    if (priority === 'sample_review')
+        return 1;
+    return 0;
+}
+const REVIEWER_REPORT_RULE_VERSION = 'reviewer-report.v1';
+function buildReviewerReport(session, invocations, generatedAt, reviewState, storyInvocations = invocations, sessionStory = buildSessionStory(session, storyInvocations)) {
+    const indicators = session.indicators;
+    const scopeReasons = reviewerScopeReasonCodes(session);
+    const scopeKind = scopeReasons.length === 0 ? 'single_skill_single_goal' : 'degraded_complex';
+    const findings = reviewerFindingsForSession(session, reviewState);
+    const attentionCount = findings.filter((finding) => finding.level === 'attention').length;
+    const possibleFalsePositiveCount = findings.filter((finding) => finding.level === 'possible_false_positive').length;
+    const tokenUsage = sumTokenUsage(invocations);
+    const title = reviewerTitle(session, attentionCount, possibleFalsePositiveCount);
+    const expectedToolCheck = expectedToolCheckForSession(session);
+    const traceLinks = uniqueEvidenceRefs([
+        ...(session.evidenceChain.firstUserMessage ? [session.evidenceChain.firstUserMessage] : []),
+        ...(session.evidenceChain.firstToolUse ? [session.evidenceChain.firstToolUse] : []),
+        ...(session.evidenceChain.firstToolFailure ? [session.evidenceChain.firstToolFailure] : []),
+        ...(session.evidenceChain.lastAssistantMessage ? [session.evidenceChain.lastAssistantMessage] : []),
+        ...findings.flatMap((finding) => finding.evidenceRefs),
+    ]).slice(0, 10);
+    return {
+        schemaVersion: 1,
+        mode: 'deterministic_session_story',
+        generatedAt,
+        title,
+        summary: reviewerSummary(session, scopeKind, attentionCount, possibleFalsePositiveCount),
+        scope: {
+            kind: scopeKind,
+            reasonCodes: scopeReasons,
+        },
+        chainSteps: [
+            reviewerStep(1, '用户期待', userGoalStepText(session), session.evidenceChain.firstUserMessage ? 'ok' : 'unknown', session.evidenceChain.firstUserMessage ? [session.evidenceChain.firstUserMessage] : []),
+            reviewerStep(2, '选择能力', skillSelectionStepText(session), 'ok', [session.evidenceChain.firstSkillContext, session.evidenceChain.firstToolUse].filter((ref) => Boolean(ref))),
+            reviewerStep(3, '执行流程', executionStepText(session, expectedToolCheck), executionStepStatus(session, expectedToolCheck), session.evidenceChain.firstToolUse ? [session.evidenceChain.firstToolUse] : []),
+            reviewerStep(4, '结果 / 产物', deliveryStepText(session), indicators.assistantDeliverySignalCount > 0 ? 'ok' : 'unknown', session.evidenceChain.lastAssistantMessage ? [session.evidenceChain.lastAssistantMessage] : []),
+            reviewerStep(5, '用户反馈', userFeedbackStepText(session), userFeedbackStepStatus(session), userFeedbackEvidenceRefs(session)),
+        ],
+        findings,
+        oneLookMetrics: {
+            toolCallCount: indicators.toolCallCount,
+            toolFailureCount: indicators.toolFailureCount,
+            userMessageCount: indicators.userMessageCount,
+            userFollowUpCount: indicators.userFollowUpCount,
+            assistantDeliverySignalCount: indicators.assistantDeliverySignalCount,
+            deliverableArtifactSignalCount: indicators.deliverableArtifactSignalCount,
+            assistantProgressUpdateCount: assistantProgressUpdateEvents(session).length,
+            selfCorrectionCount: indicators.selfCorrectionCount,
+            repeatedExecutionCount: indicators.repeatedExecutionCount,
+            finalDeliverySignalCount: assistantFinalDeliveryEvents(session).length,
+            traceEventCount: session.fullSessionTimeline.length || session.timelinePreview.length,
+            tokenUsage: {
+                ...tokenUsage,
+                attribution: 'skill_segment',
+            },
+        },
+        sessionStory,
+        authorSuggestions: reviewerAuthorSuggestions(session, findings),
+        traceLinks,
+    };
+}
+function buildSessionStory(session, invocations) {
+    const nodes = [];
+    const push = (kind, label, status, text, evidenceRefs = []) => {
+        nodes.push({
+            id: hashParts('session-story-node', session.id, kind, String(nodes.length), text),
+            order: nodes.length + 1,
+            kind,
+            label,
+            status,
+            text,
+            evidenceRefs: uniqueEvidenceRefs(evidenceRefs).slice(0, 5),
+        });
+        return nodes[nodes.length - 1];
+    };
+    const goalSlices = sessionStoryGoalSlices(session, invocations);
+    const subagentDispatches = sessionStorySubagentDispatches(session);
+    const skillLinks = sessionStorySkillLinks(session, invocations);
+    const progressUpdates = assistantProgressUpdateEvents(session);
+    const finalDeliveries = assistantFinalDeliveryEvents(session);
+    const userGoalNode = push('user_goal', '用户提出目标', session.evidenceChain.firstUserMessage ? 'ok' : 'unknown', goalSlices.length > 1
+        ? `识别到 ${goalSlices.length} 个目标段：${goalSlices.map((goal) => goal.inferredUserGoal ?? '未提取到明确目标').slice(0, 3).join('；')}${goalSlices.length > 3 ? '；...' : ''}`
+        : session.evidenceChain.firstUserMessage?.snippet
+            ? `用户目标：${session.evidenceChain.firstUserMessage.snippet}`
+            : '没有看到明确人工用户目标；当前只能按运行证据还原链路。', session.evidenceChain.firstUserMessage ? [session.evidenceChain.firstUserMessage] : []);
+    const roleSummary = skillLinks.map((link) => `${link.skillName}：${skillRoleLabel(link.role)}`).join('；');
+    const invocationText = skillLinks.length > 1
+        ? `本次链路识别到 ${skillLinks.length} 个能力：${roleSummary}。`
+        : skillLinks[0]
+            ? `本次使用能力：${skillLinks[0].skillName}，角色判断：${skillRoleLabel(skillLinks[0].role)}。`
+            : `本次使用能力：${session.skillName}。`;
+    const invocationNode = push('skill_invocation', '能力介入', 'ok', invocationText, uniqueEvidenceRefs([session.evidenceChain.firstSkillContext, session.evidenceChain.firstToolUse].filter((ref) => Boolean(ref))));
+    let subagentNode;
+    if (subagentDispatches.length > 0) {
+        subagentNode = push('subagent_branch', '分支 / 子任务', 'unknown', `检测到 ${subagentDispatches.length} 条分支或子任务执行线；主线和分支已单独列出，仍需要结合原文确认真实委派关系。`, subagentDispatches.flatMap((dispatch) => dispatch.evidenceRefs.slice(0, 1)));
+    }
+    const executionNode = push('tool_execution', '执行过程', session.indicators.toolFailureCount > 0 ? 'attention' : session.indicators.toolCallCount > 0 ? 'ok' : 'unknown', session.indicators.toolCallCount > 0
+        ? `执行中看到 ${session.indicators.toolCallCount} 次工具调用${session.indicators.toolFailureCount > 0 ? `，其中失败 ${session.indicators.toolFailureCount} 次` : ''}。`
+        : '没有看到明确工具调用；只能根据消息上下文复盘。', uniqueEvidenceRefs([
+        session.evidenceChain.firstToolUse,
+        session.evidenceChain.firstToolFailure,
+    ].filter((ref) => Boolean(ref))));
+    const deliveryNode = push('delivery', '交付产物', finalDeliveries.length > 0 ? 'ok' : 'attention', deliveryStepText(session), uniqueEvidenceRefs([
+        ...finalDeliveries.slice(-2).map(evidenceRefFromTimeline),
