Next Loop Contract Hermes 下一轮 discovery/enrichment 应优先消费这个合同，而不是自由选择目标。
Runner Mode
只补旧卡 缺报告 new 0 existing 10
Backlog has 11 incomplete cards; enrich existing cards before adding inventory.
还没有任何 loop_report；Hermes 运行后必须写入真实执行报告。
Dispatch Plan
先恢复 Hermes loop_report 健康，再继续维护 Toolkoi 内容。
new 0 existing 10 signals 0
1. 处理优先卡片 required · 5
按 priorityWorkItems 顺序处理，但先判定验证方式：不是所有 MCP 都必须 sandbox；能用官方来源、Wiki 二层关系或可回滚 relation 证明的，优先做 source/wiki/relation 验证，只有安装与 smoke test 可安全复现时才跑 sandbox。
2. 标准化多源输入 required · 3
按 sourceIntakeContract 把 RSS、AI HOT、HEX2077、网页或博客先结构化成 llm_wiki_intake，再判断进入候选、raw source、Wiki 或跳过。
3. 消化候选队列 required · 3
按 candidateQueuePriorities 打开原始 URL，优先处理高分候选；陈旧弱信号写明 skipped/no_action 原因，不要让 RSS/AI HOT/HEX2077 队列无限堆积。
4. 复查质量缺口 required · 10
运行 quality_audit 或等效检查，确认剩余低分卡片。
5. 写回闭环报告 required · 1
调用 toolkoi_loop_report 写入 completed、blocked、skipped、handledSignals 和 summary。
Priority Work Items
外部官方源 跑沙盒
卡片已有官网、官方仓库或安装链接，必须补齐官方 source、安装证据和真实 sandbox。
跑沙盒 · 冷却 最近 sandbox 已失败；先补新安装来源、改变 sandbox 策略，或由用户明确要求后再重跑。
最近 sandbox 未跑通
跑沙盒：真实 sandbox 结果必须为 passed，并满足 sandboxPlan.successSignals；失败只能作为 raw source，并在 loop_report 标记 blocked/skipped。
可验证 toolkoi_sandbox_verify python:3.12-alpine
已有 installCommand；可以用 bounded sandbox 验证最小安装或启动路径。
set -eux
pip install hermes-agent-mcp
printf '\nTOOLKOI_SANDBOX_INSTALL_COMMAND_COMPLETED\n'
printf 'Hermes must inspect command output and only mark passed when the tool starts or answers a minimal help/version/smoke probe.\n' 验收：exit_code=0
验收：输出包含 help/version/health/query 之一
验收：raw source 中有 输入/处理/输出/验证结果
外部官方源 跑沙盒
卡片已有官网、官方仓库或安装链接，必须补齐官方 source、安装证据和真实 sandbox。
跑沙盒 · 冷却 最近 sandbox 已失败；先补新安装来源、改变 sandbox 策略，或由用户明确要求后再重跑。
最近 sandbox 未跑通
跑沙盒：真实 sandbox 结果必须为 passed，并满足 sandboxPlan.successSignals；失败只能作为 raw source，并在 loop_report 标记 blocked/skipped。
可验证 toolkoi_sandbox_verify python:3.12-alpine
已有 installCommand；可以用 bounded sandbox 验证最小安装或启动路径。
set -eux
pip install sakana-fugu
printf '\nTOOLKOI_SANDBOX_INSTALL_COMMAND_COMPLETED\n'
printf 'Hermes must inspect command output and only mark passed when the tool starts or answers a minimal help/version/smoke probe.\n' 验收：exit_code=0
验收：输出包含 help/version/health/query 之一
验收：raw source 中有 输入/处理/输出/验证结果
外部官方源 跑沙盒
卡片已有官网、官方仓库或安装链接，必须补齐官方 source、安装证据和真实 sandbox。
跑沙盒 · 冷却 最近 sandbox 已失败；先补新安装来源、改变 sandbox 策略，或由用户明确要求后再重跑。
最近 sandbox 未跑通
跑沙盒：真实 sandbox 结果必须为 passed，并满足 sandboxPlan.successSignals；失败只能作为 raw source，并在 loop_report 标记 blocked/skipped。
可验证 toolkoi_sandbox_verify python:3.12-alpine
已有 installCommand；可以用 bounded sandbox 验证最小安装或启动路径。
set -eux
pip install zambo-mcp
printf '\nTOOLKOI_SANDBOX_INSTALL_COMMAND_COMPLETED\n'
printf 'Hermes must inspect command output and only mark passed when the tool starts or answers a minimal help/version/smoke probe.\n' 验收：exit_code=0
验收：输出包含 help/version/health/query 之一
验收：raw source 中有 输入/处理/输出/验证结果
外部官方源 跑沙盒
卡片已有官网、官方仓库或安装链接，必须补齐官方 source、安装证据和真实 sandbox。
跑沙盒 · 阻塞 现有 installCommand 不是可执行 shell 命令；Hermes 需要先从官方来源补安装路径，再运行 sandbox。
缺少真实 sandbox 验证
跑沙盒：真实 sandbox 结果必须为 passed，并满足 sandboxPlan.successSignals；失败只能作为 raw source，并在 loop_report 标记 blocked/skipped。
先补安装源 toolkoi_sandbox_verify
现有 installCommand 不是可执行 shell 命令；Hermes 需要先从官方来源补安装路径，再运行 sandbox。
non_executable_install_command
验收：官方安装命令
验收：无凭据最小 smoke test
验收：sandbox raw source 写入 passed
外部官方源 跑沙盒
卡片已有官网、官方仓库或安装链接，必须补齐官方 source、安装证据和真实 sandbox。
跑沙盒 · 阻塞 现有 installCommand 不是可执行 shell 命令；Hermes 需要先从官方来源补安装路径，再运行 sandbox。
缺少真实 sandbox 验证
跑沙盒：真实 sandbox 结果必须为 passed，并满足 sandboxPlan.successSignals；失败只能作为 raw source，并在 loop_report 标记 blocked/skipped。
先补安装源 toolkoi_sandbox_verify
现有 installCommand 不是可执行 shell 命令；Hermes 需要先从官方来源补安装路径，再运行 sandbox。
non_executable_install_command
验收：官方安装命令
验收：无凭据最小 smoke test
验收：sandbox raw source 写入 passed
Required Actions
跑沙盒 candidate_research
Evidence Required
sandbox result candidate_decision
{
"generatedAt": "2026-09-19T07:17:37.700Z",
"runnerMode": {
"mode": "enrichment_only",
"reason": "Backlog has 11 incomplete cards; enrich existing cards before adding inventory.",
"newCardLimit": 0,
"existingCardLimit": 10
},
"loopReportStatus": {
"status": "missing",
"label": "缺报告",
"latestReportAt": null,
"latestContentReportAt": null,
"latestSummary": null,
"ageHours": null,
"reason": "还没有任何 loop_report；Hermes 运行后必须写入真实执行报告。"
},
"dispatchPlan": {
"objective": "先恢复 Hermes loop_report 健康，再继续维护 Toolkoi 内容。",
"maxNewCards": 0,
"maxExistingCards": 10,
"maxSignalItems": 0,
"phases": [
{
"order": 1,
"key": "priority_work",
"title": "处理优先卡片",
"action": "按 priorityWorkItems 顺序处理，但先判定验证方式：不是所有 MCP 都必须 sandbox；能用官方来源、Wiki 二层关系或可回滚 relation 证明的，优先做 source/wiki/relation 验证，只有安装与 smoke test 可安全复现时才跑 sandbox。",
"itemCount": 5,
"required": true,
"exitCriteria": [
"只把 actionCriteria 已满足的动作写入 completed。",
"sandbox_failed/cooldown 项先做策略复盘，不能反复重跑同一个失败探针。",
"cooldown/blocked 只作用到对应 action，不能跳过整张卡片的其他 ready 动作。",
"未满足项必须写入 blocked 或 skipped 并说明原因。"
]
},
{
"order": 2,
"key": "source_intake",
"title": "标准化多源输入",
"action": "按 sourceIntakeContract 把 RSS、AI HOT、HEX2077、网页或博客先结构化成 llm_wiki_intake，再判断进入候选、raw source、Wiki 或跳过。",
"itemCount": 3,
"required": true,
"sourceIntakeContract": {
"mode": "multi_source_to_llm_wiki",
"sourceCount": 3,
"outputLayer": "llm_wiki_intake"
},
"exitCriteria": [
"每个来源输出结构化 JSON，包含 sourceKey、sourceItemId、title、url、summaryZh 和 candidateAction。",
"聚合 RSS 必须逐条处理 entry；网页/日报必须保留真实原文 URL。",
"缺少可验证工具价值的条目只能 skip 或保留 raw source，不能直接建卡。"
]
},
{
"order": 3,
"key": "candidate_queue",
"title": "消化候选队列",
"action": "按 candidateQueuePriorities 打开原始 URL，优先处理高分候选；陈旧弱信号写明 skipped/no_action 原因，不要让 RSS/AI HOT/HEX2077 队列无限堆积。",
"itemCount": 3,
"required": true,
"capacityPlan": {
"mode": "capacity_expansion",
"targetDecisionsPerRun": 15,
"suggestedReviewWindow": 20,
"suggestedCadenceMinutes": 30,
"sourceTargets": [
{
"sourceKey": "hex2077",
"label": "HEX2077",
"priority": 0,
"targetDecisions": 5,
"visibleCandidates": 5,
"queuedCount": 296,
"reviewLimit": 5,
"reason": "本轮处理 HEX2077 的 5 条可见候选；逐条打开原链接并写入 evidence-backed decision。"
},
{
"sourceKey": "rss",
"label": "RSS",
"priority": 1,
"targetDecisions": 5,
"visibleCandidates": 5,
"queuedCount": 274,
"reviewLimit": 5,
"reason": "本轮处理 RSS 的 5 条可见候选；逐条打开原链接并写入 evidence-backed decision。"
},
{
"sourceKey": "aihot",
"label": "AI HOT",
"priority": 2,
"targetDecisions": 5,
"visibleCandidates": 5,
"queuedCount": 196,
"reviewLimit": 5,
"reason": "本轮处理 AI HOT 的 5 条可见候选；逐条打开原链接并写入 evidence-backed decision。"
}
],
"actions": [
"运行 candidate_decision_only pass，先清 visible topCandidates，再刷新合同窗口。",
"保持 toolkoi_candidate_decision 单条写入；禁止批量猜测或为未打开候选造决策。",
"若连续两轮 outsideTopWindow 未下降，优先提升来源质量过滤或继续扩大候选窗口。"
],
"reason": "当前按可见窗口预计还需 52 轮，候选处理能力不足，需要进入候选决策专注模式。"
},
"executionRunbook": {
"mode": "candidate_decision_only",
"decisionTool": "toolkoi_candidate_decision",
"candidatesSource": "dispatchPlan.phases[key=candidate_queue].capacityPlan.sourceTargets + candidateQueuePriorities[].topCandidates",
"refreshAfterDecisions": 15,
"steps": [
{
"order": 1,
"action": "read_capacity_targets",
"required": true,
"instruction": "读取 candidate_queue phase 的 capacityPlan.sourceTargets，按 priority 从小到大处理 targetDecisions > 0 的来源。"
},
{
"order": 2,
"action": "open_candidate_source",
"required": true,
"instruction": "从对应 candidateQueuePriorities[].topCandidates 取 candidate id 和原始 URL；每条都必须打开原链接并判断来源可信度。"
},
{
"order": 3,
"action": "write_single_candidate_decision",
"required": true,
"instruction": "每研究完一条 candidate，立即调用 toolkoi_candidate_decision 写入 outcome、reason、evidence、decidedBy。"
},
{
"order": 4,
"action": "refresh_loop_contract",
"required": true,
"instruction": "达到 refreshAfterDecisions 或可见候选耗尽后，重新读取 toolkoi_loop_contract，禁止继续按旧窗口处理。"
}
],
"guardrails": [
"禁止批量写入 candidate decision。",
"不能为未打开原链接的 candidate 造 outcome。",
"sourceTargets 里 targetDecisions 为 0 的来源只能刷新窗口或检查来源质量，不能写决策。"
],
"completionCriteria": [
"本轮写入的 candidate decision 数量达到 refreshAfterDecisions，或所有 visible topCandidates 已处理。",
"每条写入都有 original_url、source_confidence、decision_reason 证据。",
"处理后必须刷新 loop_contract 并重新计算下一轮 sourceTargets。"
]
},
"executionBatch": [
{
"batchIndex": 1,
"sourceKey": "hex2077",
"sourceLabel": "HEX2077",
"priority": 0,
"candidateId": "47743b21-5878-499b-ac8a-d473a0513342",
"title": "1. The Trust Fracture: Claude's Dual Identity Crisis | 信任断裂：Claude 的双面危机——从水印风波到阿里全面禁用",
"url": "https://www.reuters.com/world/china/alibaba-ban-claude-code-workplace-over-alleged-backdoor-risks-source-says-2026-07-03/",
"urlHost": "reuters.com",
"urlClass": "web_page",
"decisionHint": "打开页面确认是否有官方产品、文档、仓库或可接入能力；证据不足时不要建卡。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "47743b21-5878-499b-ac8a-d473a0513342",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://www.reuters.com/world/china/alibaba-ban-claude-code-workplace-over-alleged-backdoor-risks-source-says-2026-07-03/",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:1"
],
"payloadSkeleton": {
"candidateId": "47743b21-5878-499b-ac8a-d473a0513342",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://www.reuters.com/world/china/alibaba-ban-claude-code-workplace-over-alleged-backdoor-risks-source-says-2026-07-03/",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:1"
],
"decidedBy": "hermes"
}
},
"score": 92,
"queuedAt": "2026-07-04T03:38:48.417Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 HEX2077 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 2,
"sourceKey": "hex2077",
"sourceLabel": "HEX2077",
"priority": 0,
"candidateId": "467d2960-b7c1-46b6-940d-45d81ff4b6e5",
"title": "3. AI Security: From Shield to Sword | 安全悖论：AI 既是盾牌，也是利刃",
"url": "https://the-decoder.com/security-vulnerability-reports-have-exploded-since-ai-models-started-hunting-for-bugs/",
"urlHost": "the-decoder.com",
"urlClass": "web_page",
"decisionHint": "打开页面确认是否有官方产品、文档、仓库或可接入能力；证据不足时不要建卡。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "467d2960-b7c1-46b6-940d-45d81ff4b6e5",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://the-decoder.com/security-vulnerability-reports-have-exploded-since-ai-models-started-hunting-for-bugs/",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:2"
],
"payloadSkeleton": {
"candidateId": "467d2960-b7c1-46b6-940d-45d81ff4b6e5",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://the-decoder.com/security-vulnerability-reports-have-exploded-since-ai-models-started-hunting-for-bugs/",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:2"
],
"decidedBy": "hermes"
}
},
"score": 92,
"queuedAt": "2026-07-04T03:38:48.417Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 HEX2077 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 3,
"sourceKey": "hex2077",
"sourceLabel": "HEX2077",
"priority": 0,
"candidateId": "5144d0d7-b3a0-4600-ba8c-113f289e45c6",
"title": "前沿大模型被曝存在同伴保护行为",
"url": "https://arxiv.org/abs/2604.19784",
"urlHost": "arxiv.org",
"urlClass": "web_page",
"decisionHint": "打开页面确认是否有官方产品、文档、仓库或可接入能力；证据不足时不要建卡。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "5144d0d7-b3a0-4600-ba8c-113f289e45c6",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://arxiv.org/abs/2604.19784",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:3"
],
"payloadSkeleton": {
"candidateId": "5144d0d7-b3a0-4600-ba8c-113f289e45c6",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://arxiv.org/abs/2604.19784",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:3"
],
"decidedBy": "hermes"
}
},
"score": 92,
"queuedAt": "2026-07-04T03:38:48.417Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 HEX2077 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 4,
"sourceKey": "hex2077",
"sourceLabel": "HEX2077",
"priority": 0,
"candidateId": "93f25c25-e779-4481-9d1b-9658bbbcab4a",
"title": "1. The Security Arms Race | AI安全从附属品升格为战略武器：GPT-5.6受控发布、Daybreak防御平台与政府审批风暴三线并行",
"url": "http://mp.weixin.qq.com/s?__biz=MjM5NjA0NjgyMA==&mid=2651341309&idx=1&sn=9afc5021305207db4b56448f2d298c80&chksm=bc4efd5d2cfde9f4d358f3c696aa65c7c341e85bb2058edb80ca29424675c1e6b473319271de&scene=0&xtrack=1#rd",
"urlHost": "mp.weixin.qq.com",
"urlClass": "news_article",
"decisionHint": "按情报线索处理：只在文章明确指向工具、MCP、Skill、模型能力或官方项目时转入 raw source；泛新闻应跳过。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "93f25c25-e779-4481-9d1b-9658bbbcab4a",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:http://mp.weixin.qq.com/s?__biz=MjM5NjA0NjgyMA==&mid=2651341309&idx=1&sn=9afc5021305207db4b56448f2d298c80&chksm=bc4efd5d2cfde9f4d358f3c696aa65c7c341e85bb2058edb80ca29424675c1e6b473319271de&scene=0&xtrack=1#rd",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:4"
],
"payloadSkeleton": {
"candidateId": "93f25c25-e779-4481-9d1b-9658bbbcab4a",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:http://mp.weixin.qq.com/s?__biz=MjM5NjA0NjgyMA==&mid=2651341309&idx=1&sn=9afc5021305207db4b56448f2d298c80&chksm=bc4efd5d2cfde9f4d358f3c696aa65c7c341e85bb2058edb80ca29424675c1e6b473319271de&scene=0&xtrack=1#rd",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:4"
],
"decidedBy": "hermes"
}
},
"score": 92,
"queuedAt": "2026-06-29T05:43:23.429Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 HEX2077 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 5,
"sourceKey": "hex2077",
"sourceLabel": "HEX2077",
"priority": 0,
"candidateId": "8c516342-7b0c-4d42-ae1e-710f4a32a1b2",
"title": "2. The Model Battleground Escalates | 模型军备竞赛白热化：字节豆包2.1、Google Gemini 3.5 Flash、智谱GLM-5.2混战升级",
"url": "https://mp.weixin.qq.com/s?__biz=MzI1MzYzMjE0MQ==&mid=2247520467&idx=1&sn=e6210ac0ffacb37351752b7fcb5259e5",
"urlHost": "mp.weixin.qq.com",
"urlClass": "news_article",
"decisionHint": "按情报线索处理：只在文章明确指向工具、MCP、Skill、模型能力或官方项目时转入 raw source；泛新闻应跳过。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "8c516342-7b0c-4d42-ae1e-710f4a32a1b2",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://mp.weixin.qq.com/s?__biz=MzI1MzYzMjE0MQ==&mid=2247520467&idx=1&sn=e6210ac0ffacb37351752b7fcb5259e5",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:5"
],
"payloadSkeleton": {
"candidateId": "8c516342-7b0c-4d42-ae1e-710f4a32a1b2",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://mp.weixin.qq.com/s?__biz=MzI1MzYzMjE0MQ==&mid=2247520467&idx=1&sn=e6210ac0ffacb37351752b7fcb5259e5",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:5"
],
"decidedBy": "hermes"
}
},
"score": 92,
"queuedAt": "2026-06-29T05:43:23.429Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 HEX2077 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 6,
"sourceKey": "rss",
"sourceLabel": "RSS",
"priority": 1,
"candidateId": "756da756-1cf4-4382-841d-90367abd6c67",
"title": "SkillOpt: Agent skills as trainable parameters",
"url": "https://github.com/microsoft/SkillOpt",
"urlHost": "github.com",
"urlClass": "code_repository",
"decisionHint": "优先检查仓库是否是可接入工具、MCP、Skill 或开发效率能力；若只是论文/样例报告，改写为 raw source 或 skipped。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "756da756-1cf4-4382-841d-90367abd6c67",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://github.com/microsoft/SkillOpt",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:6"
],
"payloadSkeleton": {
"candidateId": "756da756-1cf4-4382-841d-90367abd6c67",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://github.com/microsoft/SkillOpt",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:6"
],
"decidedBy": "hermes"
}
},
"score": 100,
"queuedAt": "2026-06-30T17:22:18.753Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 RSS 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 7,
"sourceKey": "rss",
"sourceLabel": "RSS",
"priority": 1,
"candidateId": "c9a5cc4e-6950-477f-9e58-ec69f91e72df",
"title": "Memora: A Harmonic Memory Representation Balancing Abstraction and Specificity",
"url": "https://github.com/microsoft/Memora",
"urlHost": "github.com",
"urlClass": "code_repository",
"decisionHint": "优先检查仓库是否是可接入工具、MCP、Skill 或开发效率能力；若只是论文/样例报告，改写为 raw source 或 skipped。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "c9a5cc4e-6950-477f-9e58-ec69f91e72df",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://github.com/microsoft/Memora",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:7"
],
"payloadSkeleton": {
"candidateId": "c9a5cc4e-6950-477f-9e58-ec69f91e72df",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://github.com/microsoft/Memora",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:7"
],
"decidedBy": "hermes"
}
},
"score": 100,
"queuedAt": "2026-06-29T21:18:56.775Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 RSS 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 8,
"sourceKey": "rss",
"sourceLabel": "RSS",
"priority": 1,
"candidateId": "02b406f7-e448-4fae-aadd-89cf04dd2e53",
"title": "Ire identifies another LOTUSLITE specimen",
"url": "https://github.com/microsoft/project-ire/blob/main/reports/47e51e82229e80a387c3cb100d39d3705e6360bbf9bfa1601dbc484e8d02e653.md",
"urlHost": "github.com",
"urlClass": "code_repository",
"decisionHint": "优先检查仓库是否是可接入工具、MCP、Skill 或开发效率能力；若只是论文/样例报告，改写为 raw source 或 skipped。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "02b406f7-e448-4fae-aadd-89cf04dd2e53",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://github.com/microsoft/project-ire/blob/main/reports/47e51e82229e80a387c3cb100d39d3705e6360bbf9bfa1601dbc484e8d02e653.md",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:8"
],
"payloadSkeleton": {
"candidateId": "02b406f7-e448-4fae-aadd-89cf04dd2e53",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://github.com/microsoft/project-ire/blob/main/reports/47e51e82229e80a387c3cb100d39d3705e6360bbf9bfa1601dbc484e8d02e653.md",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:8"
],
"decidedBy": "hermes"
}
},
"score": 100,
"queuedAt": "2026-06-29T16:29:08.652Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 RSS 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 9,
"sourceKey": "rss",
"sourceLabel": "RSS",
"priority": 1,
"candidateId": "162e1297-5c68-49e5-8075-33414ea72c77",
"title": "Data Formulator 0.7: AI-powered data analytics for enterprise data",
"url": "https://github.com/microsoft/data-formulator",
"urlHost": "github.com",
"urlClass": "code_repository",
"decisionHint": "优先检查仓库是否是可接入工具、MCP、Skill 或开发效率能力；若只是论文/样例报告，改写为 raw source 或 skipped。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "162e1297-5c68-49e5-8075-33414ea72c77",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://github.com/microsoft/data-formulator",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:9"
],
"payloadSkeleton": {
"candidateId": "162e1297-5c68-49e5-8075-33414ea72c77",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://github.com/microsoft/data-formulator",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:9"
],
"decidedBy": "hermes"
}
},
"score": 100,
"queuedAt": "2026-06-29T16:29:08.652Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 RSS 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 10,
"sourceKey": "rss",
"sourceLabel": "RSS",
"priority": 1,
"candidateId": "a2624854-e23b-4305-abe0-940b73f76935",
"title": "Evaluating performance and efficiency of the GitHub Copilot agentic harness across models and tasks",
"url": "https://github.com/github/copilot-sdk",
"urlHost": "github.com",
"urlClass": "code_repository",
"decisionHint": "优先检查仓库是否是可接入工具、MCP、Skill 或开发效率能力；若只是论文/样例报告，改写为 raw source 或 skipped。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "a2624854-e23b-4305-abe0-940b73f76935",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://github.com/github/copilot-sdk",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:10"
],
"payloadSkeleton": {
"candidateId": "a2624854-e23b-4305-abe0-940b73f76935",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://github.com/github/copilot-sdk",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:10"
],
"decidedBy": "hermes"
}
},
"score": 100,
"queuedAt": "2026-06-29T16:22:15.520Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 RSS 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 11,
"sourceKey": "aihot",
"sourceLabel": "AI HOT",
"priority": 2,
"candidateId": "88970f6e-610e-40bb-b796-dbcc7f7f8cf6",
"title": "语言模型中的全局工作空间",
"url": "https://www.anthropic.com/research/global-workspace",
"urlHost": "anthropic.com",
"urlClass": "web_page",
"decisionHint": "打开页面确认是否有官方产品、文档、仓库或可接入能力；证据不足时不要建卡。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "88970f6e-610e-40bb-b796-dbcc7f7f8cf6",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://www.anthropic.com/research/global-workspace",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:11"
],
"payloadSkeleton": {
"candidateId": "88970f6e-610e-40bb-b796-dbcc7f7f8cf6",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://www.anthropic.com/research/global-workspace",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:11"
],
"decidedBy": "hermes"
}
},
"score": 90,
"queuedAt": "2026-07-07T16:11:59.076Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 AI HOT 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 12,
"sourceKey": "aihot",
"sourceLabel": "AI HOT",
"priority": 2,
"candidateId": "33d0b4d1-7e8a-4413-bce1-27abc2500bfc",
"title": "开辟新赛道：小米 YU7 GT 创全球首个纽北自动驾驶圈速纪录，官方圈速榜新增\"自动驾驶\"分类",
"url": "https://www.ithome.com/0/967/234.htm",
"urlHost": "ithome.com",
"urlClass": "news_article",
"decisionHint": "按情报线索处理：只在文章明确指向工具、MCP、Skill、模型能力或官方项目时转入 raw source；泛新闻应跳过。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "33d0b4d1-7e8a-4413-bce1-27abc2500bfc",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://www.ithome.com/0/967/234.htm",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:12"
],
"payloadSkeleton": {
"candidateId": "33d0b4d1-7e8a-4413-bce1-27abc2500bfc",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://www.ithome.com/0/967/234.htm",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:12"
],
"decidedBy": "hermes"
}
},
"score": 85,
"queuedAt": "2026-06-24T15:24:25.903Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 AI HOT 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 13,
"sourceKey": "aihot",
"sourceLabel": "AI HOT",
"priority": 2,
"candidateId": "31ca07fd-4885-4d32-96ad-2cc87779b01c",
"title": "扎克伯格：建千兆瓦级AI集群，集中精英与资本",
"url": "https://x.com/rohanpaul_ai/status/2073834219659534816",
"urlHost": "x.com",
"urlClass": "social_discussion",
"decisionHint": "按二级信号处理：先找原始项目、论文、官网或官方仓库；找不到强来源时写 skipped/no_action。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "31ca07fd-4885-4d32-96ad-2cc87779b01c",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://x.com/rohanpaul_ai/status/2073834219659534816",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:13"
],
"payloadSkeleton": {
"candidateId": "31ca07fd-4885-4d32-96ad-2cc87779b01c",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://x.com/rohanpaul_ai/status/2073834219659534816",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:13"
],
"decidedBy": "hermes"
}
},
"score": 84,
"queuedAt": "2026-07-05T20:11:25.055Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 AI HOT 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 14,
"sourceKey": "aihot",
"sourceLabel": "AI HOT",
"priority": 2,
"candidateId": "a74c7bd4-4d1f-4066-9d39-f277c42b29f9",
"title": "Anthropic在Claude Code中植入隐写术代码识别中国用户",
"url": "https://mp.weixin.qq.com/s/yLb4T2UC16ebKHApdBbgWw",
"urlHost": "mp.weixin.qq.com",
"urlClass": "news_article",
"decisionHint": "按情报线索处理：只在文章明确指向工具、MCP、Skill、模型能力或官方项目时转入 raw source；泛新闻应跳过。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "a74c7bd4-4d1f-4066-9d39-f277c42b29f9",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://mp.weixin.qq.com/s/yLb4T2UC16ebKHApdBbgWw",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:14"
],
"payloadSkeleton": {
"candidateId": "a74c7bd4-4d1f-4066-9d39-f277c42b29f9",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://mp.weixin.qq.com/s/yLb4T2UC16ebKHApdBbgWw",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:14"
],
"decidedBy": "hermes"
}
},
"score": 84,
"queuedAt": "2026-07-01T02:13:04.838Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 AI HOT 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 15,
"sourceKey": "aihot",
"sourceLabel": "AI HOT",
"priority": 2,
"candidateId": "e662e047-7914-4356-8c32-6e54eedebd05",
"title": "纽约时报修订诉讼，指控微软为OpenAI建造版权侵权超级计算机",
"url": "https://arstechnica.com/tech-policy/2026/06/microsoft-built-supercomputer-to-help-openai-infringe-copyrights-nyt-alleged",
"urlHost": "arstechnica.com",
"urlClass": "web_page",
"decisionHint": "打开页面确认是否有官方产品、文档、仓库或可接入能力；证据不足时不要建卡。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "e662e047-7914-4356-8c32-6e54eedebd05",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://arstechnica.com/tech-policy/2026/06/microsoft-built-supercomputer-to-help-openai-infringe-copyrights-nyt-alleged",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:15"
],
"payloadSkeleton": {
"candidateId": "e662e047-7914-4356-8c32-6e54eedebd05",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://arstechnica.com/tech-policy/2026/06/microsoft-built-supercomputer-to-help-openai-infringe-copyrights-nyt-alleged",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:15"
],
"decidedBy": "hermes"
}
},
"score": 84,
"queuedAt": "2026-06-26T22:09:12.201Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 AI HOT 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
}
],
"executionSummary": {
"itemCount": 15,
"activeSourceCount": 3,
"emptyTargetCount": 0,
"decisionTool": "toolkoi_candidate_decision",
"refreshAfterDecisions": 15,
"highestScoreCandidate": {
"candidateId": "756da756-1cf4-4382-841d-90367abd6c67",
"sourceKey": "rss",
"title": "SkillOpt: Agent skills as trainable parameters",
"url": "https://github.com/microsoft/SkillOpt",
"score": 100
},
"oldestCandidate": {
"candidateId": "33d0b4d1-7e8a-4413-bce1-27abc2500bfc",
"sourceKey": "aihot",
"title": "开辟新赛道：小米 YU7 GT 创全球首个纽北自动驾驶圈速纪录，官方圈速榜新增\"自动驾驶\"分类",
"url": "https://www.ithome.com/0/967/234.htm",
"queuedAt": "2026-06-24T15:24:25.903Z"
},
"sourceBreakdown": [
{
"sourceKey": "hex2077",
"label": "HEX2077",
"priority": 0,
"targetDecisions": 5,
"batchCount": 5,
"queuedCount": 296,
"highestScore": 92,
"oldestQueuedAt": "2026-06-29T05:43:23.429Z",
"firstCandidateId": "47743b21-5878-499b-ac8a-d473a0513342"
},
{
"sourceKey": "rss",
"label": "RSS",
"priority": 1,
"targetDecisions": 5,
"batchCount": 5,
"queuedCount": 274,
"highestScore": 100,
"oldestQueuedAt": "2026-06-29T16:22:15.520Z",
"firstCandidateId": "756da756-1cf4-4382-841d-90367abd6c67"
},
{
"sourceKey": "aihot",
"label": "AI HOT",
"priority": 2,
"targetDecisions": 5,
"batchCount": 5,
"queuedCount": 196,
"highestScore": 90,
"oldestQueuedAt": "2026-06-24T15:24:25.903Z",
"firstCandidateId": "88970f6e-610e-40bb-b796-dbcc7f7f8cf6"
}
],
"operatorNote": "按 batchIndex 顺序逐条打开原链接；每写入 toolkoi_candidate_decision 后记录 evidence；处理 15 条或批次耗尽后刷新 loop_contract。"
},
"reportGate": {
"reportTool": "toolkoi_loop_report",
"consumedContractGeneratedAt": "2026-09-19T07:17:37.700Z",
"expectedDecisionCount": 15,
"refreshRequiredAfterDecisions": 15,
"acceptedDecisionTool": "toolkoi_candidate_decision",
"requiredCompletedFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"requiredEvidenceFields": [