+        ...progressUpdates.slice(-2).map(evidenceRefFromTimeline),
+        session.evidenceChain.lastAssistantMessage,
+    ].filter((ref) => Boolean(ref))));
+    const feedbackNode = push('user_feedback', '用户反馈', userFeedbackStepStatus(session), userFeedbackStepText(session), userFeedbackEvidenceRefs(session));
+    if (session.indicators.userGoalShiftCount > 0) {
+        push('goal_shift', '目标切换', 'unknown', `用户中途切换了 ${session.indicators.userGoalShiftCount} 次目标，后续诉求可能不属于当前 skill。`, userFeedbackEvidenceRefs(session));
+    }
+    const goalChecklistItems = checklistItemsForAnswer(session, 'goal_satisfaction');
+    const declaredBehaviorChecklistItems = checklistItemsForAnswer(session, 'declared_behavior_fit');
+    const userFeelingChecklistItems = checklistItemsForAnswer(session, 'user_feeling');
+    const answers = [
+        sessionStoryAnswer('goal_satisfaction', '用户目标有没有被满足', goalChecklistItems, [
+            session.evidenceChain.firstUserMessage,
+            session.evidenceChain.lastAssistantMessage,
+            ...userFeedbackEvidenceRefs(session),
+        ]),
+        sessionStoryAnswer('declared_behavior_fit', '行为是否符合能力用途', declaredBehaviorChecklistItems, [
+            session.evidenceChain.firstSkillContext,
+            session.evidenceChain.firstToolUse,
+            session.evidenceChain.firstToolFailure,
+        ]),
+        sessionStoryAnswer('user_feeling', '用户是否觉得有用或绕路', userFeelingChecklistItems, userFeedbackEvidenceRefs(session)),
+    ];
+    const summary = answers.some((answer) => answer.status === 'degraded')
+        ? '这次数据本身有可信度问题（比如 skill 没真的被加载），先确认数据再下判断。'
+        : answers.some((answer) => answer.status === 'attention')
+            ? '这次有几条需要看一眼的事项，从红色标记的事实和原文开始看。'
+            : answers.every((answer) => answer.status === 'ok')
+                ? '这次从目标、执行到反馈都没有明显异常，进入常规抽样。'
+                : '这次链路已按语义节点展开，但部分结论仍需要人工结合原文判断。';
+    return {
+        schemaVersion: 1,
+        summary,
+        invocationCount: invocations.length,
+        goalSliceCount: session.goalSliceIds.length,
+        branchCount: subagentDispatches.length,
+        progressUpdateCount: progressUpdates.length,
+        finalDeliverySignalCount: finalDeliveries.length,
+        mainlineNodeIds: [
+            userGoalNode.id,
+            invocationNode.id,
+            ...(subagentNode ? [subagentNode.id] : []),
+            executionNode.id,
+            deliveryNode.id,
+            feedbackNode.id,
+        ],
+        goalSlices,
+        subagentDispatches,
+        skillLinks,
+        graph: sessionStoryGraph(nodes, skillLinks),
+        nodes,
+        answers,
+    };
+}
+function assistantFinalDeliveryEvents(session) {
+    return (session.fullSessionTimeline.length > 0 ? session.fullSessionTimeline : session.timelinePreview)
+        .filter(isAssistantDeliveryEvent);
+}
+function assistantProgressUpdateEvents(session) {
+    return (session.fullSessionTimeline.length > 0 ? session.fullSessionTimeline : session.timelinePreview)
+        .filter(isAssistantProgressUpdateEvent);
+}
+function sessionStoryGoalSlices(session, invocations) {
+    const byId = new Map();
+    for (const invocation of invocations) {
+        const group = byId.get(invocation.goalSliceId) ?? [];
+        group.push(invocation);
+        byId.set(invocation.goalSliceId, group);
+    }
+    return Array.from(byId.entries()).map(([id, group], index) => {
+        const timeline = uniqueTimelineEvents(group.flatMap((invocation) => invocation.timeline)).sort(compareTimelineEvents);
+        const userEvents = timeline.filter((event) => event.kind === 'user_message');
+        const startTimestamp = minString(group.map((invocation) => invocation.startTimestamp)) ?? session.startTimestamp;
+        const endTimestamp = maxString(group.map((invocation) => invocation.endTimestamp)) ?? session.endTimestamp;
+        const hasGoalShift = userEvents.some((event) => hasUserGoalShiftSignal(event.snippet ?? ''));
+        const reasonCode = hasGoalShift
+            ? 'explicit_user_goal_shift'
+            : group.length > 1 ? 'skill_segment_boundary' : 'default_session_slice';
+        return {
+            id,
+            order: index + 1,
+            skillNames: unique(group.map((invocation) => invocation.skillName)).sort(),
+            startTimestamp,
+            endTimestamp,
+            reasonCode,
+            inferredUserGoal: inferUserGoal(userEvents),
+            evidenceRefs: userEvents.slice(0, 3).map(evidenceRefFromTimeline),
+        };
+    }).sort((a, b) => a.startTimestamp.localeCompare(b.startTimestamp));
+}
+function sessionStorySubagentDispatches(session) {
+    const branches = session.timelineTree?.branches ?? [];
+    return branches.map((branch, index) => ({
+        id: hashParts('session-story-dispatch', session.id, branch.id),
+        order: index + 1,
+        branchId: branch.id,
+        label: branch.label,
+        sourceTrace: branch.sourceTrace,
+        attachTo: branch.attachTo ? {
+            messageIndex: branch.attachTo.messageIndex,
+            toolUseId: branch.attachTo.toolUseId,
+            label: branch.attachTo.label,
+        } : undefined,
+        eventCount: branch.events.length,
+        evidenceRefs: branch.events.slice(0, 3).map(evidenceRefFromTimeline),
+    }));
+}
+function sessionStorySkillLinks(session, invocations) {
+    const bySkill = new Map();
+    for (const invocation of invocations) {
+        const group = bySkill.get(invocation.skillName) ?? [];
+        group.push(invocation);
+        bySkill.set(invocation.skillName, group);
+    }
+    return Array.from(bySkill.entries()).map(([skillName, group], index) => {
+        const role = inferSkillRole(group, invocations, session);
+        return {
+            id: hashParts('session-story-skill-link', session.id, skillName),
+            order: index + 1,
+            skillName,
+            role,
+            invocationIds: group.map((invocation) => invocation.id),
+            goalSliceIds: unique(group.map((invocation) => invocation.goalSliceId)),
+            evidenceRefs: uniqueEvidenceRefs(group.flatMap((invocation) => [
+                invocation.evidenceChain.firstSkillContext,
+                invocation.evidenceChain.firstToolUse,
+                ...routingEvidenceEvents(invocation).map(evidenceRefFromTimeline),
+            ].filter((ref) => Boolean(ref)))).slice(0, 5),
+        };
+    }).sort((a, b) => a.order - b.order);
+}
+function inferSkillRole(group, allInvocations, session) {
+    const hasRoutingSignal = group.some((invocation) => routingEvidenceEvents(invocation).length > 0);
+    const hasBranchDispatch = (session.timelineTree?.branches.length ?? 0) > 0 && group.some((invocation) => invocation.timeline.some((event) => event.kind === 'tool_use' && /^(Task|Agent|Skill)$/i.test(event.toolName ?? '')));
+    const hasExecutionSignal = group.some((invocation) => invocation.timeline.some((event) => event.kind === 'tool_use' && !/^(Task|Agent|Skill)$/i.test(event.toolName ?? '')));
+    if ((hasRoutingSignal || hasBranchDispatch) && allInvocations.length > group.length)
+        return hasExecutionSignal ? 'mixed' : 'router';