"original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index"
],
"sourceQuotas": [
{
"sourceKey": "hex2077",
"label": "HEX2077",
"expectedDecisionCount": 5,
"firstCandidateId": "47743b21-5878-499b-ac8a-d473a0513342"
},
{
"sourceKey": "rss",
"label": "RSS",
"expectedDecisionCount": 5,
"firstCandidateId": "756da756-1cf4-4382-841d-90367abd6c67"
},
{
"sourceKey": "aihot",
"label": "AI HOT",
"expectedDecisionCount": 5,
"firstCandidateId": "88970f6e-610e-40bb-b796-dbcc7f7f8cf6"
}
],
"failureModes": [
"opened_candidate_without_decision",
"decision_without_required_evidence",
"continued_after_refresh_threshold"
],
"loopReportInstruction": "处理 candidateExecutionBatch 后必须调用 toolkoi_loop_report；completed/skipped/blocked 需要覆盖每条已打开候选，按 sourceQuotas 交账，不要只写泛泛 summary。达到 15 条后必须刷新 loop_contract。"
},
"reportAudit": {
"status": "pending",
"expectedDecisionCount": 15,
"matchingDecisionCount": 0,
"missingDecisionCount": 15,
"evidenceGapCount": 0,
"reportReady": false,
"missingCandidateIds": [
"47743b21-5878-499b-ac8a-d473a0513342",
"467d2960-b7c1-46b6-940d-45d81ff4b6e5",
"5144d0d7-b3a0-4600-ba8c-113f289e45c6",
"93f25c25-e779-4481-9d1b-9658bbbcab4a",
"8c516342-7b0c-4d42-ae1e-710f4a32a1b2",
"756da756-1cf4-4382-841d-90367abd6c67",
"c9a5cc4e-6950-477f-9e58-ec69f91e72df",
"02b406f7-e448-4fae-aadd-89cf04dd2e53",
"162e1297-5c68-49e5-8075-33414ea72c77",
"a2624854-e23b-4305-abe0-940b73f76935",
"88970f6e-610e-40bb-b796-dbcc7f7f8cf6",
"33d0b4d1-7e8a-4413-bce1-27abc2500bfc",
"31ca07fd-4885-4d32-96ad-2cc87779b01c",
"a74c7bd4-4d1f-4066-9d39-f277c42b29f9",
"e662e047-7914-4356-8c32-6e54eedebd05"
],
"evidenceGaps": [],
"nextStep": "先完成 candidateExecutionBatch 中的 candidate decisions，再写 loop_report；不要只写泛泛 summary。"
},
"outcomeSimulator": {
"mode": "read_only_preflight",
"simulatedCount": 15,
"likelyOutcomeMix": [
{
"outcome": "no_action",
"count": 6
},
{
"outcome": "converted_to_source",
"count": 4
},
{
"outcome": "created_new_card",
"count": 5
}
],
"items": [
{
"candidateId": "47743b21-5878-499b-ac8a-d473a0513342",
"batchIndex": 1,
"urlClass": "web_page",
"suggestedOutcome": "no_action",
"confidence": "low",
"requiredConfirmation": [
"page_contains_public_tool_or_capability",
"evidence_is_stronger_than_generic_webpage"
],
"reason": "普通网页需要更强证据；看不到具体可接入能力时应跳过。"
},
{
"candidateId": "467d2960-b7c1-46b6-940d-45d81ff4b6e5",
"batchIndex": 2,
"urlClass": "web_page",
"suggestedOutcome": "no_action",
"confidence": "low",
"requiredConfirmation": [
"page_contains_public_tool_or_capability",
"evidence_is_stronger_than_generic_webpage"
],
"reason": "普通网页需要更强证据；看不到具体可接入能力时应跳过。"
},
{
"candidateId": "5144d0d7-b3a0-4600-ba8c-113f289e45c6",
"batchIndex": 3,
"urlClass": "web_page",
"suggestedOutcome": "no_action",
"confidence": "low",
"requiredConfirmation": [
"page_contains_public_tool_or_capability",
"evidence_is_stronger_than_generic_webpage"
],
"reason": "普通网页需要更强证据；看不到具体可接入能力时应跳过。"
},
{
"candidateId": "93f25c25-e779-4481-9d1b-9658bbbcab4a",
"batchIndex": 4,
"urlClass": "news_article",
"suggestedOutcome": "converted_to_source",
"confidence": "low",
"requiredConfirmation": [
"article_points_to_specific_tool_or_project",
"source_url_is_original_enough"
],
"reason": "新闻页通常更适合沉淀为 raw source；只有明确指向工具、项目或模型能力时才继续补卡。"
},
{
"candidateId": "8c516342-7b0c-4d42-ae1e-710f4a32a1b2",
"batchIndex": 5,
"urlClass": "news_article",
"suggestedOutcome": "converted_to_source",
"confidence": "low",
"requiredConfirmation": [
"article_points_to_specific_tool_or_project",
"source_url_is_original_enough"
],
"reason": "新闻页通常更适合沉淀为 raw source；只有明确指向工具、项目或模型能力时才继续补卡。"
},
{
"candidateId": "756da756-1cf4-4382-841d-90367abd6c67",
"batchIndex": 6,
"urlClass": "code_repository",
"suggestedOutcome": "created_new_card",
"confidence": "medium",
"requiredConfirmation": [
"repo_is_tool_mcp_skill_or_dev_productivity_capability",
"official_source_confirmed"
],
"reason": "代码仓库最有可能对应可接入工具、MCP、Skill 或开发效率能力；仍需打开仓库确认不是纯论文、样例或报告。"
},
{
"candidateId": "c9a5cc4e-6950-477f-9e58-ec69f91e72df",
"batchIndex": 7,
"urlClass": "code_repository",
"suggestedOutcome": "created_new_card",
"confidence": "medium",
"requiredConfirmation": [
"repo_is_tool_mcp_skill_or_dev_productivity_capability",
"official_source_confirmed"
],
"reason": "代码仓库最有可能对应可接入工具、MCP、Skill 或开发效率能力；仍需打开仓库确认不是纯论文、样例或报告。"
},
{
"candidateId": "02b406f7-e448-4fae-aadd-89cf04dd2e53",
"batchIndex": 8,
"urlClass": "code_repository",
"suggestedOutcome": "created_new_card",
"confidence": "medium",
"requiredConfirmation": [
"repo_is_tool_mcp_skill_or_dev_productivity_capability",
"official_source_confirmed"
],
"reason": "代码仓库最有可能对应可接入工具、MCP、Skill 或开发效率能力；仍需打开仓库确认不是纯论文、样例或报告。"
},
{
"candidateId": "162e1297-5c68-49e5-8075-33414ea72c77",
"batchIndex": 9,
"urlClass": "code_repository",
"suggestedOutcome": "created_new_card",
"confidence": "medium",
"requiredConfirmation": [
"repo_is_tool_mcp_skill_or_dev_productivity_capability",
"official_source_confirmed"
],
"reason": "代码仓库最有可能对应可接入工具、MCP、Skill 或开发效率能力；仍需打开仓库确认不是纯论文、样例或报告。"
},
{
"candidateId": "a2624854-e23b-4305-abe0-940b73f76935",
"batchIndex": 10,
"urlClass": "code_repository",
"suggestedOutcome": "created_new_card",
"confidence": "medium",
"requiredConfirmation": [
"repo_is_tool_mcp_skill_or_dev_productivity_capability",
"official_source_confirmed"
],
"reason": "代码仓库最有可能对应可接入工具、MCP、Skill 或开发效率能力；仍需打开仓库确认不是纯论文、样例或报告。"
},
{
"candidateId": "88970f6e-610e-40bb-b796-dbcc7f7f8cf6",
"batchIndex": 11,
"urlClass": "web_page",
"suggestedOutcome": "no_action",
"confidence": "low",
"requiredConfirmation": [
"page_contains_public_tool_or_capability",
"evidence_is_stronger_than_generic_webpage"
],
"reason": "普通网页需要更强证据；看不到具体可接入能力时应跳过。"
},
{
"candidateId": "33d0b4d1-7e8a-4413-bce1-27abc2500bfc",
"batchIndex": 12,
"urlClass": "news_article",
"suggestedOutcome": "converted_to_source",
"confidence": "low",
"requiredConfirmation": [
"article_points_to_specific_tool_or_project",
"source_url_is_original_enough"
],
"reason": "新闻页通常更适合沉淀为 raw source；只有明确指向工具、项目或模型能力时才继续补卡。"
},
{
"candidateId": "31ca07fd-4885-4d32-96ad-2cc87779b01c",
"batchIndex": 13,
"urlClass": "social_discussion",
"suggestedOutcome": "no_action",
"confidence": "low",
"requiredConfirmation": [
"original_project_or_official_source_found",
"social_post_not_standalone_rumor"
],
"reason": "社交帖只能作为线索；必须找到原始项目、论文、官网或官方仓库后再转来源或建卡。"
},
{
"candidateId": "a74c7bd4-4d1f-4066-9d39-f277c42b29f9",
"batchIndex": 14,
"urlClass": "news_article",
"suggestedOutcome": "converted_to_source",
"confidence": "low",
"requiredConfirmation": [
"article_points_to_specific_tool_or_project",
"source_url_is_original_enough"
],
"reason": "新闻页通常更适合沉淀为 raw source；只有明确指向工具、项目或模型能力时才继续补卡。"
},
{
"candidateId": "e662e047-7914-4356-8c32-6e54eedebd05",
"batchIndex": 15,
"urlClass": "web_page",
"suggestedOutcome": "no_action",
"confidence": "low",
"requiredConfirmation": [
"page_contains_public_tool_or_capability",
"evidence_is_stronger_than_generic_webpage"
],
"reason": "普通网页需要更强证据；看不到具体可接入能力时应跳过。"
}
],
"guardrail": "这是只读预估，不能直接写入 candidate decision；Hermes 必须打开原链接并带证据调用 toolkoi_candidate_decision。"
},
"automationReadiness": {
"status": "browser_review_required",
"canAutoWrite": false,
"candidateCount": 15,
"browserReviewRequiredCount": 15,
"evidenceRequired": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index"
],
"blockers": [
"simulator_is_not_evidence",
"candidate_decision_requires_original_url_review"
],
"nextStep": "逐条打开 candidateExecutionBatch 原链接，完成 evidence-backed decision 后再刷新 loop_contract。"
},
"browserReviewPlan": {
"mode": "original_url_review",
"browserRequired": true,
"reviewCount": 15,
"guardrail": "浏览器核查计划只产生证据，不自动写入 candidate decision。",
"items": [
{
"candidateId": "47743b21-5878-499b-ac8a-d473a0513342",
"batchIndex": 1,
"title": "1. The Trust Fracture: Claude's Dual Identity Crisis | 信任断裂：Claude 的双面危机——从水印风波到阿里全面禁用",
"url": "https://www.reuters.com/world/china/alibaba-ban-claude-code-workplace-over-alleged-backdoor-risks-source-says-2026-07-03/",
"urlHost": "reuters.com",
"urlClass": "web_page",
"checklist": [
"打开网页并确认是否存在明确公开的工具、产品、文档或接入能力。",
"检查页面是否提供官方来源、安装/使用方式或可验证的功能说明。",
"证据不足时不要建卡，优先 skipped/no_action。"
],
"passSignals": [
"页面直接说明可接入工具、Skill、MCP 或开发效率能力。",
"页面提供官方文档、安装链接、仓库或可验证功能。"
],
"skipSignals": [
"页面是泛介绍、目录页、营销页或无法确认实际能力。",
"没有足够公开证据支撑 Toolkoi 卡片或 raw source。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:1"
]
},
{
"candidateId": "467d2960-b7c1-46b6-940d-45d81ff4b6e5",
"batchIndex": 2,
"title": "3. AI Security: From Shield to Sword | 安全悖论：AI 既是盾牌，也是利刃",
"url": "https://the-decoder.com/security-vulnerability-reports-have-exploded-since-ai-models-started-hunting-for-bugs/",
"urlHost": "the-decoder.com",
"urlClass": "web_page",
"checklist": [
"打开网页并确认是否存在明确公开的工具、产品、文档或接入能力。",
"检查页面是否提供官方来源、安装/使用方式或可验证的功能说明。",
"证据不足时不要建卡，优先 skipped/no_action。"
],
"passSignals": [
"页面直接说明可接入工具、Skill、MCP 或开发效率能力。",
"页面提供官方文档、安装链接、仓库或可验证功能。"
],
"skipSignals": [
"页面是泛介绍、目录页、营销页或无法确认实际能力。",
"没有足够公开证据支撑 Toolkoi 卡片或 raw source。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:2"
]
},
{
"candidateId": "5144d0d7-b3a0-4600-ba8c-113f289e45c6",
"batchIndex": 3,
"title": "前沿大模型被曝存在同伴保护行为",
"url": "https://arxiv.org/abs/2604.19784",
"urlHost": "arxiv.org",
"urlClass": "web_page",
"checklist": [
"打开网页并确认是否存在明确公开的工具、产品、文档或接入能力。",
"检查页面是否提供官方来源、安装/使用方式或可验证的功能说明。",
"证据不足时不要建卡，优先 skipped/no_action。"
],
"passSignals": [
"页面直接说明可接入工具、Skill、MCP 或开发效率能力。",
"页面提供官方文档、安装链接、仓库或可验证功能。"
],
"skipSignals": [
"页面是泛介绍、目录页、营销页或无法确认实际能力。",
"没有足够公开证据支撑 Toolkoi 卡片或 raw source。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:3"
]
},
{
"candidateId": "93f25c25-e779-4481-9d1b-9658bbbcab4a",
"batchIndex": 4,
"title": "1. The Security Arms Race | AI安全从附属品升格为战略武器：GPT-5.6受控发布、Daybreak防御平台与政府审批风暴三线并行",
"url": "http://mp.weixin.qq.com/s?__biz=MjM5NjA0NjgyMA==&mid=2651341309&idx=1&sn=9afc5021305207db4b56448f2d298c80&chksm=bc4efd5d2cfde9f4d358f3c696aa65c7c341e85bb2058edb80ca29424675c1e6b473319271de&scene=0&xtrack=1#rd",
"urlHost": "mp.weixin.qq.com",
"urlClass": "news_article",
"checklist": [
"打开原始新闻页并确认具体提到的项目、工具、模型能力或官方链接。",
"寻找文章中的原始项目链接；没有原始来源时只能作为弱 raw source。",
"判断这条新闻是否和 Toolkoi 能力卡有关，而不是泛 AI 资讯。"
],
"passSignals": [
"文章明确指向工具、MCP、Skill、模型能力或官方发布。",
"能提取可回溯的原文 URL 和来源可信度。"
],
"skipSignals": [
"只是宏观观点、融资、榜单或无具体工具入口。",
"需要付费墙或登录且无法确认原文内容。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:4"
]
},
{
"candidateId": "8c516342-7b0c-4d42-ae1e-710f4a32a1b2",
"batchIndex": 5,
"title": "2. The Model Battleground Escalates | 模型军备竞赛白热化：字节豆包2.1、Google Gemini 3.5 Flash、智谱GLM-5.2混战升级",
"url": "https://mp.weixin.qq.com/s?__biz=MzI1MzYzMjE0MQ==&mid=2247520467&idx=1&sn=e6210ac0ffacb37351752b7fcb5259e5",
"urlHost": "mp.weixin.qq.com",
"urlClass": "news_article",
"checklist": [
"打开原始新闻页并确认具体提到的项目、工具、模型能力或官方链接。",
"寻找文章中的原始项目链接；没有原始来源时只能作为弱 raw source。",
"判断这条新闻是否和 Toolkoi 能力卡有关，而不是泛 AI 资讯。"
],
"passSignals": [
"文章明确指向工具、MCP、Skill、模型能力或官方发布。",
"能提取可回溯的原文 URL 和来源可信度。"
],
"skipSignals": [
"只是宏观观点、融资、榜单或无具体工具入口。",
"需要付费墙或登录且无法确认原文内容。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:5"
]
},
{
"candidateId": "756da756-1cf4-4382-841d-90367abd6c67",
"batchIndex": 6,
"title": "SkillOpt: Agent skills as trainable parameters",
"url": "https://github.com/microsoft/SkillOpt",
"urlHost": "github.com",
"urlClass": "code_repository",
"checklist": [
"打开原始 URL 并确认页面可访问。",
"确认仓库 README、描述、releases 或文档是否说明它是工具、MCP、Skill 或开发效率能力。",
"确认项目来源可信度和维护状态，不能只凭标题建卡。"
],
"passSignals": [
"README 或官方文档说明可安装、调用、接入或作为开发工具使用。",
"仓库 owner、stars、recent activity 或 release 能支撑 source_confidence。"
],
"skipSignals": [
"只是论文、新闻、榜单、样例报告或一次性 demo，不构成 Toolkoi 能力卡。",
"页面无法访问或没有足够公开证据。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:6"
]
},
{
"candidateId": "c9a5cc4e-6950-477f-9e58-ec69f91e72df",
"batchIndex": 7,
"title": "Memora: A Harmonic Memory Representation Balancing Abstraction and Specificity",
"url": "https://github.com/microsoft/Memora",
"urlHost": "github.com",
"urlClass": "code_repository",
"checklist": [
"打开原始 URL 并确认页面可访问。",
"确认仓库 README、描述、releases 或文档是否说明它是工具、MCP、Skill 或开发效率能力。",
"确认项目来源可信度和维护状态，不能只凭标题建卡。"
],
"passSignals": [
"README 或官方文档说明可安装、调用、接入或作为开发工具使用。",
"仓库 owner、stars、recent activity 或 release 能支撑 source_confidence。"
],
"skipSignals": [
"只是论文、新闻、榜单、样例报告或一次性 demo，不构成 Toolkoi 能力卡。",
"页面无法访问或没有足够公开证据。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:7"
]
},
{
"candidateId": "02b406f7-e448-4fae-aadd-89cf04dd2e53",
"batchIndex": 8,
"title": "Ire identifies another LOTUSLITE specimen",
"url": "https://github.com/microsoft/project-ire/blob/main/reports/47e51e82229e80a387c3cb100d39d3705e6360bbf9bfa1601dbc484e8d02e653.md",
"urlHost": "github.com",
"urlClass": "code_repository",
"checklist": [
"打开原始 URL 并确认页面可访问。",
"确认仓库 README、描述、releases 或文档是否说明它是工具、MCP、Skill 或开发效率能力。",
"确认项目来源可信度和维护状态，不能只凭标题建卡。"
],
"passSignals": [
"README 或官方文档说明可安装、调用、接入或作为开发工具使用。",
"仓库 owner、stars、recent activity 或 release 能支撑 source_confidence。"
],
"skipSignals": [
"只是论文、新闻、榜单、样例报告或一次性 demo，不构成 Toolkoi 能力卡。",
"页面无法访问或没有足够公开证据。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:8"
]
},
{
"candidateId": "162e1297-5c68-49e5-8075-33414ea72c77",
"batchIndex": 9,
"title": "Data Formulator 0.7: AI-powered data analytics for enterprise data",
"url": "https://github.com/microsoft/data-formulator",
"urlHost": "github.com",
"urlClass": "code_repository",
"checklist": [
"打开原始 URL 并确认页面可访问。",
"确认仓库 README、描述、releases 或文档是否说明它是工具、MCP、Skill 或开发效率能力。",
"确认项目来源可信度和维护状态，不能只凭标题建卡。"
],
"passSignals": [
"README 或官方文档说明可安装、调用、接入或作为开发工具使用。",
"仓库 owner、stars、recent activity 或 release 能支撑 source_confidence。"
],
"skipSignals": [
"只是论文、新闻、榜单、样例报告或一次性 demo，不构成 Toolkoi 能力卡。",
"页面无法访问或没有足够公开证据。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:9"
]
},
{
"candidateId": "a2624854-e23b-4305-abe0-940b73f76935",
"batchIndex": 10,
"title": "Evaluating performance and efficiency of the GitHub Copilot agentic harness across models and tasks",
"url": "https://github.com/github/copilot-sdk",
"urlHost": "github.com",
"urlClass": "code_repository",
"checklist": [
"打开原始 URL 并确认页面可访问。",
"确认仓库 README、描述、releases 或文档是否说明它是工具、MCP、Skill 或开发效率能力。",
"确认项目来源可信度和维护状态，不能只凭标题建卡。"
],
"passSignals": [
"README 或官方文档说明可安装、调用、接入或作为开发工具使用。",
"仓库 owner、stars、recent activity 或 release 能支撑 source_confidence。"
],
"skipSignals": [
"只是论文、新闻、榜单、样例报告或一次性 demo，不构成 Toolkoi 能力卡。",
"页面无法访问或没有足够公开证据。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:10"
]
},
{
"candidateId": "88970f6e-610e-40bb-b796-dbcc7f7f8cf6",
"batchIndex": 11,
"title": "语言模型中的全局工作空间",
"url": "https://www.anthropic.com/research/global-workspace",
"urlHost": "anthropic.com",
"urlClass": "web_page",
"checklist": [
"打开网页并确认是否存在明确公开的工具、产品、文档或接入能力。",
"检查页面是否提供官方来源、安装/使用方式或可验证的功能说明。",
"证据不足时不要建卡，优先 skipped/no_action。"
],
"passSignals": [
"页面直接说明可接入工具、Skill、MCP 或开发效率能力。",
"页面提供官方文档、安装链接、仓库或可验证功能。"
],
"skipSignals": [
"页面是泛介绍、目录页、营销页或无法确认实际能力。",
"没有足够公开证据支撑 Toolkoi 卡片或 raw source。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:11"
]
},
{
"candidateId": "33d0b4d1-7e8a-4413-bce1-27abc2500bfc",
"batchIndex": 12,
"title": "开辟新赛道：小米 YU7 GT 创全球首个纽北自动驾驶圈速纪录，官方圈速榜新增\"自动驾驶\"分类",
"url": "https://www.ithome.com/0/967/234.htm",
"urlHost": "ithome.com",
"urlClass": "news_article",
"checklist": [
"打开原始新闻页并确认具体提到的项目、工具、模型能力或官方链接。",
"寻找文章中的原始项目链接；没有原始来源时只能作为弱 raw source。",
"判断这条新闻是否和 Toolkoi 能力卡有关，而不是泛 AI 资讯。"
],
"passSignals": [
"文章明确指向工具、MCP、Skill、模型能力或官方发布。",
"能提取可回溯的原文 URL 和来源可信度。"
],
"skipSignals": [
"只是宏观观点、融资、榜单或无具体工具入口。",
"需要付费墙或登录且无法确认原文内容。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:12"
]
},
{
"candidateId": "31ca07fd-4885-4d32-96ad-2cc87779b01c",
"batchIndex": 13,
"title": "扎克伯格：建千兆瓦级AI集群，集中精英与资本",
"url": "https://x.com/rohanpaul_ai/status/2073834219659534816",
"urlHost": "x.com",
"urlClass": "social_discussion",
"checklist": [
"打开社交帖并识别它引用的原始项目、论文、官网或仓库。",
"不要把单条社交帖当作官方证据；先找二级原始来源。",
"确认讨论是否包含可验证的接入价值。"
],
"passSignals": [
"社交帖链接到官方项目、仓库、论文或产品页。",
"讨论内容能解释工具价值，并能被原始来源复核。"
],
"skipSignals": [
"只是观点、传闻、吐槽或没有原始链接。",
"无法确认作者、项目或具体工具能力。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:13"
]
},
{
"candidateId": "a74c7bd4-4d1f-4066-9d39-f277c42b29f9",
"batchIndex": 14,
"title": "Anthropic在Claude Code中植入隐写术代码识别中国用户",
"url": "https://mp.weixin.qq.com/s/yLb4T2UC16ebKHApdBbgWw",
"urlHost": "mp.weixin.qq.com",
"urlClass": "news_article",
"checklist": [
"打开原始新闻页并确认具体提到的项目、工具、模型能力或官方链接。",
"寻找文章中的原始项目链接；没有原始来源时只能作为弱 raw source。",
"判断这条新闻是否和 Toolkoi 能力卡有关，而不是泛 AI 资讯。"
],
"passSignals": [
"文章明确指向工具、MCP、Skill、模型能力或官方发布。",
"能提取可回溯的原文 URL 和来源可信度。"
],
"skipSignals": [
"只是宏观观点、融资、榜单或无具体工具入口。",
"需要付费墙或登录且无法确认原文内容。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:14"
]
},
{
"candidateId": "e662e047-7914-4356-8c32-6e54eedebd05",
"batchIndex": 15,
"title": "纽约时报修订诉讼，指控微软为OpenAI建造版权侵权超级计算机",
"url": "https://arstechnica.com/tech-policy/2026/06/microsoft-built-supercomputer-to-help-openai-infringe-copyrights-nyt-alleged",
"urlHost": "arstechnica.com",
"urlClass": "web_page",
"checklist": [
"打开网页并确认是否存在明确公开的工具、产品、文档或接入能力。",
"检查页面是否提供官方来源、安装/使用方式或可验证的功能说明。",
"证据不足时不要建卡，优先 skipped/no_action。"
],
"passSignals": [
"页面直接说明可接入工具、Skill、MCP 或开发效率能力。",
"页面提供官方文档、安装链接、仓库或可验证功能。"
],
"skipSignals": [
"页面是泛介绍、目录页、营销页或无法确认实际能力。",
"没有足够公开证据支撑 Toolkoi 卡片或 raw source。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:15"
]
}
]
},
"browserReviewReceiptSpec": {
"mode": "browser_evidence_receipt",
"expectedReceiptCount": 15,
"writePolicy": "read_only_before_candidate_decision"
},
"exitCriteria": [
"每个主动查看的 queued candidate 必须调用 toolkoi_candidate_decision，写入 outcome、reason 和 evidence。",
"创建新卡前必须有原始 URL、来源可信度判断和中文卡片字段。",
"不足以入库的候选必须写 skipped/no_action 原因。"
]
},
{
"order": 4,
"key": "quality_audit",
"title": "复查质量缺口",
"action": "运行 quality_audit 或等效检查，确认剩余低分卡片。",
"itemCount": 10,
"required": true,
"exitCriteria": [
"记录剩余缺口数量和下一轮优先项。"
]
},
{
"order": 5,
"key": "loop_report",
"title": "写回闭环报告",
"action": "调用 toolkoi_loop_report 写入 completed、blocked、skipped、handledSignals 和 summary。",
"itemCount": 1,
"required": true,
"exitCriteria": [
"loop_report 包含 consumed contractGeneratedAt。",
"报告不包含空 actions/evidence 的 completed 项。"
]
}
],
"stopConditions": [
"遇到生产凭据、不可公开来源或需要任意 shell/Docker 权限时停止并标记 blocked。",
"来源不足时只写 raw source 或 skipped，不发布 Wiki。",
"每轮必须写回 loop_report，否则下一轮视为闭环不健康。"
]
},
"sourceIntakeContract": {
"mode": "multi_source_to_llm_wiki",
"operator": "hermes",
"outputLayer": "llm_wiki_intake",
"writePolicy": "normalize_first_publish_after_review",
"inputKinds": [
"rss_feed",
"ai_daily",
"web_page",
"blog_post",
"community_aggregate"
],
"requiredFields": [
"sourceKey",
"sourceItemId",
"title",
"url",
"summaryZh",
"tags",
"sourceConfidence",
"candidateAction",
"rawSource"
],
"candidateActions": [
"queue_candidate",
"convert_to_raw_source",
"publish_wiki_page",
"skip"
],
"outputTargets": [
"toolkoi_candidates",
"tool_sources",
"tool_wiki_pages"
],
"normalizationSubmission": {
"method": "POST",
"endpoint": "/api/admin/toolkoi/source-intake-normalization",
"contentType": "application/json",
"auth": "TOOLKOI_ADMIN_TOKEN",
"writePolicy": "read_only_normalization_plan",
"expectedResponseFields": [
"actionCounts",
"items",
"guardrail",
"writePolicy"
]
},
"dedupSubmission": {
"method": "POST",
"endpoint": "/api/admin/toolkoi/source-intake-dedup",
"contentType": "application/json",
"auth": "TOOLKOI_ADMIN_TOKEN",
"writePolicy": "read_only_dedup_check",
"expectedResponseFields": [
"duplicateCount",
"newItemCount",
"items",
"guardrail"
]
},
"applyPlanSubmission": {
"method": "POST",
"endpoint": "/api/admin/toolkoi/source-intake-apply-plan",
"contentType": "application/json",
"auth": "TOOLKOI_ADMIN_TOKEN",
"writePolicy": "read_only_submit_plan",
"expectedResponseFields": [
"submitQueue",
"skippedItems",
"blockedItems",
"guardrail"
]
},
"applyAuditSubmission": {
"method": "POST",
"endpoint": "/api/admin/toolkoi/source-intake-apply-audit",
"contentType": "application/json",
"auth": "TOOLKOI_ADMIN_TOKEN",
"writePolicy": "read_only_apply_audit",
"expectedResponseFields": [
"matchingSubmitCount",
"missingSubmitCount",
"readyForLoopReport",
"missingItems"
]
},
"guardrails": [
"保留原始 URL；不要只保存聚合页跳转链接。",
"输出必须是结构化 JSON；不要把抓取过程写成对外播报。",
"入 Wiki 或卡片前必须经过来源可信度和回滚边界检查。"
],
"sources": [
{
"sourceKey": "rss",
"label": "RSS",
"inputKind": "rss_feed",
"queuedCount": 274,
"topCandidateCount": 5,
"status": "stale",
"normalizationInstruction": "RSS 聚合源按 entry 逐条解析，保留每条原始 URL、标题、摘要、发布时间和来源标签，再决定 queue_candidate 或 skip。"
},
{
"sourceKey": "hex2077",
"label": "HEX2077",
"inputKind": "web_page",
"queuedCount": 296,
"topCandidateCount": 5,
"status": "stale",
"normalizationInstruction": "HEX2077 页面用浏览器读取正文和出站链接，落库时保存真实原文 URL，不把跳转页当最终来源。"
},
{
"sourceKey": "aihot",
"label": "AI HOT",
"inputKind": "ai_daily",
"queuedCount": 196,
"topCandidateCount": 5,
"status": "stale",
"normalizationInstruction": "AI HOT 日报先抽取条目级原链接和中文摘要；只把明确指向工具、MCP、Skill、模型能力或项目的内容转入候选。"
}
]
},
"llmWikiOutputContract": {
"mode": "strict_llm_wiki_output",
"publishPolicy": "auto_publish_with_rollback_revision",
"rollbackRequired": true,
"outputTargets": [
"tool_sources",
"tool_wiki_pages",
"tool_relations",
"tool_wiki_revisions"
],
"rawSourceFields": [
"toolSlug",
"sourceType",
"title",
"url",
"rawContent",
"trustLevel",
"fetchedAt"
],
"wikiPageFields": [
"toolSlug",
"summary",
"bestFor",
"notFor",
"setupNotes",
"wikiLinks",
"sourceUrls"
],
"relationFields": [
"sourceSlug",
"targetSlug",
"relationReason",
"confidence"
],
"revisionFields": [
"toolSlug",
"publishedBy",
"model",
"createdAt",
"rollbackKey"
],
"guardrails": [
"所有明细卡片和 Wiki 页面说明必须输出中文。",
"每次自动发布必须留下 tool_wiki_revisions 回滚记录。",
"没有 raw source 或来源可信度不足时，只能输出 skip/blocked，不能发布 Wiki。"
],
"draftInputs": {
"existingToolDraftCount": 5,
"candidateDraftCount": 15,
"candidateSourceKeys": [
"hex2077",
"rss",
"aihot"
]
},
"completionCriteria": [
"raw source、wiki page、relation、revision 四类输出字段齐全或明确 skip。",
"每个自动发布项都能通过 rollbackKey 找到对应 revision。",
"二层关系必须带 relationReason 和 confidence。"
]
},
"llmWikiPublishGate": {
"status": "blocked_until_verified_outputs",
"canAutoPublish": false,
"draftCount": 20,
"requiredEvidence": [
"raw_source_url",
"source_confidence",
"chinese_card_or_wiki_copy",
"relation_confidence",
"rollback_key"
],
"blockers": [
"llm_wiki_output_contract_is_schema_only",
"candidate_browser_review_receipts_missing",
"candidate_decisions_missing"
],
"nextStep": "先完成 sourceIntakeContract 输出和 browser review receipt，再写 source/wiki/revision。"
},
"signalPriorities": [],
"candidateQueuePriorities": [
{
"sourceKey": "hex2077",
"label": "HEX2077",
"priority": 0,
"action": "candidate_research",
"reviewLimit": 5,
"queuedCount": 296,
"highScoreCount": 98,
"staleCount": 296,
"freshCount": 0,
"averageScore": 82,
"status": "stale",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 HEX2077 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。",
"topCandidates": [
{
"id": "47743b21-5878-499b-ac8a-d473a0513342",
"title": "1. The Trust Fracture: Claude's Dual Identity Crisis | 信任断裂：Claude 的双面危机——从水印风波到阿里全面禁用",
"url": "https://www.reuters.com/world/china/alibaba-ban-claude-code-workplace-over-alleged-backdoor-risks-source-says-2026-07-03/",
"score": 92,
"queuedAt": "2026-07-04T03:38:48.417Z"
},
{
"id": "467d2960-b7c1-46b6-940d-45d81ff4b6e5",
"title": "3. AI Security: From Shield to Sword | 安全悖论：AI 既是盾牌，也是利刃",
"url": "https://the-decoder.com/security-vulnerability-reports-have-exploded-since-ai-models-started-hunting-for-bugs/",
"score": 92,
"queuedAt": "2026-07-04T03:38:48.417Z"
},
{
"id": "5144d0d7-b3a0-4600-ba8c-113f289e45c6",
"title": "前沿大模型被曝存在同伴保护行为",
"url": "https://arxiv.org/abs/2604.19784",
"score": 92,
"queuedAt": "2026-07-04T03:38:48.417Z"
},
{
"id": "93f25c25-e779-4481-9d1b-9658bbbcab4a",
"title": "1. The Security Arms Race | AI安全从附属品升格为战略武器：GPT-5.6受控发布、Daybreak防御平台与政府审批风暴三线并行",
"url": "http://mp.weixin.qq.com/s?__biz=MjM5NjA0NjgyMA==&mid=2651341309&idx=1&sn=9afc5021305207db4b56448f2d298c80&chksm=bc4efd5d2cfde9f4d358f3c696aa65c7c341e85bb2058edb80ca29424675c1e6b473319271de&scene=0&xtrack=1#rd",
"score": 92,
"queuedAt": "2026-06-29T05:43:23.429Z"
},
{
"id": "8c516342-7b0c-4d42-ae1e-710f4a32a1b2",
"title": "2. The Model Battleground Escalates | 模型军备竞赛白热化：字节豆包2.1、Google Gemini 3.5 Flash、智谱GLM-5.2混战升级",
"url": "https://mp.weixin.qq.com/s?__biz=MzI1MzYzMjE0MQ==&mid=2247520467&idx=1&sn=e6210ac0ffacb37351752b7fcb5259e5",
"score": 92,
"queuedAt": "2026-06-29T05:43:23.429Z"
}
]
},
{
"sourceKey": "rss",
"label": "RSS",
"priority": 1,
"action": "candidate_research",
"reviewLimit": 5,
"queuedCount": 274,
"highScoreCount": 67,
"staleCount": 274,
"freshCount": 0,
"averageScore": 81,
"status": "stale",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 RSS 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。",
"topCandidates": [
{
"id": "756da756-1cf4-4382-841d-90367abd6c67",
"title": "SkillOpt: Agent skills as trainable parameters",
"url": "https://github.com/microsoft/SkillOpt",
"score": 100,
"queuedAt": "2026-06-30T17:22:18.753Z"
},
{
"id": "c9a5cc4e-6950-477f-9e58-ec69f91e72df",
"title": "Memora: A Harmonic Memory Representation Balancing Abstraction and Specificity",
"url": "https://github.com/microsoft/Memora",
"score": 100,
"queuedAt": "2026-06-29T21:18:56.775Z"
},
{
"id": "02b406f7-e448-4fae-aadd-89cf04dd2e53",
"title": "Ire identifies another LOTUSLITE specimen",
"url": "https://github.com/microsoft/project-ire/blob/main/reports/47e51e82229e80a387c3cb100d39d3705e6360bbf9bfa1601dbc484e8d02e653.md",
"score": 100,
"queuedAt": "2026-06-29T16:29:08.652Z"
},
{
"id": "162e1297-5c68-49e5-8075-33414ea72c77",
"title": "Data Formulator 0.7: AI-powered data analytics for enterprise data",
"url": "https://github.com/microsoft/data-formulator",
"score": 100,
"queuedAt": "2026-06-29T16:29:08.652Z"
},
{
"id": "a2624854-e23b-4305-abe0-940b73f76935",
"title": "Evaluating performance and efficiency of the GitHub Copilot agentic harness across models and tasks",
"url": "https://github.com/github/copilot-sdk",
"score": 100,
"queuedAt": "2026-06-29T16:22:15.520Z"
}
]
},
{
"sourceKey": "aihot",
"label": "AI HOT",
"priority": 2,
"action": "candidate_research",
"reviewLimit": 5,
"queuedCount": 196,
"highScoreCount": 2,
"staleCount": 196,
"freshCount": 0,
"averageScore": 74,
"status": "stale",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 AI HOT 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。",
"topCandidates": [
{
"id": "88970f6e-610e-40bb-b796-dbcc7f7f8cf6",
"title": "语言模型中的全局工作空间",
"url": "https://www.anthropic.com/research/global-workspace",
"score": 90,
"queuedAt": "2026-07-07T16:11:59.076Z"
},
{
"id": "33d0b4d1-7e8a-4413-bce1-27abc2500bfc",
"title": "开辟新赛道：小米 YU7 GT 创全球首个纽北自动驾驶圈速纪录，官方圈速榜新增\"自动驾驶\"分类",
"url": "https://www.ithome.com/0/967/234.htm",
"score": 85,
"queuedAt": "2026-06-24T15:24:25.903Z"
},
{
"id": "31ca07fd-4885-4d32-96ad-2cc87779b01c",
"title": "扎克伯格：建千兆瓦级AI集群，集中精英与资本",
"url": "https://x.com/rohanpaul_ai/status/2073834219659534816",
"score": 84,
"queuedAt": "2026-07-05T20:11:25.055Z"
},
{
"id": "a74c7bd4-4d1f-4066-9d39-f277c42b29f9",
"title": "Anthropic在Claude Code中植入隐写术代码识别中国用户",
"url": "https://mp.weixin.qq.com/s/yLb4T2UC16ebKHApdBbgWw",