+    if (hasRoutingSignal || hasBranchDispatch)
+        return 'router';
+    if (hasExecutionSignal)
+        return 'executor';
+    return 'unknown';
+}
+function routingEvidenceEvents(invocation) {
+    return invocation.timeline.filter((event) => {
+        if (event.kind !== 'assistant_message' && event.kind !== 'tool_use')
+            return false;
+        const text = event.fullText ?? event.snippet ?? '';
+        return /子\s*Claude|subagent|子任务|分发|委派|调用.+skill|走\s*`?[\w-]+`?\s*skill|\/consult|task-runner/i.test(text);
+    }).slice(0, 3);
+}
+function skillRoleLabel(role) {
+    if (role === 'router')
+        return '路由';
+    if (role === 'executor')
+        return '执行';
+    if (role === 'mixed')
+        return '路由 + 执行';
+    return '未确认';
+}
+function sessionStoryGraph(nodes, skillLinks) {
+    const graphNodes = nodes.map((node) => ({
+        id: node.id,
+        label: node.label,
+        kind: node.kind,
+        status: node.status,
+        detailNodeId: node.id,
+    }));
+    for (const link of skillLinks) {
+        graphNodes.push({
+            id: link.id,
+            label: `${link.skillName}（${skillRoleLabel(link.role)}）`,
+            kind: 'skill_invocation',
+            status: link.role === 'unknown' ? 'unknown' : 'ok',
+            role: link.role,
+        });
+    }
+    const edges = [];
+    for (let index = 1; index < nodes.length; index += 1) {
+        edges.push({ fromId: nodes[index - 1].id, toId: nodes[index].id, label: '下一步' });
+    }
+    const goalNode = nodes.find((node) => node.kind === 'user_goal');
+    const executionNode = nodes.find((node) => node.kind === 'tool_execution');
+    if (goalNode && executionNode) {
+        for (const link of skillLinks) {
+            edges.push({ fromId: goalNode.id, toId: link.id, label: '触发' });
+            edges.push({ fromId: link.id, toId: executionNode.id, label: skillRoleLabel(link.role) });
+        }
+    }
+    return { nodes: graphNodes, edges };
+}
+function sessionStoryAnswer(key, label, checklistItems, evidenceRefs) {
+    const folded = foldExperienceChecklistItems(checklistItems);
+    return {
+        key,
+        label,
+        status: folded.status,
+        reason: folded.reason,
+        sourceItemKeys: folded.sourceItemKeys,
+        text: sessionStoryAnswerText(key, folded.reason),
+        evidenceRefs: uniqueEvidenceRefs(evidenceRefs.filter((ref) => Boolean(ref))).slice(0, 5),
+        checklistItems,
+    };
+}
+function sessionStoryAnswerText(key, reason) {
+    const texts = {
+        goal_satisfaction: {
+            data_degraded: '这次没看到 skill 真的被加载，先确认数据，再判断目标是否满足。',
+            blocking_failed: '用户出现了不满或叫停，目标没真的满足。',
+            attention_accumulated: '有几条要看一眼，打开原文确认目标是否真的达成。',
+            unknown_dominant: '现有证据不够，判断不了目标是否满足。',
+            all_passed: '关键信号都没问题，看起来目标已满足。',
+            not_applicable: '当前场景不适合回答这个问题。',
+        },
+        declared_behavior_fit: {
+            data_degraded: '没看到 skill 真的被加载，先确认这次任务是不是真的命中了你的 skill。',
+            blocking_failed: '关键执行项没过，skill 行为需要优先看一眼。',
+            attention_accumulated: 'SKILL.md 声明的标准流程、硬性规则或核心工具有几条要看一眼。',
+            unknown_dominant: 'SKILL.md 声明不全或执行证据不够，判不出行为是否符合用途。',
+            all_passed: '执行情况看起来符合 skill 声明的用途。',
+            not_applicable: '当前场景不适合回答这个问题。',
+        },
+        user_feeling: {
+            data_degraded: '这次数据本身有问题，用户感受判断先放一放。',
+            blocking_failed: '看到用户不满或主动叫停，可能觉得没用或绕路了。',
+            attention_accumulated: '看到用户纠正或追问，结合原文判断用户感受。',
+            unknown_dominant: '用户没给明确反馈，判断不了是否觉得有用。',
+            all_passed: '没看到明显不满，用户体感看起来正常。',
+            not_applicable: '当前场景不适合回答这个问题。',
+        },
+    };
+    return texts[key][reason];
+}
+function checklistItemsForAnswer(session, key) {
+    if (key === 'goal_satisfaction')
+        return goalSatisfactionChecklistItems(session);
+    if (key === 'declared_behavior_fit')
+        return declaredBehaviorChecklistItems(session);
+    return userFeelingChecklistItems(session);
+}
+function attributionSourcesToLabel(sources) {
+    if (sources.length === 0)
+        return '未识别（旧数据可能没记录来源）';
+    const map = {
+        'skill-tool': 'assistant 调用 Skill 工具',
+        'command-name': '用户用 slash command',
+        'aima-cmd': '用户用 aima-cmd 块',
+        'skill-script': '跑了 skills/<name>/scripts 脚本',
+        'read-skill-md': 'LLM 主动 Read SKILL.md',
+        unknown: '未知',
+    };
+    return sources.map((s) => map[s] ?? s).join(' + ');
+}
+export function isExperienceTraceInProgress(session) {
+    if (session.indicators.assistantDeliverySignalCount > 0)
+        return false;
+    if (session.indicators.deliverableArtifactSignalCount > 0)
+        return false;
+    // 用户已经主动表达不满（纠正 / 负向 / 中断）属于「被打断」, 不是「在途中」, 仍需触发 final_delivery_absent。
+    if (session.indicators.userCorrectionCount > 0)
+        return false;
+    if (session.indicators.negativeFeedbackCount > 0)
+        return false;
+    if (session.indicators.userInterruptionCount > 0)
+        return false;
+    const last = session.evidenceChain.lastAssistantMessage?.snippet ?? '';
+    // 完全没有最后助手回复时不强判 in-progress, 走原 failed / unknown 路径。
+    if (!last.trim())
+        return false;
+    if (isAssistantProgressUpdateText(last))
+        return true;
+    // 预备语（让我看 / 先看看 / 先读 / 接下来）也算 in-progress, 仍未给出最终回复。
+    return /让我(?:先|来|看|读|检查|分析|确认|拉|获取|继续)|先(?:看(?:看|一?下)|读(?:取|一下)?|确认|检查|分析|拉取|获取)|接下来/i.test(last);
+}
+function checklistItem(input) {
+    const { statusCandidates, ...item } = input;
+    return {
+        ...item,
+        status: aggregateExperienceChecklistItemStatus([input.status, ...(statusCandidates ?? [])]),
+        source: input.source ?? 'deterministic_rule',
+        evidenceRefs: uniqueEvidenceRefs((input.evidenceRefs ?? []).filter((ref) => Boolean(ref))).slice(0, 5),
+    };
+}
+export function hasRecognizableUserGoalText(value) {
+    const text = value?.trim() ?? '';
+    if (!text)
+        return false;
+    if (/^(嗯+|啊+|好的|好|继续|可以|收到|ok|OK|yes|no|不用|不需要)[。.!！?？\s]*$/.test(text))
+        return false;
+    if (/(帮我|请|需要|想要|我要|给我|看下|看一下|基于|根据|重新|继续|先|把|将)/.test(text)
+        && /(生成|创建|写|实现|修复|优化|调整|修改|新增|删除|分析|review|评价|检查|看|拉取|执行|运行|验证|整理|总结|回复|评论|设计|拆分|合并|标注|定位|排查|上传|导出|发布|查询|对齐|沉淀|补充|改|做)/i.test(text)) {
+        return true;
+    }
+    if (/(生成|创建|写一个|实现|修复|优化|调整|修改|新增|分析|review|检查|排查|验证|总结|设计|标注|定位|查询|对齐|补充|改一下|做一个)/i.test(text))
+        return true;
+    return text.length >= 12 && /[？?]/.test(text) && !/^(为什么|怎么|哪里)[？?]?$/.test(text);
+}
+function hasRecognizableUserGoal(ref) {
+    return hasRecognizableUserGoalText(ref?.snippet);
+}
+function goalSatisfactionChecklistItems(session) {
+    const feedbackRefs = userFeedbackEvidenceRefs(session);
+    const goalIdentified = hasRecognizableUserGoal(session.evidenceChain.firstUserMessage);
+    const inProgress = isExperienceTraceInProgress(session);
+    const hasDelivery = session.indicators.assistantDeliverySignalCount > 0;
+    const hasArtifact = session.indicators.deliverableArtifactSignalCount > 0;
+    return [
+        checklistItem({
+            key: 'goal_identified',