"score": 84,
"queuedAt": "2026-07-01T02:13:04.838Z"
},
{
"id": "e662e047-7914-4356-8c32-6e54eedebd05",
"title": "纽约时报修订诉讼，指控微软为OpenAI建造版权侵权超级计算机",
"url": "https://arstechnica.com/tech-policy/2026/06/microsoft-built-supercomputer-to-help-openai-infringe-copyrights-nyt-alleged",
"score": 84,
"queuedAt": "2026-06-26T22:09:12.201Z"
}
]
}
],
"candidateDecisionPolicy": {
"mcpTool": "toolkoi_candidate_decision",
"candidatesSource": "candidateQueuePriorities[].topCandidates[].id",
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"allowedOutcomes": [
{
"outcome": "converted_to_source",
"status": "published"
},
{
"outcome": "updated_existing_card",
"status": "published"
},
{
"outcome": "created_new_card",
"status": "published"
},
{
"outcome": "skipped",
"status": "skipped"
},
{
"outcome": "no_action",
"status": "skipped"
},
{
"outcome": "blocked",
"status": "failed"
}
],
"evidenceRequired": [
"original_url",
"source_confidence",
"decision_reason"
],
"writeRule": "每个被打开研究的 queued candidate 只写一条决策；不要批量猜测、不要为未查看候选造决策。",
"noBulkDecision": true
},
"recentCandidateDecisions": [],
"candidateDecisionQueuePreview": [
{
"candidateId": "47743b21-5878-499b-ac8a-d473a0513342",
"sourceKey": "hex2077",
"sourceLabel": "HEX2077",
"priority": 0,
"title": "1. The Trust Fracture: Claude's Dual Identity Crisis | 信任断裂：Claude 的双面危机——从水印风波到阿里全面禁用",
"url": "https://www.reuters.com/world/china/alibaba-ban-claude-code-workplace-over-alleged-backdoor-risks-source-says-2026-07-03/",
"score": 92,
"queuedAt": "2026-07-04T03:38:48.417Z",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 HEX2077 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"candidateId": "467d2960-b7c1-46b6-940d-45d81ff4b6e5",
"sourceKey": "hex2077",
"sourceLabel": "HEX2077",
"priority": 0,
"title": "3. AI Security: From Shield to Sword | 安全悖论：AI 既是盾牌，也是利刃",
"url": "https://the-decoder.com/security-vulnerability-reports-have-exploded-since-ai-models-started-hunting-for-bugs/",
"score": 92,
"queuedAt": "2026-07-04T03:38:48.417Z",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 HEX2077 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"candidateId": "5144d0d7-b3a0-4600-ba8c-113f289e45c6",
"sourceKey": "hex2077",
"sourceLabel": "HEX2077",
"priority": 0,
"title": "前沿大模型被曝存在同伴保护行为",
"url": "https://arxiv.org/abs/2604.19784",
"score": 92,
"queuedAt": "2026-07-04T03:38:48.417Z",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 HEX2077 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"candidateId": "93f25c25-e779-4481-9d1b-9658bbbcab4a",
"sourceKey": "hex2077",
"sourceLabel": "HEX2077",
"priority": 0,
"title": "1. The Security Arms Race | AI安全从附属品升格为战略武器：GPT-5.6受控发布、Daybreak防御平台与政府审批风暴三线并行",
"url": "http://mp.weixin.qq.com/s?__biz=MjM5NjA0NjgyMA==&mid=2651341309&idx=1&sn=9afc5021305207db4b56448f2d298c80&chksm=bc4efd5d2cfde9f4d358f3c696aa65c7c341e85bb2058edb80ca29424675c1e6b473319271de&scene=0&xtrack=1#rd",
"score": 92,
"queuedAt": "2026-06-29T05:43:23.429Z",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 HEX2077 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"candidateId": "8c516342-7b0c-4d42-ae1e-710f4a32a1b2",
"sourceKey": "hex2077",
"sourceLabel": "HEX2077",
"priority": 0,
"title": "2. The Model Battleground Escalates | 模型军备竞赛白热化：字节豆包2.1、Google Gemini 3.5 Flash、智谱GLM-5.2混战升级",
"url": "https://mp.weixin.qq.com/s?__biz=MzI1MzYzMjE0MQ==&mid=2247520467&idx=1&sn=e6210ac0ffacb37351752b7fcb5259e5",
"score": 92,
"queuedAt": "2026-06-29T05:43:23.429Z",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 HEX2077 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"candidateId": "756da756-1cf4-4382-841d-90367abd6c67",
"sourceKey": "rss",
"sourceLabel": "RSS",
"priority": 1,
"title": "SkillOpt: Agent skills as trainable parameters",
"url": "https://github.com/microsoft/SkillOpt",
"score": 100,
"queuedAt": "2026-06-30T17:22:18.753Z",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 RSS 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"candidateId": "c9a5cc4e-6950-477f-9e58-ec69f91e72df",
"sourceKey": "rss",
"sourceLabel": "RSS",
"priority": 1,
"title": "Memora: A Harmonic Memory Representation Balancing Abstraction and Specificity",
"url": "https://github.com/microsoft/Memora",
"score": 100,
"queuedAt": "2026-06-29T21:18:56.775Z",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 RSS 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"candidateId": "02b406f7-e448-4fae-aadd-89cf04dd2e53",
"sourceKey": "rss",
"sourceLabel": "RSS",
"priority": 1,
"title": "Ire identifies another LOTUSLITE specimen",
"url": "https://github.com/microsoft/project-ire/blob/main/reports/47e51e82229e80a387c3cb100d39d3705e6360bbf9bfa1601dbc484e8d02e653.md",
"score": 100,
"queuedAt": "2026-06-29T16:29:08.652Z",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 RSS 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
}
],
"candidateDecisionGap": {
"queuedTotal": 766,
"topCandidateWindow": 15,
"pendingDecisionCount": 15,
"recentDecisionCount": 0,
"outsideTopWindow": 751,
"coveragePercent": 2,
"estimatedReviewCycles": 52,
"backlogPressure": "critical",
"status": "backlog",
"reason": "还有 751 条 queued candidate 未进入当前 topCandidates 窗口；Hermes 需要逐轮消化或扩大 reviewLimit。"
},
"candidateCapacityPlan": {
"mode": "capacity_expansion",
"targetDecisionsPerRun": 15,
"suggestedReviewWindow": 20,
"suggestedCadenceMinutes": 30,
"sourceTargets": [
{
"sourceKey": "hex2077",
"label": "HEX2077",
"priority": 0,
"targetDecisions": 5,
"visibleCandidates": 5,
"queuedCount": 296,
"reviewLimit": 5,
"reason": "本轮处理 HEX2077 的 5 条可见候选；逐条打开原链接并写入 evidence-backed decision。"
},
{
"sourceKey": "rss",
"label": "RSS",
"priority": 1,
"targetDecisions": 5,
"visibleCandidates": 5,
"queuedCount": 274,
"reviewLimit": 5,
"reason": "本轮处理 RSS 的 5 条可见候选；逐条打开原链接并写入 evidence-backed decision。"
},
{
"sourceKey": "aihot",
"label": "AI HOT",
"priority": 2,
"targetDecisions": 5,
"visibleCandidates": 5,
"queuedCount": 196,
"reviewLimit": 5,
"reason": "本轮处理 AI HOT 的 5 条可见候选；逐条打开原链接并写入 evidence-backed decision。"
}
],
"actions": [
"运行 candidate_decision_only pass，先清 visible topCandidates，再刷新合同窗口。",
"保持 toolkoi_candidate_decision 单条写入；禁止批量猜测或为未打开候选造决策。",
"若连续两轮 outsideTopWindow 未下降，优先提升来源质量过滤或继续扩大候选窗口。"
],
"reason": "当前按可见窗口预计还需 52 轮，候选处理能力不足，需要进入候选决策专注模式。"
},
"candidateExecutionRunbook": {
"mode": "candidate_decision_only",
"decisionTool": "toolkoi_candidate_decision",
"candidatesSource": "dispatchPlan.phases[key=candidate_queue].capacityPlan.sourceTargets + candidateQueuePriorities[].topCandidates",
"refreshAfterDecisions": 15,
"steps": [
{
"order": 1,
"action": "read_capacity_targets",
"required": true,
"instruction": "读取 candidate_queue phase 的 capacityPlan.sourceTargets，按 priority 从小到大处理 targetDecisions > 0 的来源。"
},
{
"order": 2,
"action": "open_candidate_source",
"required": true,
"instruction": "从对应 candidateQueuePriorities[].topCandidates 取 candidate id 和原始 URL；每条都必须打开原链接并判断来源可信度。"
},
{
"order": 3,
"action": "write_single_candidate_decision",
"required": true,
"instruction": "每研究完一条 candidate，立即调用 toolkoi_candidate_decision 写入 outcome、reason、evidence、decidedBy。"
},
{
"order": 4,
"action": "refresh_loop_contract",
"required": true,
"instruction": "达到 refreshAfterDecisions 或可见候选耗尽后，重新读取 toolkoi_loop_contract，禁止继续按旧窗口处理。"
}
],
"guardrails": [
"禁止批量写入 candidate decision。",
"不能为未打开原链接的 candidate 造 outcome。",
"sourceTargets 里 targetDecisions 为 0 的来源只能刷新窗口或检查来源质量，不能写决策。"
],
"completionCriteria": [
"本轮写入的 candidate decision 数量达到 refreshAfterDecisions，或所有 visible topCandidates 已处理。",
"每条写入都有 original_url、source_confidence、decision_reason 证据。",
"处理后必须刷新 loop_contract 并重新计算下一轮 sourceTargets。"
]
},
"candidateExecutionBatch": [
{
"batchIndex": 1,
"sourceKey": "hex2077",
"sourceLabel": "HEX2077",
"priority": 0,
"candidateId": "47743b21-5878-499b-ac8a-d473a0513342",
"title": "1. The Trust Fracture: Claude's Dual Identity Crisis | 信任断裂：Claude 的双面危机——从水印风波到阿里全面禁用",
"url": "https://www.reuters.com/world/china/alibaba-ban-claude-code-workplace-over-alleged-backdoor-risks-source-says-2026-07-03/",
"urlHost": "reuters.com",
"urlClass": "web_page",
"decisionHint": "打开页面确认是否有官方产品、文档、仓库或可接入能力；证据不足时不要建卡。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "47743b21-5878-499b-ac8a-d473a0513342",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://www.reuters.com/world/china/alibaba-ban-claude-code-workplace-over-alleged-backdoor-risks-source-says-2026-07-03/",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:1"
],
"payloadSkeleton": {
"candidateId": "47743b21-5878-499b-ac8a-d473a0513342",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://www.reuters.com/world/china/alibaba-ban-claude-code-workplace-over-alleged-backdoor-risks-source-says-2026-07-03/",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:1"
],
"decidedBy": "hermes"
}
},
"score": 92,
"queuedAt": "2026-07-04T03:38:48.417Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 HEX2077 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 2,
"sourceKey": "hex2077",
"sourceLabel": "HEX2077",
"priority": 0,
"candidateId": "467d2960-b7c1-46b6-940d-45d81ff4b6e5",
"title": "3. AI Security: From Shield to Sword | 安全悖论：AI 既是盾牌，也是利刃",
"url": "https://the-decoder.com/security-vulnerability-reports-have-exploded-since-ai-models-started-hunting-for-bugs/",
"urlHost": "the-decoder.com",
"urlClass": "web_page",
"decisionHint": "打开页面确认是否有官方产品、文档、仓库或可接入能力；证据不足时不要建卡。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "467d2960-b7c1-46b6-940d-45d81ff4b6e5",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://the-decoder.com/security-vulnerability-reports-have-exploded-since-ai-models-started-hunting-for-bugs/",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:2"
],
"payloadSkeleton": {
"candidateId": "467d2960-b7c1-46b6-940d-45d81ff4b6e5",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://the-decoder.com/security-vulnerability-reports-have-exploded-since-ai-models-started-hunting-for-bugs/",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:2"
],
"decidedBy": "hermes"
}
},
"score": 92,
"queuedAt": "2026-07-04T03:38:48.417Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 HEX2077 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 3,
"sourceKey": "hex2077",
"sourceLabel": "HEX2077",
"priority": 0,
"candidateId": "5144d0d7-b3a0-4600-ba8c-113f289e45c6",
"title": "前沿大模型被曝存在同伴保护行为",
"url": "https://arxiv.org/abs/2604.19784",
"urlHost": "arxiv.org",
"urlClass": "web_page",
"decisionHint": "打开页面确认是否有官方产品、文档、仓库或可接入能力；证据不足时不要建卡。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "5144d0d7-b3a0-4600-ba8c-113f289e45c6",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://arxiv.org/abs/2604.19784",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:3"
],
"payloadSkeleton": {
"candidateId": "5144d0d7-b3a0-4600-ba8c-113f289e45c6",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://arxiv.org/abs/2604.19784",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:3"
],
"decidedBy": "hermes"
}
},
"score": 92,
"queuedAt": "2026-07-04T03:38:48.417Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 HEX2077 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 4,
"sourceKey": "hex2077",
"sourceLabel": "HEX2077",
"priority": 0,
"candidateId": "93f25c25-e779-4481-9d1b-9658bbbcab4a",
"title": "1. The Security Arms Race | AI安全从附属品升格为战略武器：GPT-5.6受控发布、Daybreak防御平台与政府审批风暴三线并行",
"url": "http://mp.weixin.qq.com/s?__biz=MjM5NjA0NjgyMA==&mid=2651341309&idx=1&sn=9afc5021305207db4b56448f2d298c80&chksm=bc4efd5d2cfde9f4d358f3c696aa65c7c341e85bb2058edb80ca29424675c1e6b473319271de&scene=0&xtrack=1#rd",
"urlHost": "mp.weixin.qq.com",
"urlClass": "news_article",
"decisionHint": "按情报线索处理：只在文章明确指向工具、MCP、Skill、模型能力或官方项目时转入 raw source；泛新闻应跳过。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "93f25c25-e779-4481-9d1b-9658bbbcab4a",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:http://mp.weixin.qq.com/s?__biz=MjM5NjA0NjgyMA==&mid=2651341309&idx=1&sn=9afc5021305207db4b56448f2d298c80&chksm=bc4efd5d2cfde9f4d358f3c696aa65c7c341e85bb2058edb80ca29424675c1e6b473319271de&scene=0&xtrack=1#rd",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:4"
],
"payloadSkeleton": {
"candidateId": "93f25c25-e779-4481-9d1b-9658bbbcab4a",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:http://mp.weixin.qq.com/s?__biz=MjM5NjA0NjgyMA==&mid=2651341309&idx=1&sn=9afc5021305207db4b56448f2d298c80&chksm=bc4efd5d2cfde9f4d358f3c696aa65c7c341e85bb2058edb80ca29424675c1e6b473319271de&scene=0&xtrack=1#rd",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:4"
],
"decidedBy": "hermes"
}
},
"score": 92,
"queuedAt": "2026-06-29T05:43:23.429Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 HEX2077 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 5,
"sourceKey": "hex2077",
"sourceLabel": "HEX2077",
"priority": 0,
"candidateId": "8c516342-7b0c-4d42-ae1e-710f4a32a1b2",
"title": "2. The Model Battleground Escalates | 模型军备竞赛白热化：字节豆包2.1、Google Gemini 3.5 Flash、智谱GLM-5.2混战升级",
"url": "https://mp.weixin.qq.com/s?__biz=MzI1MzYzMjE0MQ==&mid=2247520467&idx=1&sn=e6210ac0ffacb37351752b7fcb5259e5",
"urlHost": "mp.weixin.qq.com",
"urlClass": "news_article",
"decisionHint": "按情报线索处理：只在文章明确指向工具、MCP、Skill、模型能力或官方项目时转入 raw source；泛新闻应跳过。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "8c516342-7b0c-4d42-ae1e-710f4a32a1b2",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://mp.weixin.qq.com/s?__biz=MzI1MzYzMjE0MQ==&mid=2247520467&idx=1&sn=e6210ac0ffacb37351752b7fcb5259e5",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:5"
],
"payloadSkeleton": {
"candidateId": "8c516342-7b0c-4d42-ae1e-710f4a32a1b2",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://mp.weixin.qq.com/s?__biz=MzI1MzYzMjE0MQ==&mid=2247520467&idx=1&sn=e6210ac0ffacb37351752b7fcb5259e5",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:5"
],
"decidedBy": "hermes"
}
},
"score": 92,
"queuedAt": "2026-06-29T05:43:23.429Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 HEX2077 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 6,
"sourceKey": "rss",
"sourceLabel": "RSS",
"priority": 1,
"candidateId": "756da756-1cf4-4382-841d-90367abd6c67",
"title": "SkillOpt: Agent skills as trainable parameters",
"url": "https://github.com/microsoft/SkillOpt",
"urlHost": "github.com",
"urlClass": "code_repository",
"decisionHint": "优先检查仓库是否是可接入工具、MCP、Skill 或开发效率能力；若只是论文/样例报告，改写为 raw source 或 skipped。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "756da756-1cf4-4382-841d-90367abd6c67",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://github.com/microsoft/SkillOpt",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:6"
],
"payloadSkeleton": {
"candidateId": "756da756-1cf4-4382-841d-90367abd6c67",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://github.com/microsoft/SkillOpt",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:6"
],
"decidedBy": "hermes"
}
},
"score": 100,
"queuedAt": "2026-06-30T17:22:18.753Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 RSS 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 7,
"sourceKey": "rss",
"sourceLabel": "RSS",
"priority": 1,
"candidateId": "c9a5cc4e-6950-477f-9e58-ec69f91e72df",
"title": "Memora: A Harmonic Memory Representation Balancing Abstraction and Specificity",
"url": "https://github.com/microsoft/Memora",
"urlHost": "github.com",
"urlClass": "code_repository",
"decisionHint": "优先检查仓库是否是可接入工具、MCP、Skill 或开发效率能力；若只是论文/样例报告，改写为 raw source 或 skipped。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "c9a5cc4e-6950-477f-9e58-ec69f91e72df",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://github.com/microsoft/Memora",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:7"
],
"payloadSkeleton": {
"candidateId": "c9a5cc4e-6950-477f-9e58-ec69f91e72df",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://github.com/microsoft/Memora",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:7"
],
"decidedBy": "hermes"
}
},
"score": 100,
"queuedAt": "2026-06-29T21:18:56.775Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 RSS 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 8,
"sourceKey": "rss",
"sourceLabel": "RSS",
"priority": 1,
"candidateId": "02b406f7-e448-4fae-aadd-89cf04dd2e53",
"title": "Ire identifies another LOTUSLITE specimen",
"url": "https://github.com/microsoft/project-ire/blob/main/reports/47e51e82229e80a387c3cb100d39d3705e6360bbf9bfa1601dbc484e8d02e653.md",
"urlHost": "github.com",
"urlClass": "code_repository",
"decisionHint": "优先检查仓库是否是可接入工具、MCP、Skill 或开发效率能力；若只是论文/样例报告，改写为 raw source 或 skipped。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "02b406f7-e448-4fae-aadd-89cf04dd2e53",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://github.com/microsoft/project-ire/blob/main/reports/47e51e82229e80a387c3cb100d39d3705e6360bbf9bfa1601dbc484e8d02e653.md",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:8"
],
"payloadSkeleton": {
"candidateId": "02b406f7-e448-4fae-aadd-89cf04dd2e53",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://github.com/microsoft/project-ire/blob/main/reports/47e51e82229e80a387c3cb100d39d3705e6360bbf9bfa1601dbc484e8d02e653.md",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:8"
],
"decidedBy": "hermes"
}
},
"score": 100,
"queuedAt": "2026-06-29T16:29:08.652Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 RSS 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 9,
"sourceKey": "rss",
"sourceLabel": "RSS",
"priority": 1,
"candidateId": "162e1297-5c68-49e5-8075-33414ea72c77",
"title": "Data Formulator 0.7: AI-powered data analytics for enterprise data",
"url": "https://github.com/microsoft/data-formulator",
"urlHost": "github.com",
"urlClass": "code_repository",
"decisionHint": "优先检查仓库是否是可接入工具、MCP、Skill 或开发效率能力；若只是论文/样例报告，改写为 raw source 或 skipped。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "162e1297-5c68-49e5-8075-33414ea72c77",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://github.com/microsoft/data-formulator",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:9"
],
"payloadSkeleton": {
"candidateId": "162e1297-5c68-49e5-8075-33414ea72c77",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://github.com/microsoft/data-formulator",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:9"
],
"decidedBy": "hermes"
}
},
"score": 100,
"queuedAt": "2026-06-29T16:29:08.652Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 RSS 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 10,
"sourceKey": "rss",
"sourceLabel": "RSS",
"priority": 1,
"candidateId": "a2624854-e23b-4305-abe0-940b73f76935",
"title": "Evaluating performance and efficiency of the GitHub Copilot agentic harness across models and tasks",
"url": "https://github.com/github/copilot-sdk",
"urlHost": "github.com",
"urlClass": "code_repository",
"decisionHint": "优先检查仓库是否是可接入工具、MCP、Skill 或开发效率能力；若只是论文/样例报告，改写为 raw source 或 skipped。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "a2624854-e23b-4305-abe0-940b73f76935",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://github.com/github/copilot-sdk",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:10"
],
"payloadSkeleton": {
"candidateId": "a2624854-e23b-4305-abe0-940b73f76935",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://github.com/github/copilot-sdk",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:10"
],
"decidedBy": "hermes"
}
},
"score": 100,
"queuedAt": "2026-06-29T16:22:15.520Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 RSS 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 11,
"sourceKey": "aihot",
"sourceLabel": "AI HOT",
"priority": 2,
"candidateId": "88970f6e-610e-40bb-b796-dbcc7f7f8cf6",
"title": "语言模型中的全局工作空间",
"url": "https://www.anthropic.com/research/global-workspace",
"urlHost": "anthropic.com",
"urlClass": "web_page",
"decisionHint": "打开页面确认是否有官方产品、文档、仓库或可接入能力；证据不足时不要建卡。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "88970f6e-610e-40bb-b796-dbcc7f7f8cf6",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://www.anthropic.com/research/global-workspace",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:11"
],
"payloadSkeleton": {
"candidateId": "88970f6e-610e-40bb-b796-dbcc7f7f8cf6",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://www.anthropic.com/research/global-workspace",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:11"
],
"decidedBy": "hermes"
}
},
"score": 90,
"queuedAt": "2026-07-07T16:11:59.076Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 AI HOT 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 12,
"sourceKey": "aihot",
"sourceLabel": "AI HOT",
"priority": 2,
"candidateId": "33d0b4d1-7e8a-4413-bce1-27abc2500bfc",
"title": "开辟新赛道：小米 YU7 GT 创全球首个纽北自动驾驶圈速纪录，官方圈速榜新增\"自动驾驶\"分类",
"url": "https://www.ithome.com/0/967/234.htm",
"urlHost": "ithome.com",
"urlClass": "news_article",
"decisionHint": "按情报线索处理：只在文章明确指向工具、MCP、Skill、模型能力或官方项目时转入 raw source；泛新闻应跳过。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "33d0b4d1-7e8a-4413-bce1-27abc2500bfc",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://www.ithome.com/0/967/234.htm",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:12"
],
"payloadSkeleton": {
"candidateId": "33d0b4d1-7e8a-4413-bce1-27abc2500bfc",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://www.ithome.com/0/967/234.htm",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:12"
],
"decidedBy": "hermes"
}
},
"score": 85,
"queuedAt": "2026-06-24T15:24:25.903Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 AI HOT 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 13,
"sourceKey": "aihot",
"sourceLabel": "AI HOT",
"priority": 2,
"candidateId": "31ca07fd-4885-4d32-96ad-2cc87779b01c",
"title": "扎克伯格：建千兆瓦级AI集群，集中精英与资本",
"url": "https://x.com/rohanpaul_ai/status/2073834219659534816",
"urlHost": "x.com",
"urlClass": "social_discussion",
"decisionHint": "按二级信号处理：先找原始项目、论文、官网或官方仓库；找不到强来源时写 skipped/no_action。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "31ca07fd-4885-4d32-96ad-2cc87779b01c",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://x.com/rohanpaul_ai/status/2073834219659534816",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:13"
],
"payloadSkeleton": {
"candidateId": "31ca07fd-4885-4d32-96ad-2cc87779b01c",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://x.com/rohanpaul_ai/status/2073834219659534816",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:13"
],
"decidedBy": "hermes"
}
},
"score": 84,
"queuedAt": "2026-07-05T20:11:25.055Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 AI HOT 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 14,
"sourceKey": "aihot",
"sourceLabel": "AI HOT",
"priority": 2,
"candidateId": "a74c7bd4-4d1f-4066-9d39-f277c42b29f9",
"title": "Anthropic在Claude Code中植入隐写术代码识别中国用户",
"url": "https://mp.weixin.qq.com/s/yLb4T2UC16ebKHApdBbgWw",
"urlHost": "mp.weixin.qq.com",
"urlClass": "news_article",
"decisionHint": "按情报线索处理：只在文章明确指向工具、MCP、Skill、模型能力或官方项目时转入 raw source；泛新闻应跳过。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "a74c7bd4-4d1f-4066-9d39-f277c42b29f9",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://mp.weixin.qq.com/s/yLb4T2UC16ebKHApdBbgWw",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:14"
],
"payloadSkeleton": {
"candidateId": "a74c7bd4-4d1f-4066-9d39-f277c42b29f9",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://mp.weixin.qq.com/s/yLb4T2UC16ebKHApdBbgWw",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:14"
],
"decidedBy": "hermes"
}
},
"score": 84,
"queuedAt": "2026-07-01T02:13:04.838Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 AI HOT 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
},
{
"batchIndex": 15,
"sourceKey": "aihot",
"sourceLabel": "AI HOT",
"priority": 2,
"candidateId": "e662e047-7914-4356-8c32-6e54eedebd05",
"title": "纽约时报修订诉讼，指控微软为OpenAI建造版权侵权超级计算机",
"url": "https://arstechnica.com/tech-policy/2026/06/microsoft-built-supercomputer-to-help-openai-infringe-copyrights-nyt-alleged",
"urlHost": "arstechnica.com",
"urlClass": "web_page",
"decisionHint": "打开页面确认是否有官方产品、文档、仓库或可接入能力；证据不足时不要建卡。",
"decisionTemplate": {
"tool": "toolkoi_candidate_decision",
"candidateId": "e662e047-7914-4356-8c32-6e54eedebd05",
"allowedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"requiredFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"evidenceTemplate": [
"original_url:https://arstechnica.com/tech-policy/2026/06/microsoft-built-supercomputer-to-help-openai-infringe-copyrights-nyt-alleged",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:15"
],
"payloadSkeleton": {
"candidateId": "e662e047-7914-4356-8c32-6e54eedebd05",
"outcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"reason": "<中文判断原因>",
"evidence": [
"original_url:https://arstechnica.com/tech-policy/2026/06/microsoft-built-supercomputer-to-help-openai-infringe-copyrights-nyt-alleged",
"source_confidence:<high|medium|low>",
"decision_reason:<why_this_outcome>",
"candidate_batch_index:15"
],
"decidedBy": "hermes"
}
},
"score": 84,
"queuedAt": "2026-06-26T22:09:12.201Z",
"decisionTool": "toolkoi_candidate_decision",
"requiredEvidence": [
"original_url",
"source_confidence",
"handled_signal_or_candidate_outcome"
],
"nextStep": "从 AI HOT 最高分候选开始打开原始 URL；只在来源强、可解释、可回滚时创建卡片、raw source 或 Wiki。"
}
],
"candidateExecutionSummary": {
"itemCount": 15,
"activeSourceCount": 3,
"emptyTargetCount": 0,
"decisionTool": "toolkoi_candidate_decision",
"refreshAfterDecisions": 15,
"highestScoreCandidate": {
"candidateId": "756da756-1cf4-4382-841d-90367abd6c67",
"sourceKey": "rss",
"title": "SkillOpt: Agent skills as trainable parameters",
"url": "https://github.com/microsoft/SkillOpt",
"score": 100
},
"oldestCandidate": {
"candidateId": "33d0b4d1-7e8a-4413-bce1-27abc2500bfc",
"sourceKey": "aihot",
"title": "开辟新赛道：小米 YU7 GT 创全球首个纽北自动驾驶圈速纪录，官方圈速榜新增\"自动驾驶\"分类",
"url": "https://www.ithome.com/0/967/234.htm",
"queuedAt": "2026-06-24T15:24:25.903Z"
},
"sourceBreakdown": [
{
"sourceKey": "hex2077",
"label": "HEX2077",
"priority": 0,
"targetDecisions": 5,
"batchCount": 5,
"queuedCount": 296,
"highestScore": 92,
"oldestQueuedAt": "2026-06-29T05:43:23.429Z",
"firstCandidateId": "47743b21-5878-499b-ac8a-d473a0513342"
},
{
"sourceKey": "rss",
"label": "RSS",
"priority": 1,
"targetDecisions": 5,
"batchCount": 5,
"queuedCount": 274,
"highestScore": 100,
"oldestQueuedAt": "2026-06-29T16:22:15.520Z",
"firstCandidateId": "756da756-1cf4-4382-841d-90367abd6c67"
},
{
"sourceKey": "aihot",
"label": "AI HOT",
"priority": 2,
"targetDecisions": 5,
"batchCount": 5,
"queuedCount": 196,
"highestScore": 90,
"oldestQueuedAt": "2026-06-24T15:24:25.903Z",
"firstCandidateId": "88970f6e-610e-40bb-b796-dbcc7f7f8cf6"
}
],
"operatorNote": "按 batchIndex 顺序逐条打开原链接；每写入 toolkoi_candidate_decision 后记录 evidence；处理 15 条或批次耗尽后刷新 loop_contract。"
},
"candidateBatchReportGate": {
"reportTool": "toolkoi_loop_report",
"consumedContractGeneratedAt": "2026-09-19T07:17:37.700Z",
"expectedDecisionCount": 15,
"refreshRequiredAfterDecisions": 15,
"acceptedDecisionTool": "toolkoi_candidate_decision",
"requiredCompletedFields": [
"candidateId",
"outcome",
"reason",
"evidence",
"decidedBy"