+            label: goalIdentified ? '目标已识别' : '目标不明确',
+            status: goalIdentified ? 'passed' : 'unknown',
+            contribution: 'informational',
+            reason: goalIdentified
+                ? '真实用户原文里能识别出目标动作或明确请求。'
+                : session.evidenceChain.firstUserMessage ? '看到真实用户原文，但目标动作不够明确。' : '没有看到真实用户目标原文。',
+            evidenceRefs: [session.evidenceChain.firstUserMessage],
+        }),
+        checklistItem({
+            key: 'completion_result_present',
+            label: hasDelivery ? '给了用户最终答复' : inProgress ? '会话进行中' : '没给用户最终答复',
+            status: hasDelivery ? 'passed' : inProgress ? 'not_applicable' : 'failed',
+            contribution: hasDelivery ? 'attention' : inProgress ? 'neutral' : 'attention',
+            reason: hasDelivery
+                ? '看到 assistant 给出了明确的完成话术或结果反馈。'
+                : inProgress
+                    ? '最后一句还是过程态（「先看看」「让我」），任务还没收尾，先不判定。'
+                    : '没看到 assistant 给用户明确的完成话术或结果反馈。',
+            evidenceRefs: [session.evidenceChain.lastAssistantMessage],
+            suggestionKey: hasDelivery || inProgress ? undefined : 'final_delivery_absent',
+        }),
+        checklistItem({
+            key: 'deliverable_artifact_present',
+            label: hasArtifact ? '给了可点开的产物' : inProgress ? '会话进行中' : '没给可点开的产物',
+            status: hasArtifact ? 'passed' : inProgress ? 'not_applicable' : 'unknown',
+            contribution: hasArtifact ? 'informational' : inProgress ? 'neutral' : 'informational',
+            reason: hasArtifact
+                ? '看到 assistant 回复里附了链接、路径、代码块或文件。'
+                : inProgress
+                    ? '任务还没收尾，先不判定产物。'
+                    : '没看到明确的链接、路径、代码块或文件；不一定失败，得按 skill 目标判断。',
+            evidenceRefs: [session.evidenceChain.lastAssistantMessage],
+            suggestionKey: hasArtifact || inProgress ? undefined : 'artifact_absent',
+        }),
+        checklistItem({
+            key: 'negative_feedback_seen',
+            label: session.indicators.negativeFeedbackCount > 0 ? '看到用户负向反馈' : '未见用户负向反馈',
+            status: session.indicators.negativeFeedbackCount > 0 ? 'failed' : 'passed',
+            contribution: 'blocking',
+            reason: session.indicators.negativeFeedbackCount > 0 ? '看到用户负向表达，不能直接认为目标已满足。' : '没有看到用户负向表达。',
+            evidenceRefs: feedbackRefs,
+            suggestionKey: session.indicators.negativeFeedbackCount > 0 ? 'negative_feedback_review' : undefined,
+        }),
+        checklistItem({
+            key: 'user_correction_seen',
+            label: session.indicators.userCorrectionCount > 0 ? '看到用户纠正' : '未见用户纠正',
+            status: session.indicators.userCorrectionCount > 0 ? 'failed' : 'passed',
+            contribution: 'attention',
+            reason: session.indicators.userCorrectionCount > 0 ? '用户中途纠正了方向，目标是否满足要打开原文看。' : '没有看到用户纠正。',
+            evidenceRefs: feedbackRefs,
+            suggestionKey: session.indicators.userCorrectionCount > 0 ? 'user_correction_review' : undefined,
+        }),
+        checklistItem({
+            key: 'user_interruption_seen',
+            label: session.indicators.userInterruptionCount > 0 ? '看到用户中断' : '未见用户中断',
+            status: session.indicators.userInterruptionCount > 0 ? 'failed' : 'passed',
+            contribution: 'blocking',
+            reason: session.indicators.userInterruptionCount > 0 ? '看到用户中断或停止任务信号，不能认为执行链路自然完成。' : '没有看到用户中断信号。',
+            evidenceRefs: feedbackRefs,
+            suggestionKey: session.indicators.userInterruptionCount > 0 ? 'user_interruption_review' : undefined,
+        }),
+        checklistItem({
+            key: 'goal_shift_seen',
+            label: session.indicators.userGoalShiftCount > 0 ? '看到目标切换' : '未见目标切换',
+            status: session.indicators.userGoalShiftCount > 0 ? 'failed' : 'passed',
+            contribution: 'attention',
+            reason: session.indicators.userGoalShiftCount > 0 ? '用户中途切换了目标，后续诉求可能不属于这个 skill。' : '没有看到目标切换。',
+            evidenceRefs: feedbackRefs,
+            suggestionKey: session.indicators.userGoalShiftCount > 0 ? 'goal_shift_review' : undefined,
+        }),
+    ];
+}
+function declaredBehaviorChecklistItems(session) {
+    const expectedToolCheck = expectedToolCheckForSession(session);
+    const declarations = skillDeclarationCheckForSession(session);
+    const hasSkillRead = session.evidenceChain.skillContextCount > 0;
+    const attributionLabel = attributionSourcesToLabel(session.attributionSources ?? []);
+    const items = [
+        checklistItem({
+            key: 'attribution_source',
+            label: hasSkillRead
+                ? 'LLM 读了 SKILL.md'
+                : `skill 判定来源：${attributionLabel}`,
+            status: 'passed',
+            contribution: 'informational',
+            reason: hasSkillRead
+                ? `看到 ${session.evidenceChain.skillContextCount} 次 SKILL.md 加载事件，可作为能力归因证据。`
+                : `日志里没看到 LLM 主动 Read SKILL.md。这次 skill 归因来自：${attributionLabel}。OpenClaw 脚本 / Skill 工具 / slash command 触发时这是常态，不代表 skill 没用上。`,
+            evidenceRefs: [session.evidenceChain.firstSkillContext, session.evidenceChain.firstToolUse],
+        }),
+        checklistItem({
+            key: 'workflow_declared',
+            label: declarations.workflows.declared ? '标准流程已声明' : '标准流程未声明',
+            status: declarations.workflows.declared ? 'passed' : 'not_declared',
+            contribution: declarations.workflows.declared ? 'informational' : 'attention',
+            reason: declarations.workflows.declared ? `SKILL.md 里声明了 ${declarations.workflows.count} 个标准流程节点。` : 'SKILL.md 没声明标准流程，运行时只能猜流程是否完整。',
+            suggestionKey: declarations.workflows.declared ? undefined : 'workflow_not_declared',
+        }),
+    ];
+    if (declarations.workflows.declared) {
+        const executed = session.indicators.toolCallCount > 0;
+        items.push(checklistItem({
+            key: 'workflow_executed',
+            label: executed ? '标准流程已执行' : '标准流程未执行',
+            status: executed ? 'unknown' : 'failed',
+            contribution: 'attention',
+            reason: executed ? '看到了工具调用，但是不是真的覆盖了完整流程，还要打开原文确认。' : '声明了标准流程，但没看到工具执行的证据。',
+            evidenceRefs: [session.evidenceChain.firstToolUse],
+            suggestionKey: 'workflow_execution_review',
+        }));
+    }
+    items.push(checklistItem({
+        key: 'hardrule_declared',
+        label: declarations.hardRules.declared ? '硬性规则已声明' : '硬性规则未声明',
+        status: declarations.hardRules.declared ? 'passed' : 'not_declared',
+        contribution: declarations.hardRules.declared ? 'informational' : 'attention',
+        reason: declarations.hardRules.declared ? `SKILL.md 里声明了 ${declarations.hardRules.count} 条硬性规则。` : 'SKILL.md 没声明硬性规则。',
+        suggestionKey: declarations.hardRules.declared ? undefined : 'hardrule_not_declared',
+    }));
+    if (declarations.hardRules.declared) {
+        items.push(checklistItem({
+            key: 'hardrule_executed',
+            label: '硬性规则执行情况需打开原文看',
+            status: 'unknown',
+            contribution: 'attention',
+            reason: '声明了硬性规则，但当前规则没法完整证明每条都执行了，要打开原文看。',
+            suggestionKey: 'hardrule_execution_review',
+        }));
+    }
+    items.push(checklistItem({
+        key: 'core_tools_declared',
+        label: expectedToolCheck.declared ? '核心工具已声明' : '核心工具未声明',
+        status: expectedToolCheck.declared ? 'passed' : 'not_declared',
+        contribution: expectedToolCheck.declared ? 'informational' : 'attention',
+        reason: expectedToolCheck.declared ? `SKILL.md 声明的核心工具：${expectedToolCheck.expectedTools.join('、')}。` : 'SKILL.md 没声明核心工具，分不出「真用上了 skill 工具」还是「只是随便调了个工具」。',
+        suggestionKey: expectedToolCheck.declared ? undefined : 'expected_tools_not_declared',
+    }));
+    if (expectedToolCheck.declared) {
+        const hit = expectedToolCheck.matchedTools.length > 0;
+        items.push(checklistItem({
+            key: 'core_tools_hit',
+            label: hit ? '核心工具用上了' : '核心工具没用上',
+            status: hit ? 'passed' : 'failed',
+            contribution: 'blocking',
+            reason: hit ? `用上了核心工具：${expectedToolCheck.matchedTools.join('、')}。` : '没用上 SKILL.md 声明的核心工具。',
+            evidenceRefs: [session.evidenceChain.firstToolUse],
+            suggestionKey: hit ? undefined : 'expected_tools_missed',
+        }));
+    }
+    return items;
+}
+function userFeelingChecklistItems(session) {
+    const feedbackRefs = userFeedbackEvidenceRefs(session);
+    const hasAnyFeedback = session.indicators.positiveFeedbackCount > 0
+        || session.indicators.negativeFeedbackCount > 0
+        || session.indicators.userCorrectionCount > 0
+        || session.indicators.userInterruptionCount > 0
+        || session.indicators.userFollowUpCount > 0
+        || session.indicators.userGoalShiftCount > 0;
+    return [
+        checklistItem({
+            key: 'user_feedback_signal_present',
+            label: hasAnyFeedback ? '看到用户反馈信号' : '未见用户反馈信号',
+            status: hasAnyFeedback ? 'passed' : 'unknown',
+            contribution: 'informational',
+            reason: hasAnyFeedback ? '看到至少一种用户反馈或后续行为信号。' : '没有看到明确用户反馈信号。',
+            evidenceRefs: feedbackRefs,
+        }),
+        checklistItem({
+            key: 'positive_feedback_seen',
+            label: session.indicators.positiveFeedbackCount > 0 ? '看到用户正向反馈' : '未见用户正向反馈',
+            status: session.indicators.positiveFeedbackCount > 0 ? 'passed' : 'unknown',
+            contribution: session.indicators.positiveFeedbackCount > 0 ? 'positive' : 'neutral',
+            reason: session.indicators.positiveFeedbackCount > 0 ? '看到用户认可或正向反馈。' : '没有看到明确正向反馈。',
+            evidenceRefs: feedbackRefs,
+        }),
+        checklistItem({
+            key: 'negative_feedback_seen',
+            label: session.indicators.negativeFeedbackCount > 0 ? '看到用户负向反馈' : '未见用户负向反馈',
+            status: session.indicators.negativeFeedbackCount > 0 ? 'failed' : 'passed',
+            contribution: session.indicators.negativeFeedbackCount > 0 ? 'blocking' : 'neutral',
+            reason: session.indicators.negativeFeedbackCount > 0 ? '看到用户负向表达。' : '没有看到用户负向表达。',
+            evidenceRefs: feedbackRefs,
+            suggestionKey: session.indicators.negativeFeedbackCount > 0 ? 'negative_feedback_review' : undefined,
+        }),
+        checklistItem({
+            key: 'user_correction_seen',
+            label: session.indicators.userCorrectionCount > 0 ? '看到用户纠正' : '未见用户纠正',
+            status: session.indicators.userCorrectionCount > 0 ? 'failed' : 'passed',
+            contribution: session.indicators.userCorrectionCount > 0 ? 'attention' : 'neutral',
+            reason: session.indicators.userCorrectionCount > 0 ? '看到用户重新解释或要求修正。' : '没有看到用户纠正信号。',
+            evidenceRefs: feedbackRefs,
+            suggestionKey: session.indicators.userCorrectionCount > 0 ? 'user_correction_review' : undefined,
+        }),
+        checklistItem({
+            key: 'user_follow_up_seen',
+            label: session.indicators.userFollowUpCount > 0 ? '看到用户追问' : '未见用户追问',
+            status: session.indicators.userFollowUpCount > 0 ? 'unknown' : 'passed',
+            contribution: session.indicators.userFollowUpCount > 0 ? 'informational' : 'neutral',
+            reason: session.indicators.userFollowUpCount > 0 ? '看到用户追问；需要结合上下文区分推进使用还是不满意。' : '没有看到用户追问。',
+            evidenceRefs: feedbackRefs,
+            suggestionKey: session.indicators.userFollowUpCount > 0 ? 'follow_up_review' : undefined,
+        }),
+        checklistItem({
+            key: 'user_interruption_seen',
+            label: session.indicators.userInterruptionCount > 0 ? '看到用户中断' : '未见用户中断',
+            status: session.indicators.userInterruptionCount > 0 ? 'failed' : 'passed',
+            contribution: session.indicators.userInterruptionCount > 0 ? 'blocking' : 'neutral',
+            reason: session.indicators.userInterruptionCount > 0 ? '看到用户中断或停止任务信号。' : '没有看到用户中断信号。',
+            evidenceRefs: feedbackRefs,
+            suggestionKey: session.indicators.userInterruptionCount > 0 ? 'user_interruption_review' : undefined,
+        }),
+    ];
+}
+export function foldExperienceChecklistItems(items) {
+    const relevant = items.filter((item) => item.contribution !== 'neutral');
+    const active = relevant.length > 0 ? relevant : items;
+    const degraded = active.filter((item) => item.status === 'degraded');
+    if (degraded.length > 0)
+        return { status: 'degraded', reason: 'data_degraded', sourceItemKeys: degraded.map((item) => item.key) };
+    const blockingFailed = active.filter((item) => item.contribution === 'blocking' && item.status === 'failed');
+    if (blockingFailed.length > 0)
+        return { status: 'attention', reason: 'blocking_failed', sourceItemKeys: blockingFailed.map((item) => item.key) };
+    const attentionFailed = active.filter((item) => item.contribution === 'attention' && item.status === 'failed');
+    if (attentionFailed.length > 0)
+        return { status: 'attention', reason: 'attention_accumulated', sourceItemKeys: attentionFailed.map((item) => item.key) };
+    const unknown = active.filter((item) => item.status === 'unknown' || item.status === 'not_declared');
+    const positivePassed = active.filter((item) => item.status === 'passed' && item.contribution === 'positive');
+    const decisivePassed = active.filter((item) => item.status === 'passed' && (item.contribution === 'blocking' || item.contribution === 'attention' || item.contribution === 'positive'));
+    const informationalPassed = active.filter((item) => item.status === 'passed' && item.contribution === 'informational');
+    const passed = [...positivePassed, ...decisivePassed.filter((item) => item.contribution !== 'positive'), ...informationalPassed];
+    if (unknown.length > 0 && (decisivePassed.length === 0 || unknown.length >= passed.length)) {
+        return { status: 'unknown', reason: 'unknown_dominant', sourceItemKeys: unknown.map((item) => item.key) };
+    }
+    if (passed.length > 0)
+        return { status: 'ok', reason: 'all_passed', sourceItemKeys: passed.map((item) => item.key) };
+    return { status: 'not_applicable', reason: 'not_applicable', sourceItemKeys: active.map((item) => item.key) };
+}