],
"requiredEvidenceFields": [
"original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index"
],
"sourceQuotas": [
{
"sourceKey": "hex2077",
"label": "HEX2077",
"expectedDecisionCount": 5,
"firstCandidateId": "47743b21-5878-499b-ac8a-d473a0513342"
},
{
"sourceKey": "rss",
"label": "RSS",
"expectedDecisionCount": 5,
"firstCandidateId": "756da756-1cf4-4382-841d-90367abd6c67"
},
{
"sourceKey": "aihot",
"label": "AI HOT",
"expectedDecisionCount": 5,
"firstCandidateId": "88970f6e-610e-40bb-b796-dbcc7f7f8cf6"
}
],
"failureModes": [
"opened_candidate_without_decision",
"decision_without_required_evidence",
"continued_after_refresh_threshold"
],
"loopReportInstruction": "处理 candidateExecutionBatch 后必须调用 toolkoi_loop_report；completed/skipped/blocked 需要覆盖每条已打开候选，按 sourceQuotas 交账，不要只写泛泛 summary。达到 15 条后必须刷新 loop_contract。"
},
"candidateBatchReportAudit": {
"status": "pending",
"expectedDecisionCount": 15,
"matchingDecisionCount": 0,
"missingDecisionCount": 15,
"evidenceGapCount": 0,
"reportReady": false,
"missingCandidateIds": [
"47743b21-5878-499b-ac8a-d473a0513342",
"467d2960-b7c1-46b6-940d-45d81ff4b6e5",
"5144d0d7-b3a0-4600-ba8c-113f289e45c6",
"93f25c25-e779-4481-9d1b-9658bbbcab4a",
"8c516342-7b0c-4d42-ae1e-710f4a32a1b2",
"756da756-1cf4-4382-841d-90367abd6c67",
"c9a5cc4e-6950-477f-9e58-ec69f91e72df",
"02b406f7-e448-4fae-aadd-89cf04dd2e53",
"162e1297-5c68-49e5-8075-33414ea72c77",
"a2624854-e23b-4305-abe0-940b73f76935",
"88970f6e-610e-40bb-b796-dbcc7f7f8cf6",
"33d0b4d1-7e8a-4413-bce1-27abc2500bfc",
"31ca07fd-4885-4d32-96ad-2cc87779b01c",
"a74c7bd4-4d1f-4066-9d39-f277c42b29f9",
"e662e047-7914-4356-8c32-6e54eedebd05"
],
"evidenceGaps": [],
"nextStep": "先完成 candidateExecutionBatch 中的 candidate decisions，再写 loop_report；不要只写泛泛 summary。"
},
"candidateOutcomeSimulator": {
"mode": "read_only_preflight",
"simulatedCount": 15,
"likelyOutcomeMix": [
{
"outcome": "no_action",
"count": 6
},
{
"outcome": "converted_to_source",
"count": 4
},
{
"outcome": "created_new_card",
"count": 5
}
],
"items": [
{
"candidateId": "47743b21-5878-499b-ac8a-d473a0513342",
"batchIndex": 1,
"urlClass": "web_page",
"suggestedOutcome": "no_action",
"confidence": "low",
"requiredConfirmation": [
"page_contains_public_tool_or_capability",
"evidence_is_stronger_than_generic_webpage"
],
"reason": "普通网页需要更强证据；看不到具体可接入能力时应跳过。"
},
{
"candidateId": "467d2960-b7c1-46b6-940d-45d81ff4b6e5",
"batchIndex": 2,
"urlClass": "web_page",
"suggestedOutcome": "no_action",
"confidence": "low",
"requiredConfirmation": [
"page_contains_public_tool_or_capability",
"evidence_is_stronger_than_generic_webpage"
],
"reason": "普通网页需要更强证据；看不到具体可接入能力时应跳过。"
},
{
"candidateId": "5144d0d7-b3a0-4600-ba8c-113f289e45c6",
"batchIndex": 3,
"urlClass": "web_page",
"suggestedOutcome": "no_action",
"confidence": "low",
"requiredConfirmation": [
"page_contains_public_tool_or_capability",
"evidence_is_stronger_than_generic_webpage"
],
"reason": "普通网页需要更强证据；看不到具体可接入能力时应跳过。"
},
{
"candidateId": "93f25c25-e779-4481-9d1b-9658bbbcab4a",
"batchIndex": 4,
"urlClass": "news_article",
"suggestedOutcome": "converted_to_source",
"confidence": "low",
"requiredConfirmation": [
"article_points_to_specific_tool_or_project",
"source_url_is_original_enough"
],
"reason": "新闻页通常更适合沉淀为 raw source；只有明确指向工具、项目或模型能力时才继续补卡。"
},
{
"candidateId": "8c516342-7b0c-4d42-ae1e-710f4a32a1b2",
"batchIndex": 5,
"urlClass": "news_article",
"suggestedOutcome": "converted_to_source",
"confidence": "low",
"requiredConfirmation": [
"article_points_to_specific_tool_or_project",
"source_url_is_original_enough"
],
"reason": "新闻页通常更适合沉淀为 raw source；只有明确指向工具、项目或模型能力时才继续补卡。"
},
{
"candidateId": "756da756-1cf4-4382-841d-90367abd6c67",
"batchIndex": 6,
"urlClass": "code_repository",
"suggestedOutcome": "created_new_card",
"confidence": "medium",
"requiredConfirmation": [
"repo_is_tool_mcp_skill_or_dev_productivity_capability",
"official_source_confirmed"
],
"reason": "代码仓库最有可能对应可接入工具、MCP、Skill 或开发效率能力；仍需打开仓库确认不是纯论文、样例或报告。"
},
{
"candidateId": "c9a5cc4e-6950-477f-9e58-ec69f91e72df",
"batchIndex": 7,
"urlClass": "code_repository",
"suggestedOutcome": "created_new_card",
"confidence": "medium",
"requiredConfirmation": [
"repo_is_tool_mcp_skill_or_dev_productivity_capability",
"official_source_confirmed"
],
"reason": "代码仓库最有可能对应可接入工具、MCP、Skill 或开发效率能力；仍需打开仓库确认不是纯论文、样例或报告。"
},
{
"candidateId": "02b406f7-e448-4fae-aadd-89cf04dd2e53",
"batchIndex": 8,
"urlClass": "code_repository",
"suggestedOutcome": "created_new_card",
"confidence": "medium",
"requiredConfirmation": [
"repo_is_tool_mcp_skill_or_dev_productivity_capability",
"official_source_confirmed"
],
"reason": "代码仓库最有可能对应可接入工具、MCP、Skill 或开发效率能力；仍需打开仓库确认不是纯论文、样例或报告。"
},
{
"candidateId": "162e1297-5c68-49e5-8075-33414ea72c77",
"batchIndex": 9,
"urlClass": "code_repository",
"suggestedOutcome": "created_new_card",
"confidence": "medium",
"requiredConfirmation": [
"repo_is_tool_mcp_skill_or_dev_productivity_capability",
"official_source_confirmed"
],
"reason": "代码仓库最有可能对应可接入工具、MCP、Skill 或开发效率能力；仍需打开仓库确认不是纯论文、样例或报告。"
},
{
"candidateId": "a2624854-e23b-4305-abe0-940b73f76935",
"batchIndex": 10,
"urlClass": "code_repository",
"suggestedOutcome": "created_new_card",
"confidence": "medium",
"requiredConfirmation": [
"repo_is_tool_mcp_skill_or_dev_productivity_capability",
"official_source_confirmed"
],
"reason": "代码仓库最有可能对应可接入工具、MCP、Skill 或开发效率能力；仍需打开仓库确认不是纯论文、样例或报告。"
},
{
"candidateId": "88970f6e-610e-40bb-b796-dbcc7f7f8cf6",
"batchIndex": 11,
"urlClass": "web_page",
"suggestedOutcome": "no_action",
"confidence": "low",
"requiredConfirmation": [
"page_contains_public_tool_or_capability",
"evidence_is_stronger_than_generic_webpage"
],
"reason": "普通网页需要更强证据；看不到具体可接入能力时应跳过。"
},
{
"candidateId": "33d0b4d1-7e8a-4413-bce1-27abc2500bfc",
"batchIndex": 12,
"urlClass": "news_article",
"suggestedOutcome": "converted_to_source",
"confidence": "low",
"requiredConfirmation": [
"article_points_to_specific_tool_or_project",
"source_url_is_original_enough"
],
"reason": "新闻页通常更适合沉淀为 raw source；只有明确指向工具、项目或模型能力时才继续补卡。"
},
{
"candidateId": "31ca07fd-4885-4d32-96ad-2cc87779b01c",
"batchIndex": 13,
"urlClass": "social_discussion",
"suggestedOutcome": "no_action",
"confidence": "low",
"requiredConfirmation": [
"original_project_or_official_source_found",
"social_post_not_standalone_rumor"
],
"reason": "社交帖只能作为线索；必须找到原始项目、论文、官网或官方仓库后再转来源或建卡。"
},
{
"candidateId": "a74c7bd4-4d1f-4066-9d39-f277c42b29f9",
"batchIndex": 14,
"urlClass": "news_article",
"suggestedOutcome": "converted_to_source",
"confidence": "low",
"requiredConfirmation": [
"article_points_to_specific_tool_or_project",
"source_url_is_original_enough"
],
"reason": "新闻页通常更适合沉淀为 raw source；只有明确指向工具、项目或模型能力时才继续补卡。"
},
{
"candidateId": "e662e047-7914-4356-8c32-6e54eedebd05",
"batchIndex": 15,
"urlClass": "web_page",
"suggestedOutcome": "no_action",
"confidence": "low",
"requiredConfirmation": [
"page_contains_public_tool_or_capability",
"evidence_is_stronger_than_generic_webpage"
],
"reason": "普通网页需要更强证据；看不到具体可接入能力时应跳过。"
}
],
"guardrail": "这是只读预估，不能直接写入 candidate decision；Hermes 必须打开原链接并带证据调用 toolkoi_candidate_decision。"
},
"candidateAutomationReadiness": {
"status": "browser_review_required",
"canAutoWrite": false,
"candidateCount": 15,
"browserReviewRequiredCount": 15,
"evidenceRequired": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index"
],
"blockers": [
"simulator_is_not_evidence",
"candidate_decision_requires_original_url_review"
],
"nextStep": "逐条打开 candidateExecutionBatch 原链接，完成 evidence-backed decision 后再刷新 loop_contract。"
},
"candidateBrowserReviewPlan": {
"mode": "original_url_review",
"browserRequired": true,
"reviewCount": 15,
"guardrail": "浏览器核查计划只产生证据，不自动写入 candidate decision。",
"items": [
{
"candidateId": "47743b21-5878-499b-ac8a-d473a0513342",
"batchIndex": 1,
"title": "1. The Trust Fracture: Claude's Dual Identity Crisis | 信任断裂：Claude 的双面危机——从水印风波到阿里全面禁用",
"url": "https://www.reuters.com/world/china/alibaba-ban-claude-code-workplace-over-alleged-backdoor-risks-source-says-2026-07-03/",
"urlHost": "reuters.com",
"urlClass": "web_page",
"checklist": [
"打开网页并确认是否存在明确公开的工具、产品、文档或接入能力。",
"检查页面是否提供官方来源、安装/使用方式或可验证的功能说明。",
"证据不足时不要建卡，优先 skipped/no_action。"
],
"passSignals": [
"页面直接说明可接入工具、Skill、MCP 或开发效率能力。",
"页面提供官方文档、安装链接、仓库或可验证功能。"
],
"skipSignals": [
"页面是泛介绍、目录页、营销页或无法确认实际能力。",
"没有足够公开证据支撑 Toolkoi 卡片或 raw source。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:1"
]
},
{
"candidateId": "467d2960-b7c1-46b6-940d-45d81ff4b6e5",
"batchIndex": 2,
"title": "3. AI Security: From Shield to Sword | 安全悖论：AI 既是盾牌，也是利刃",
"url": "https://the-decoder.com/security-vulnerability-reports-have-exploded-since-ai-models-started-hunting-for-bugs/",
"urlHost": "the-decoder.com",
"urlClass": "web_page",
"checklist": [
"打开网页并确认是否存在明确公开的工具、产品、文档或接入能力。",
"检查页面是否提供官方来源、安装/使用方式或可验证的功能说明。",
"证据不足时不要建卡，优先 skipped/no_action。"
],
"passSignals": [
"页面直接说明可接入工具、Skill、MCP 或开发效率能力。",
"页面提供官方文档、安装链接、仓库或可验证功能。"
],
"skipSignals": [
"页面是泛介绍、目录页、营销页或无法确认实际能力。",
"没有足够公开证据支撑 Toolkoi 卡片或 raw source。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:2"
]
},
{
"candidateId": "5144d0d7-b3a0-4600-ba8c-113f289e45c6",
"batchIndex": 3,
"title": "前沿大模型被曝存在同伴保护行为",
"url": "https://arxiv.org/abs/2604.19784",
"urlHost": "arxiv.org",
"urlClass": "web_page",
"checklist": [
"打开网页并确认是否存在明确公开的工具、产品、文档或接入能力。",
"检查页面是否提供官方来源、安装/使用方式或可验证的功能说明。",
"证据不足时不要建卡，优先 skipped/no_action。"
],
"passSignals": [
"页面直接说明可接入工具、Skill、MCP 或开发效率能力。",
"页面提供官方文档、安装链接、仓库或可验证功能。"
],
"skipSignals": [
"页面是泛介绍、目录页、营销页或无法确认实际能力。",
"没有足够公开证据支撑 Toolkoi 卡片或 raw source。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:3"
]
},
{
"candidateId": "93f25c25-e779-4481-9d1b-9658bbbcab4a",
"batchIndex": 4,
"title": "1. The Security Arms Race | AI安全从附属品升格为战略武器：GPT-5.6受控发布、Daybreak防御平台与政府审批风暴三线并行",
"url": "http://mp.weixin.qq.com/s?__biz=MjM5NjA0NjgyMA==&mid=2651341309&idx=1&sn=9afc5021305207db4b56448f2d298c80&chksm=bc4efd5d2cfde9f4d358f3c696aa65c7c341e85bb2058edb80ca29424675c1e6b473319271de&scene=0&xtrack=1#rd",
"urlHost": "mp.weixin.qq.com",
"urlClass": "news_article",
"checklist": [
"打开原始新闻页并确认具体提到的项目、工具、模型能力或官方链接。",
"寻找文章中的原始项目链接；没有原始来源时只能作为弱 raw source。",
"判断这条新闻是否和 Toolkoi 能力卡有关，而不是泛 AI 资讯。"
],
"passSignals": [
"文章明确指向工具、MCP、Skill、模型能力或官方发布。",
"能提取可回溯的原文 URL 和来源可信度。"
],
"skipSignals": [
"只是宏观观点、融资、榜单或无具体工具入口。",
"需要付费墙或登录且无法确认原文内容。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:4"
]
},
{
"candidateId": "8c516342-7b0c-4d42-ae1e-710f4a32a1b2",
"batchIndex": 5,
"title": "2. The Model Battleground Escalates | 模型军备竞赛白热化：字节豆包2.1、Google Gemini 3.5 Flash、智谱GLM-5.2混战升级",
"url": "https://mp.weixin.qq.com/s?__biz=MzI1MzYzMjE0MQ==&mid=2247520467&idx=1&sn=e6210ac0ffacb37351752b7fcb5259e5",
"urlHost": "mp.weixin.qq.com",
"urlClass": "news_article",
"checklist": [
"打开原始新闻页并确认具体提到的项目、工具、模型能力或官方链接。",
"寻找文章中的原始项目链接；没有原始来源时只能作为弱 raw source。",
"判断这条新闻是否和 Toolkoi 能力卡有关，而不是泛 AI 资讯。"
],
"passSignals": [
"文章明确指向工具、MCP、Skill、模型能力或官方发布。",
"能提取可回溯的原文 URL 和来源可信度。"
],
"skipSignals": [
"只是宏观观点、融资、榜单或无具体工具入口。",
"需要付费墙或登录且无法确认原文内容。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:5"
]
},
{
"candidateId": "756da756-1cf4-4382-841d-90367abd6c67",
"batchIndex": 6,
"title": "SkillOpt: Agent skills as trainable parameters",
"url": "https://github.com/microsoft/SkillOpt",
"urlHost": "github.com",
"urlClass": "code_repository",
"checklist": [
"打开原始 URL 并确认页面可访问。",
"确认仓库 README、描述、releases 或文档是否说明它是工具、MCP、Skill 或开发效率能力。",
"确认项目来源可信度和维护状态，不能只凭标题建卡。"
],
"passSignals": [
"README 或官方文档说明可安装、调用、接入或作为开发工具使用。",
"仓库 owner、stars、recent activity 或 release 能支撑 source_confidence。"
],
"skipSignals": [
"只是论文、新闻、榜单、样例报告或一次性 demo，不构成 Toolkoi 能力卡。",
"页面无法访问或没有足够公开证据。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:6"
]
},
{
"candidateId": "c9a5cc4e-6950-477f-9e58-ec69f91e72df",
"batchIndex": 7,
"title": "Memora: A Harmonic Memory Representation Balancing Abstraction and Specificity",
"url": "https://github.com/microsoft/Memora",
"urlHost": "github.com",
"urlClass": "code_repository",
"checklist": [
"打开原始 URL 并确认页面可访问。",
"确认仓库 README、描述、releases 或文档是否说明它是工具、MCP、Skill 或开发效率能力。",
"确认项目来源可信度和维护状态，不能只凭标题建卡。"
],
"passSignals": [
"README 或官方文档说明可安装、调用、接入或作为开发工具使用。",
"仓库 owner、stars、recent activity 或 release 能支撑 source_confidence。"
],
"skipSignals": [
"只是论文、新闻、榜单、样例报告或一次性 demo，不构成 Toolkoi 能力卡。",
"页面无法访问或没有足够公开证据。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:7"
]
},
{
"candidateId": "02b406f7-e448-4fae-aadd-89cf04dd2e53",
"batchIndex": 8,
"title": "Ire identifies another LOTUSLITE specimen",
"url": "https://github.com/microsoft/project-ire/blob/main/reports/47e51e82229e80a387c3cb100d39d3705e6360bbf9bfa1601dbc484e8d02e653.md",
"urlHost": "github.com",
"urlClass": "code_repository",
"checklist": [
"打开原始 URL 并确认页面可访问。",
"确认仓库 README、描述、releases 或文档是否说明它是工具、MCP、Skill 或开发效率能力。",
"确认项目来源可信度和维护状态，不能只凭标题建卡。"
],
"passSignals": [
"README 或官方文档说明可安装、调用、接入或作为开发工具使用。",
"仓库 owner、stars、recent activity 或 release 能支撑 source_confidence。"
],
"skipSignals": [
"只是论文、新闻、榜单、样例报告或一次性 demo，不构成 Toolkoi 能力卡。",
"页面无法访问或没有足够公开证据。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:8"
]
},
{
"candidateId": "162e1297-5c68-49e5-8075-33414ea72c77",
"batchIndex": 9,
"title": "Data Formulator 0.7: AI-powered data analytics for enterprise data",
"url": "https://github.com/microsoft/data-formulator",
"urlHost": "github.com",
"urlClass": "code_repository",
"checklist": [
"打开原始 URL 并确认页面可访问。",
"确认仓库 README、描述、releases 或文档是否说明它是工具、MCP、Skill 或开发效率能力。",
"确认项目来源可信度和维护状态，不能只凭标题建卡。"
],
"passSignals": [
"README 或官方文档说明可安装、调用、接入或作为开发工具使用。",