+function reviewerScopeReasonCodes(session) {
+    const reasons = [];
+    if (session.invocationIds.length !== 1)
+        reasons.push('multiple_invocations');
+    if (session.goalSliceIds.length !== 1)
+        reasons.push('multiple_goal_slices');
+    if ((session.timelineTree?.branches.length ?? 0) > 0)
+        reasons.push('subagent_branches_present');
+    if (session.pluginNames.length > 1 || session.commandNames.length > 1)
+        reasons.push('multiple_skill_entrypoints');
+    return reasons;
+}
+function reviewerStep(order, label, text, status, evidenceRefs = []) {
+    return {
+        order,
+        label,
+        status,
+        text,
+        evidenceRefs: uniqueEvidenceRefs(evidenceRefs).slice(0, 4),
+    };
+}
+function userGoalStepText(session) {
+    const goal = session.evidenceChain.firstUserMessage?.snippet;
+    if (!goal)
+        return '没有看到明确人工用户目标；当前只能按运行证据做常规复盘。';
+    return `用户目标：${goal}`;
+}
+function skillSelectionStepText(session) {
+    const entrypoint = session.commandNames.length > 0 ? `，入口 ${session.commandNames.join('、')}` : session.entrypoint ? `，入口 ${session.entrypoint}` : '';
+    return `本次使用的能力：${session.skillName}${entrypoint}。`;
+}
+function executionStepStatus(session, expectedToolCheck) {
+    if (session.indicators.toolFailureCount > 0 || expectedToolCheck.declared && expectedToolCheck.matchedTools.length === 0)
+        return 'attention';
+    if (session.indicators.toolCallCount > 0)
+        return 'ok';
+    return 'unknown';
+}
+function executionStepText(session, expectedToolCheck = expectedToolCheckForSession(session)) {
+    const failures = session.indicators.toolFailureCount > 0 ? `，其中失败 ${session.indicators.toolFailureCount} 次` : '';
+    const expected = expectedToolCheck.declared
+        ? expectedToolCheck.matchedTools.length > 0
+            ? `命中声明的核心工具：${expectedToolCheck.matchedTools.join('、')}。`
+            : `但没有命中能力声明的核心工具：${expectedToolCheck.expectedTools.join('、')}。`
+        : '';
+    return `执行中看到 ${session.indicators.toolCallCount} 次工具调用${failures}。${expected}`;
+}
+function deliveryStepText(session) {
+    if (session.indicators.assistantDeliverySignalCount > 0) {
+        const artifact = session.indicators.deliverableArtifactSignalCount > 0
+            ? `，其中 ${session.indicators.deliverableArtifactSignalCount} 次包含具体产物线索`
+            : '，但未看到明确产物线索';
+        return `看到 ${session.indicators.assistantDeliverySignalCount} 次完成态或结果反馈${artifact}。`;
+    }
+    return '没有发现最后结果反馈；当前不能把过程进展当成完成。';
+}
+function userFeedbackStepText(session) {
+    const parts = [
+        session.indicators.userFollowUpCount > 0 ? `追问/补充 ${session.indicators.userFollowUpCount} 次` : '',
+        session.indicators.userCorrectionCount > 0 ? `纠正 ${session.indicators.userCorrectionCount} 次` : '',
+        session.indicators.negativeFeedbackCount > 0 ? `负向反馈 ${session.indicators.negativeFeedbackCount} 次` : '',
+        session.indicators.positiveFeedbackCount > 0 ? `正向反馈 ${session.indicators.positiveFeedbackCount} 次` : '',
+        session.indicators.userGoalShiftCount > 0 ? `目标切换 ${session.indicators.userGoalShiftCount} 次` : '',
+    ].filter(Boolean);
+    return parts.length > 0 ? `用户反馈信号：${parts.join('，')}。` : '原始记录里没有看到人工追问、纠正、负向反馈或目标切换。';
+}
+function userFeedbackStepStatus(session) {
+    if (session.indicators.userCorrectionCount > 0 || session.indicators.negativeFeedbackCount > 0 || session.indicators.userInterruptionCount > 0)
+        return 'attention';
+    if (session.indicators.positiveFeedbackCount > 0)
+        return 'ok';
+    return 'unknown';
+}
+function userFeedbackEvidenceRefs(session) {
+    return uniqueEvidenceRefs(session.ruleFindings
+        .filter((finding) => finding.code === 'user_correction_seen' || finding.code === 'negative_feedback_seen' || finding.code === 'positive_feedback_seen' || finding.code === 'user_goal_shift_seen' || finding.code === 'user_interruption_seen')
+        .flatMap((finding) => finding.evidenceRefs)).slice(0, 5);
+}
+function expectedToolCheckForSession(session) {
+    const expectedTools = loadExpectedToolsForSkill(session.skillName, session.cwd);
+    if (expectedTools.length === 0)
+        return { expectedTools: [], matchedTools: [], declared: false };
+    const events = session.fullSessionTimeline.length > 0 ? session.fullSessionTimeline : session.timelinePreview;
+    const matchedTools = expectedTools.filter((tool) => events.some((event) => eventMatchesExpectedTool(event, tool)));
+    return {
+        expectedTools,
+        matchedTools,
+        declared: true,
+    };
+}
+function skillDeclarationCheckForSession(session) {
+    const path = findSkillMdPathForExperience(session.skillName, session.cwd ?? process.cwd());
+    if (!path) {
+        return {
+            hardRules: { declared: false, count: 0 },
+            workflows: { declared: false, count: 0 },
+        };
+    }
+    try {
+        const content = readFileSync(path, 'utf-8');
+        const hardRules = validateSkillHardRules(content);
+        const workflows = validateSkillWorkflows(content);
+        return {
+            hardRules: { declared: hardRules.declared, count: hardRules.rules.length },
+            workflows: { declared: workflows.declared, count: workflows.workflows.reduce((sum, workflow) => sum + workflow.nodes.length, 0) },
+        };
+    }
+    catch {
+        return {
+            hardRules: { declared: false, count: 0 },
+            workflows: { declared: false, count: 0 },
+        };
+    }
+}
+function loadExpectedToolsForSkill(skillName, cwd) {
+    const path = findSkillMdPathForExperience(skillName, cwd ?? process.cwd());
+    if (!path)
+        return [];
+    try {
+        const parsed = parseSkillFrontmatter(readFileSync(path, 'utf-8'));
+        const data = parsed.ok ? parsed.data : undefined;
+        const value = data?.expected_tools ?? data?.expectedTools;
+        return Array.isArray(value)
+            ? unique(value.filter((item) => typeof item === 'string' && Boolean(item.trim())).map((item) => item.trim()))
+            : [];
+    }
+    catch {
+        return [];
+    }
+}
+function eventMatchesExpectedTool(event, expectedTool) {
+    if (event.kind !== 'tool_use')
+        return false;
+    const text = `${event.toolName ?? ''}\n${event.label ?? ''}\n${event.snippet ?? ''}\n${event.fullText ?? ''}`.toLowerCase();
+    const normalized = expectedTool.toLowerCase().trim();
+    const aliases = unique([
+        normalized,
+        normalized.replace(/[-_]?cli$/, ''),
+    ].filter(Boolean));
+    return aliases.some((alias) => new RegExp(`(^|[^a-z0-9_-])${escapeRegExp(alias)}([^a-z0-9_-]|$)`, 'i').test(text));
+}
+function escapeRegExp(value) {
+    return value.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
+}
+function findSkillMdPathForExperience(skillName, cwd) {
+    if (!/^[A-Za-z0-9_.-]+$/.test(skillName))
+        return undefined;
+    const candidates = [
+        join(cwd, '.claude', 'skills', skillName, 'SKILL.md'),
+        join(cwd, '.openclaw', 'workspace', 'skills', skillName, 'SKILL.md'),
+        join(cwd, 'workspace', 'skills', skillName, 'SKILL.md'),
+        join(cwd, 'skills', skillName, 'SKILL.md'),
+        join(homedir(), '.openclaw', 'workspace', 'skills', skillName, 'SKILL.md'),
+        join(homedir(), '.claude', 'skills', skillName, 'SKILL.md'),
+        join(homedir(), '.codex', 'skills', skillName, 'SKILL.md'),
+        join(homedir(), '.agents', 'skills', skillName, 'SKILL.md'),
+    ];
+    return candidates.find((candidate) => existsSync(candidate));
+}
+function reviewerFindingsForSession(session, reviewState) {
+    const findings = [];
+    const push = (level, title, body, ruleSource, evidenceRefs = []) => {
+        const id = hashParts('reviewer-finding', session.id, ruleSource, title);
+        const judgmentId = hashParts('reviewer-judgment', session.id, ruleSource, title, evidenceRefs.map((ref) => ref.id).join('|'));
+        const reviewEntry = reviewState?.entries[observationReviewStateKey('reviewer_judgment', judgmentId)];
+        findings.push({
+            id,
+            judgmentId,
+            source: 'deterministic_rule',
+            level,
+            title,
+            body,
+            ruleSource,
+            ruleVersion: REVIEWER_REPORT_RULE_VERSION,
+            evidenceRefs: uniqueEvidenceRefs(evidenceRefs).slice(0, 5),
+            reviewStateRef: {
+                targetType: 'reviewer_judgment',
+                targetId: judgmentId,
+                ...(reviewEntry?.verdict ? { verdict: reviewEntry.verdict } : {}),
+                ...(reviewEntry?.reason ? { reason: reviewEntry.reason } : {}),
+                ...(reviewEntry?.note ? { note: reviewEntry.note } : {}),
+                ...(reviewEntry?.reviewedAt ? { reviewedAt: reviewEntry.reviewedAt } : {}),
+            },
+        });
+    };
+    const findingRefs = (code) => session.ruleFindings.filter((finding) => finding.code === code).flatMap((finding) => finding.evidenceRefs);
+    const expectedToolCheck = expectedToolCheckForSession(session);
+    if (session.indicators.toolFailureCount > 0) {
+        push('attention', `工具调用失败 ${session.indicators.toolFailureCount} 次`, '执行中遇到工具报错。看下失败的步骤是否在 SKILL.md 里写明了重试或回退方式。', 'tool_error_recovery', findingRefs('tool_failure_seen'));
+    }
+    if (session.indicators.assistantDeliverySignalCount === 0) {
+        push('attention', '没看到给用户的最终答复', 'assistant 没说「完成 / 结果如下」这种收尾，可能任务还没跑完，或收尾文案不够清楚让用户知道事情结束了。', 'final_delivery_absent', session.evidenceChain.lastAssistantMessage ? [session.evidenceChain.lastAssistantMessage] : []);
+    }
+    if (session.indicators.sessionInterruptedCount > 0) {
+        push('attention', `会话异常断开 ${session.indicators.sessionInterruptedCount} 次`, '任务中途被异常中断或重启。如果是网络/超时，看是否要在 skill 里加重试；如果是程序原因，跟开发反馈。', 'session_interrupted', findingRefs('session_interrupted_seen'));
+    }
+    if (expectedToolCheck.declared && expectedToolCheck.matchedTools.length === 0) {
+        push('attention', '没用上 SKILL.md 声明的核心工具', `SKILL.md 里声明 ${expectedToolCheck.expectedTools.join('、')} 是核心工具，但这次没看到调用。要么 description 指引不够清楚，要么用户的诉求不属于这个 skill 的场景。`, 'expected_tools_missed', session.evidenceChain.firstToolUse ? [session.evidenceChain.firstToolUse] : []);
+    }
+    if (session.indicators.userCorrectionCount > 0) {
+        push('attention', `用户纠正 ${session.indicators.userCorrectionCount} 次`, '用户在过程中纠正了方向。看原文确认是 skill 理解偏差，还是 skill 不该处理这种诉求。', 'user_correction', findingRefs('user_correction_seen'));
+    }
+    if (session.indicators.userInterruptionCount > 0) {
+        push('attention', `用户手动叫停 ${session.indicators.userInterruptionCount} 次`, '用户主动喊停了执行。常见原因：跑偏 / 太慢 / 用错工具。看原文定位是哪一步触发的。', 'user_interruption', findingRefs('user_interruption_seen'));
+    }
+    if (session.indicators.negativeFeedbackCount > 0) {
+        push('attention', `用户说了 ${session.indicators.negativeFeedbackCount} 次不满意`, '用户出现了「不对 / 错了 / 不行」等负向表达。先看是 skill 给的结果不达预期，还是用户对方向本身有疑问。', 'negative_feedback', findingRefs('negative_feedback_seen'));
+    }
+    if (session.indicators.hardRuleTextHitCount > 0) {
+        push('note', `用户提了 ${session.indicators.hardRuleTextHitCount} 次硬性要求`, '用户在对话里强调了某些必须做/不能做的规则。如果同类要求反复出现，可以沉淀到 SKILL.md 的 hardRules。', 'user_hard_rule', findingRefs('hard_rule_seen'));
+    }
+    if (reviewerScopeReasonCodes(session).length > 0) {
+        push('note', '复杂链路降级展示', '本次不是严格的 1 次会话 × 1 个目标 × 1 个能力场景。当前先做通用展示，不强行拆分多能力、子任务或目标切换。', 'complex_scope_degraded', []);
+    }
+    if (findings.length === 0) {
+        push('note', '未命中优先问题信号', '没有看到需要优先关注的纠正、中断、负向反馈、工具失败或没收尾的信号。', 'no_priority_signal', []);
+    }
+    return findings;
+}
+function reviewerTitle(session, attentionCount, possibleFalsePositiveCount) {
+    const suffix = possibleFalsePositiveCount > 0 ? ` · ${possibleFalsePositiveCount} 项疑似误判` : '';
+    if (attentionCount > 0)
+        return `${session.skillName} · ${attentionCount} 项要看一眼${suffix}`;
+    if (session.indicators.assistantDeliverySignalCount > 0)
+        return `${session.skillName} · 看起来有结果 · 常规抽样${suffix}`;
+    return `${session.skillName} · 常规抽样 · 未见高优先级信号${suffix}`;
+}
+function reviewerSummary(session, scopeKind, attentionCount, possibleFalsePositiveCount) {
+    const scopeText = scopeKind === 'single_skill_single_goal'
+        ? '本次属于单个能力 / 单个目标报告范围。'
+        : '本次是复杂链路，当前先做降级展示，不强行拆分语义分支。';
+    const reviewText = attentionCount > 0
+        ? `发现 ${attentionCount} 条事实层复核点。`
+        : '没有发现优先级较高的事实层复核点。';
+    const falsePositiveText = possibleFalsePositiveCount > 0 ? `另有 ${possibleFalsePositiveCount} 条疑似误判需要人工确认。` : '';
+    return [scopeText, reviewText, falsePositiveText].filter(Boolean).join(' ');
+}
+function reviewerAuthorSuggestions(session, findings) {
+    const suggestions = new Map();
+    const pushSuggestion = (key, text, severity) => {
+        const existing = suggestions.get(key);
+        if (!existing || severity > existing.severity)
+            suggestions.set(key, { text, severity });
+    };
+    for (const answer of session.sessionStory?.answers ?? []) {
+        for (const item of answer.checklistItems ?? []) {
+            if (!item.suggestionKey)
+                continue;
+            const text = suggestionTextForChecklistItem(item.suggestionKey);
+            if (!text)
+                continue;
+            pushSuggestion(item.suggestionKey, text, severityForChecklistStatus(item.status));
+        }
+    }
+    if (findings.some((finding) => finding.ruleSource === 'final_delivery_absent')) {
+        pushSuggestion('final_delivery_absent', '补充明确的产物交付表达或交付标记，避免过程进展被当成完成。', 4);
+    }
+    if (findings.some((finding) => finding.ruleSource === 'tool_error_recovery')) {
+        pushSuggestion('tool_error_recovery', '复查失败工具调用前后的执行流程，必要时把稳定路径写入能力说明文档。', 4);
+    }
+    if (findings.some((finding) => finding.ruleSource === 'session_interrupted')) {
+        pushSuggestion('session_interrupted', '复查会话异常中断前后的上下文，确认是否需要补充中断恢复或重跑策略。', 4);
+    }
+    if (findings.some((finding) => finding.ruleSource === 'expected_tools_missed')) {
+        pushSuggestion('expected_tools_missed', '如果能力依赖核心工具，请在能力定义里维护 expected_tools，并确认运行链路实际命中这些工具。', 4);
+    }
+    if (session.indicators.hardRuleTextHitCount > 0) {
+        pushSuggestion('user_hard_rule', '把反复出现的用户硬性要求沉淀为能力规则，并在后续观测中追踪是否减少纠偏。', 2);
+    }
+    if (session.indicators.userCorrectionCount > 0 || session.indicators.negativeFeedbackCount > 0) {
+        pushSuggestion('user_negative_review', '优先打开原始片段，确认用户纠正/负向反馈发生在交付前还是交付后。', 4);
+    }
+    if (reviewerScopeReasonCodes(session).length > 0) {
+        pushSuggestion('complex_scope_review', '复杂链路暂按降级报告处理；后续再拆多能力、子任务或目标切换。', 1);
+    }
+    if (suggestions.size === 0)
+        pushSuggestion('routine_sample', '进入常规抽样池，保留 evidenceRef 以便人工抽查。', 0);
+    return Array.from(suggestions.values())
+        .sort((a, b) => b.severity - a.severity || a.text.localeCompare(b.text))
+        .map((entry) => entry.text);
+}
+function severityForChecklistStatus(status) {
+    if (status === 'degraded')
+        return 5;
+    if (status === 'failed')
+        return 4;
+    if (status === 'unknown')
+        return 3;
+    if (status === 'not_declared')
+        return 2;
+    if (status === 'passed')
+        return 1;
+    return 0;
+}
+function suggestionTextForChecklistItem(key) {
+    const suggestions = {
+        final_delivery_absent: '在最后回复里加上「已完成 / 结果如下」之类的明确收尾，让用户知道任务跑完了。',
+        artifact_absent: '如果 skill 应该产出文档、demo、代码或报告，最终回复里要附上文件路径、链接或代码块。',
+        goal_shift_review: '用户中途切了目标，后续追问不属于这个 skill。看下是否要在 description 里说清楚 skill 的边界。',
+        user_negative_or_interrupted: '用户出现了不满 / 纠正 / 叫停。先看原文是哪一步触发的，再决定改 description、补标准流程还是补硬性规则。',
+        workflow_not_declared: '在 SKILL.md 里补一个标准流程声明，把这个 skill 的执行步骤写清楚。否则报告只能猜流程是否完整。',
+        workflow_execution_review: '声明了标准流程但执行证据不够。补一下每个步骤的输出形态，让运行时能验证是否真的跑过。',
+        hardrule_not_declared: '在 SKILL.md 里补硬性规则声明，把那些「必须做 / 不能做」的约束写明。',
+        hardrule_execution_review: '声明了硬性规则但执行证据不够。补一下每条规则的触发场景，让运行时能验证。',
+        expected_tools_not_declared: '在 SKILL.md frontmatter 里声明 expected_tools。否则报告分不出「真用上了 skill 工具」还是「只是随便调了个工具」。',
+        expected_tools_missed: '声明了核心工具但没用上。先确认 description 是否清楚指引到这些工具，或者用户的诉求不属于这个 skill。',
+        negative_feedback_review: '打开用户负向反馈的原文，看问题出在理解目标、执行过程还是最后没收尾。',
+        user_correction_review: '用户纠正了多次。把纠正内容沉淀到 SKILL.md 的标准流程或硬性规则，避免下次同类返工。',
+        follow_up_review: '用户追问比较多。看是围绕产物继续推进（好事），还是因为没拿到结果而反复问（要改）。',
+        user_interruption_review: '用户叫停了执行。看下断的那一步是不是 skill 没声明标准流程导致跑偏。',
+    };
+    return suggestions[key];
+}
+function sumTokenUsage(invocations) {
+    return invocations.reduce((sum, invocation) => ({
+        inputTokens: sum.inputTokens + (invocation.metrics.inputTokens ?? 0),
+        outputTokens: sum.outputTokens + (invocation.metrics.outputTokens ?? 0),
+        cacheReadTokens: sum.cacheReadTokens + (invocation.metrics.cacheReadTokens ?? 0),
+        cacheCreationTokens: sum.cacheCreationTokens + (invocation.metrics.cacheCreationTokens ?? 0),
+    }), {
+        inputTokens: 0,
+        outputTokens: 0,
+        cacheReadTokens: 0,
+        cacheCreationTokens: 0,
+    });
+}
 function summarizeExperienceSkills(sessions, invocations) {
     const bySkill = new Map();
     for (const session of sessions) {
@@ -885,6 +2029,7 @@ function scoreForIndicators(indicators) {
         + indicators.mediumObservationCount
         + indicators.userCorrectionCount * 2
         + indicators.userInterruptionCount * 2
+        + indicators.sessionInterruptedCount * 2
         + indicators.negativeFeedbackCount * 2
         + indicators.hardRuleTextHitCount
         + indicators.toolFailureCount
@@ -898,6 +2043,20 @@ function priorityForScore(score) {
         return 'sample_review';
     return 'routine_sample';
 }
+function priorityForReviewerFindings(session, findings) {
+    const fallback = priorityForScore(session.reviewPriorityScore);
+    const attentionFindings = findings.filter((finding) => finding.level === 'attention');
+    if (attentionFindings.length === 0)
+        return fallback;
+    const criticalMissing = attentionFindings.some((finding) => finding.ruleSource === 'final_delivery_absent'
+        || finding.ruleSource === 'session_interrupted'
+        || finding.ruleSource === 'expected_tools_missed')
+        || session.indicators.userMessageCount === 0
+        || session.indicators.toolCallCount === 0;
+    if (criticalMissing)
+        return 'review_first';
+    return fallback === 'routine_sample' ? 'sample_review' : fallback;
+}
 function basisCodesForIndicators(indicators) {
     const codes = [];
     if (indicators.highObservationCount > 0)
@@ -908,6 +2067,8 @@ function basisCodesForIndicators(indicators) {
         codes.push('user_correction');
     if (indicators.userInterruptionCount > 0)
         codes.push('user_interruption');
+    if (indicators.sessionInterruptedCount > 0)
+        codes.push('session_interrupted');
     if (indicators.negativeFeedbackCount > 0)
         codes.push('negative_feedback');
     if (indicators.hardRuleTextHitCount > 0)
@@ -949,11 +2110,15 @@ function sumIndicators(values) {
         userFollowUpCount: acc.userFollowUpCount + value.userFollowUpCount,
         userCorrectionCount: acc.userCorrectionCount + value.userCorrectionCount,
         userInterruptionCount: acc.userInterruptionCount + value.userInterruptionCount,
+        sessionInterruptedCount: acc.sessionInterruptedCount + (value.sessionInterruptedCount ?? 0),
         negativeFeedbackCount: acc.negativeFeedbackCount + (value.negativeFeedbackCount ?? 0),
         positiveFeedbackCount: acc.positiveFeedbackCount + (value.positiveFeedbackCount ?? 0),
         userGoalShiftCount: acc.userGoalShiftCount + (value.userGoalShiftCount ?? 0),
         hardRuleTextHitCount: acc.hardRuleTextHitCount + value.hardRuleTextHitCount,
         assistantDeliverySignalCount: acc.assistantDeliverySignalCount + (value.assistantDeliverySignalCount ?? 0),
+        deliverableArtifactSignalCount: acc.deliverableArtifactSignalCount + (value.deliverableArtifactSignalCount ?? 0),
+        selfCorrectionCount: acc.selfCorrectionCount + (value.selfCorrectionCount ?? 0),
+        repeatedExecutionCount: acc.repeatedExecutionCount + (value.repeatedExecutionCount ?? 0),
         toolCallCount: acc.toolCallCount + value.toolCallCount,
         toolFailureCount: acc.toolFailureCount + value.toolFailureCount,
         highObservationCount: acc.highObservationCount + value.highObservationCount,
@@ -1126,7 +2291,7 @@ export function hasUserGoalShiftSignal(value) {
 }
 export function findNegativeFeedbackMatches(value) {
     const ranges = [];
-    for (const term of [...NEGATIVE_FEEDBACK_TERMS].sort((a, b) => b.length - a.length)) {
+    for (const term of [...NEGATIVE_FEEDBACK_TERMS, ...ADDITIONAL_NEGATIVE_FEEDBACK_TERMS].sort((a, b) => b.length - a.length)) {
         pushTermMatches(value, term, ranges, false);
     }
     for (const term of [...BOUNDED_NEGATIVE_FEEDBACK_TERMS].sort((a, b) => b.length - a.length)) {
@@ -1173,6 +2338,8 @@ function isSkillContextRecord(record, text) {
 }
 function isRuntimeContextRecord(record, text) {
     const meta = record;
+    if (isRuntimeProtocolPromptText(text))
+        return true;
     if (/^Conversation info \(untrusted metadata\):\s*```json/i.test(text))
         return true;
     if (meta.entrypoint !== 'sdk-ts' || typeof meta.promptId !== 'string')