"仓库 owner、stars、recent activity 或 release 能支撑 source_confidence。"
],
"skipSignals": [
"只是论文、新闻、榜单、样例报告或一次性 demo，不构成 Toolkoi 能力卡。",
"页面无法访问或没有足够公开证据。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:9"
]
},
{
"candidateId": "a2624854-e23b-4305-abe0-940b73f76935",
"batchIndex": 10,
"title": "Evaluating performance and efficiency of the GitHub Copilot agentic harness across models and tasks",
"url": "https://github.com/github/copilot-sdk",
"urlHost": "github.com",
"urlClass": "code_repository",
"checklist": [
"打开原始 URL 并确认页面可访问。",
"确认仓库 README、描述、releases 或文档是否说明它是工具、MCP、Skill 或开发效率能力。",
"确认项目来源可信度和维护状态，不能只凭标题建卡。"
],
"passSignals": [
"README 或官方文档说明可安装、调用、接入或作为开发工具使用。",
"仓库 owner、stars、recent activity 或 release 能支撑 source_confidence。"
],
"skipSignals": [
"只是论文、新闻、榜单、样例报告或一次性 demo，不构成 Toolkoi 能力卡。",
"页面无法访问或没有足够公开证据。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:10"
]
},
{
"candidateId": "88970f6e-610e-40bb-b796-dbcc7f7f8cf6",
"batchIndex": 11,
"title": "语言模型中的全局工作空间",
"url": "https://www.anthropic.com/research/global-workspace",
"urlHost": "anthropic.com",
"urlClass": "web_page",
"checklist": [
"打开网页并确认是否存在明确公开的工具、产品、文档或接入能力。",
"检查页面是否提供官方来源、安装/使用方式或可验证的功能说明。",
"证据不足时不要建卡，优先 skipped/no_action。"
],
"passSignals": [
"页面直接说明可接入工具、Skill、MCP 或开发效率能力。",
"页面提供官方文档、安装链接、仓库或可验证功能。"
],
"skipSignals": [
"页面是泛介绍、目录页、营销页或无法确认实际能力。",
"没有足够公开证据支撑 Toolkoi 卡片或 raw source。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:11"
]
},
{
"candidateId": "33d0b4d1-7e8a-4413-bce1-27abc2500bfc",
"batchIndex": 12,
"title": "开辟新赛道：小米 YU7 GT 创全球首个纽北自动驾驶圈速纪录，官方圈速榜新增\"自动驾驶\"分类",
"url": "https://www.ithome.com/0/967/234.htm",
"urlHost": "ithome.com",
"urlClass": "news_article",
"checklist": [
"打开原始新闻页并确认具体提到的项目、工具、模型能力或官方链接。",
"寻找文章中的原始项目链接；没有原始来源时只能作为弱 raw source。",
"判断这条新闻是否和 Toolkoi 能力卡有关，而不是泛 AI 资讯。"
],
"passSignals": [
"文章明确指向工具、MCP、Skill、模型能力或官方发布。",
"能提取可回溯的原文 URL 和来源可信度。"
],
"skipSignals": [
"只是宏观观点、融资、榜单或无具体工具入口。",
"需要付费墙或登录且无法确认原文内容。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:12"
]
},
{
"candidateId": "31ca07fd-4885-4d32-96ad-2cc87779b01c",
"batchIndex": 13,
"title": "扎克伯格：建千兆瓦级AI集群，集中精英与资本",
"url": "https://x.com/rohanpaul_ai/status/2073834219659534816",
"urlHost": "x.com",
"urlClass": "social_discussion",
"checklist": [
"打开社交帖并识别它引用的原始项目、论文、官网或仓库。",
"不要把单条社交帖当作官方证据；先找二级原始来源。",
"确认讨论是否包含可验证的接入价值。"
],
"passSignals": [
"社交帖链接到官方项目、仓库、论文或产品页。",
"讨论内容能解释工具价值，并能被原始来源复核。"
],
"skipSignals": [
"只是观点、传闻、吐槽或没有原始链接。",
"无法确认作者、项目或具体工具能力。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:13"
]
},
{
"candidateId": "a74c7bd4-4d1f-4066-9d39-f277c42b29f9",
"batchIndex": 14,
"title": "Anthropic在Claude Code中植入隐写术代码识别中国用户",
"url": "https://mp.weixin.qq.com/s/yLb4T2UC16ebKHApdBbgWw",
"urlHost": "mp.weixin.qq.com",
"urlClass": "news_article",
"checklist": [
"打开原始新闻页并确认具体提到的项目、工具、模型能力或官方链接。",
"寻找文章中的原始项目链接；没有原始来源时只能作为弱 raw source。",
"判断这条新闻是否和 Toolkoi 能力卡有关，而不是泛 AI 资讯。"
],
"passSignals": [
"文章明确指向工具、MCP、Skill、模型能力或官方发布。",
"能提取可回溯的原文 URL 和来源可信度。"
],
"skipSignals": [
"只是宏观观点、融资、榜单或无具体工具入口。",
"需要付费墙或登录且无法确认原文内容。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:14"
]
},
{
"candidateId": "e662e047-7914-4356-8c32-6e54eedebd05",
"batchIndex": 15,
"title": "纽约时报修订诉讼，指控微软为OpenAI建造版权侵权超级计算机",
"url": "https://arstechnica.com/tech-policy/2026/06/microsoft-built-supercomputer-to-help-openai-infringe-copyrights-nyt-alleged",
"urlHost": "arstechnica.com",
"urlClass": "web_page",
"checklist": [
"打开网页并确认是否存在明确公开的工具、产品、文档或接入能力。",
"检查页面是否提供官方来源、安装/使用方式或可验证的功能说明。",
"证据不足时不要建卡，优先 skipped/no_action。"
],
"passSignals": [
"页面直接说明可接入工具、Skill、MCP 或开发效率能力。",
"页面提供官方文档、安装链接、仓库或可验证功能。"
],
"skipSignals": [
"页面是泛介绍、目录页、营销页或无法确认实际能力。",
"没有足够公开证据支撑 Toolkoi 卡片或 raw source。"
],
"evidenceToCapture": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:15"
]
}
]
},
"candidateBrowserReviewReceiptSpec": {
"mode": "browser_evidence_receipt",
"version": "v1",
"expectedReceiptCount": 15,
"writePolicy": "read_only_before_candidate_decision",
"requiredFields": [
"candidateId",
"batchIndex",
"openedOriginalUrl",
"finalUrl",
"sourceConfidence",
"decisionReason",
"suggestedOutcome",
"evidence",
"reviewedAt"
],
"acceptedOutcomes": [
"converted_to_source",
"updated_existing_card",
"created_new_card",
"skipped",
"blocked",
"no_action"
],
"confidenceScale": [
"high",
"medium",
"low"
],
"guardrail": "浏览器核查回执只记录证据，不能替代 toolkoi_candidate_decision 写入。",
"submission": {
"method": "POST",
"endpoint": "/api/admin/toolkoi/browser-review-receipt",
"contentType": "application/json",
"auth": "TOOLKOI_ADMIN_TOKEN",
"writePolicy": "read_only_before_candidate_decision",
"expectedResponseFields": [
"accepted",
"missingEvidence",
"decisionDraft",
"nextTool",
"writePolicy"
]
},
"batchSubmission": {
"method": "POST",
"endpoint": "/api/admin/toolkoi/browser-review-receipts",
"contentType": "application/json",
"auth": "TOOLKOI_ADMIN_TOKEN",
"writePolicy": "read_only_before_candidate_decision",
"expectedResponseFields": [
"acceptedCount",
"rejectedCount",
"coveragePercent",
"readyForCandidateDecision",
"decisionDrafts",
"promotionPlan"
]
},
"promotionSubmission": {
"method": "POST",
"endpoint": "/api/admin/toolkoi/decision-promotion",
"contentType": "application/json",
"auth": "TOOLKOI_ADMIN_TOKEN",
"writePolicy": "read_only_submit_plan",
"expectedResponseFields": [
"canPromote",
"submitQueue",
"stopReason",
"guardrail"
]
},
"promotionAudit": {
"method": "POST",
"endpoint": "/api/admin/toolkoi/decision-promotion-audit",
"contentType": "application/json",
"auth": "TOOLKOI_ADMIN_TOKEN",
"writePolicy": "read_only_decision_audit",
"expectedResponseFields": [
"status",
"matchingDecisionCount",
"missingDecisionCount",
"mismatchCount",
"readyForLoopReport"
]
},
"items": [
{
"candidateId": "47743b21-5878-499b-ac8a-d473a0513342",
"batchIndex": 1,
"title": "1. The Trust Fracture: Claude's Dual Identity Crisis | 信任断裂：Claude 的双面危机——从水印风波到阿里全面禁用",
"url": "https://www.reuters.com/world/china/alibaba-ban-claude-code-workplace-over-alleged-backdoor-risks-source-says-2026-07-03/",
"urlClass": "web_page",
"receiptTemplate": {
"candidateId": "47743b21-5878-499b-ac8a-d473a0513342",
"batchIndex": 1,
"openedOriginalUrl": true,
"finalUrl": "https://www.reuters.com/world/china/alibaba-ban-claude-code-workplace-over-alleged-backdoor-risks-source-says-2026-07-03/",
"sourceConfidence": "<high|medium|low>",
"decisionReason": "<browser_evidence_summary>",
"suggestedOutcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"evidence": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:1"
],
"reviewedAt": "2026-09-19T07:17:37.700Z"
}
},
{
"candidateId": "467d2960-b7c1-46b6-940d-45d81ff4b6e5",
"batchIndex": 2,
"title": "3. AI Security: From Shield to Sword | 安全悖论：AI 既是盾牌，也是利刃",
"url": "https://the-decoder.com/security-vulnerability-reports-have-exploded-since-ai-models-started-hunting-for-bugs/",
"urlClass": "web_page",
"receiptTemplate": {
"candidateId": "467d2960-b7c1-46b6-940d-45d81ff4b6e5",
"batchIndex": 2,
"openedOriginalUrl": true,
"finalUrl": "https://the-decoder.com/security-vulnerability-reports-have-exploded-since-ai-models-started-hunting-for-bugs/",
"sourceConfidence": "<high|medium|low>",
"decisionReason": "<browser_evidence_summary>",
"suggestedOutcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"evidence": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:2"
],
"reviewedAt": "2026-09-19T07:17:37.700Z"
}
},
{
"candidateId": "5144d0d7-b3a0-4600-ba8c-113f289e45c6",
"batchIndex": 3,
"title": "前沿大模型被曝存在同伴保护行为",
"url": "https://arxiv.org/abs/2604.19784",
"urlClass": "web_page",
"receiptTemplate": {
"candidateId": "5144d0d7-b3a0-4600-ba8c-113f289e45c6",
"batchIndex": 3,
"openedOriginalUrl": true,
"finalUrl": "https://arxiv.org/abs/2604.19784",
"sourceConfidence": "<high|medium|low>",
"decisionReason": "<browser_evidence_summary>",
"suggestedOutcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"evidence": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:3"
],
"reviewedAt": "2026-09-19T07:17:37.700Z"
}
},
{
"candidateId": "93f25c25-e779-4481-9d1b-9658bbbcab4a",
"batchIndex": 4,
"title": "1. The Security Arms Race | AI安全从附属品升格为战略武器：GPT-5.6受控发布、Daybreak防御平台与政府审批风暴三线并行",
"url": "http://mp.weixin.qq.com/s?__biz=MjM5NjA0NjgyMA==&mid=2651341309&idx=1&sn=9afc5021305207db4b56448f2d298c80&chksm=bc4efd5d2cfde9f4d358f3c696aa65c7c341e85bb2058edb80ca29424675c1e6b473319271de&scene=0&xtrack=1#rd",
"urlClass": "news_article",
"receiptTemplate": {
"candidateId": "93f25c25-e779-4481-9d1b-9658bbbcab4a",
"batchIndex": 4,
"openedOriginalUrl": true,
"finalUrl": "http://mp.weixin.qq.com/s?__biz=MjM5NjA0NjgyMA==&mid=2651341309&idx=1&sn=9afc5021305207db4b56448f2d298c80&chksm=bc4efd5d2cfde9f4d358f3c696aa65c7c341e85bb2058edb80ca29424675c1e6b473319271de&scene=0&xtrack=1#rd",
"sourceConfidence": "<high|medium|low>",
"decisionReason": "<browser_evidence_summary>",
"suggestedOutcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"evidence": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:4"
],
"reviewedAt": "2026-09-19T07:17:37.700Z"
}
},
{
"candidateId": "8c516342-7b0c-4d42-ae1e-710f4a32a1b2",
"batchIndex": 5,
"title": "2. The Model Battleground Escalates | 模型军备竞赛白热化：字节豆包2.1、Google Gemini 3.5 Flash、智谱GLM-5.2混战升级",
"url": "https://mp.weixin.qq.com/s?__biz=MzI1MzYzMjE0MQ==&mid=2247520467&idx=1&sn=e6210ac0ffacb37351752b7fcb5259e5",
"urlClass": "news_article",
"receiptTemplate": {
"candidateId": "8c516342-7b0c-4d42-ae1e-710f4a32a1b2",
"batchIndex": 5,
"openedOriginalUrl": true,
"finalUrl": "https://mp.weixin.qq.com/s?__biz=MzI1MzYzMjE0MQ==&mid=2247520467&idx=1&sn=e6210ac0ffacb37351752b7fcb5259e5",
"sourceConfidence": "<high|medium|low>",
"decisionReason": "<browser_evidence_summary>",
"suggestedOutcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"evidence": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:5"
],
"reviewedAt": "2026-09-19T07:17:37.700Z"
}
},
{
"candidateId": "756da756-1cf4-4382-841d-90367abd6c67",
"batchIndex": 6,
"title": "SkillOpt: Agent skills as trainable parameters",
"url": "https://github.com/microsoft/SkillOpt",
"urlClass": "code_repository",
"receiptTemplate": {
"candidateId": "756da756-1cf4-4382-841d-90367abd6c67",
"batchIndex": 6,
"openedOriginalUrl": true,
"finalUrl": "https://github.com/microsoft/SkillOpt",
"sourceConfidence": "<high|medium|low>",
"decisionReason": "<browser_evidence_summary>",
"suggestedOutcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"evidence": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:6"
],
"reviewedAt": "2026-09-19T07:17:37.700Z"
}
},
{
"candidateId": "c9a5cc4e-6950-477f-9e58-ec69f91e72df",
"batchIndex": 7,
"title": "Memora: A Harmonic Memory Representation Balancing Abstraction and Specificity",
"url": "https://github.com/microsoft/Memora",
"urlClass": "code_repository",
"receiptTemplate": {
"candidateId": "c9a5cc4e-6950-477f-9e58-ec69f91e72df",
"batchIndex": 7,
"openedOriginalUrl": true,
"finalUrl": "https://github.com/microsoft/Memora",
"sourceConfidence": "<high|medium|low>",
"decisionReason": "<browser_evidence_summary>",
"suggestedOutcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"evidence": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:7"
],
"reviewedAt": "2026-09-19T07:17:37.700Z"
}
},
{
"candidateId": "02b406f7-e448-4fae-aadd-89cf04dd2e53",
"batchIndex": 8,
"title": "Ire identifies another LOTUSLITE specimen",
"url": "https://github.com/microsoft/project-ire/blob/main/reports/47e51e82229e80a387c3cb100d39d3705e6360bbf9bfa1601dbc484e8d02e653.md",
"urlClass": "code_repository",
"receiptTemplate": {
"candidateId": "02b406f7-e448-4fae-aadd-89cf04dd2e53",
"batchIndex": 8,
"openedOriginalUrl": true,
"finalUrl": "https://github.com/microsoft/project-ire/blob/main/reports/47e51e82229e80a387c3cb100d39d3705e6360bbf9bfa1601dbc484e8d02e653.md",
"sourceConfidence": "<high|medium|low>",
"decisionReason": "<browser_evidence_summary>",
"suggestedOutcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"evidence": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:8"
],
"reviewedAt": "2026-09-19T07:17:37.700Z"
}
},
{
"candidateId": "162e1297-5c68-49e5-8075-33414ea72c77",
"batchIndex": 9,
"title": "Data Formulator 0.7: AI-powered data analytics for enterprise data",
"url": "https://github.com/microsoft/data-formulator",
"urlClass": "code_repository",
"receiptTemplate": {
"candidateId": "162e1297-5c68-49e5-8075-33414ea72c77",
"batchIndex": 9,
"openedOriginalUrl": true,
"finalUrl": "https://github.com/microsoft/data-formulator",
"sourceConfidence": "<high|medium|low>",
"decisionReason": "<browser_evidence_summary>",
"suggestedOutcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"evidence": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:9"
],
"reviewedAt": "2026-09-19T07:17:37.700Z"
}
},
{
"candidateId": "a2624854-e23b-4305-abe0-940b73f76935",
"batchIndex": 10,
"title": "Evaluating performance and efficiency of the GitHub Copilot agentic harness across models and tasks",
"url": "https://github.com/github/copilot-sdk",
"urlClass": "code_repository",
"receiptTemplate": {
"candidateId": "a2624854-e23b-4305-abe0-940b73f76935",
"batchIndex": 10,
"openedOriginalUrl": true,
"finalUrl": "https://github.com/github/copilot-sdk",
"sourceConfidence": "<high|medium|low>",
"decisionReason": "<browser_evidence_summary>",
"suggestedOutcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"evidence": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:10"
],
"reviewedAt": "2026-09-19T07:17:37.700Z"
}
},
{
"candidateId": "88970f6e-610e-40bb-b796-dbcc7f7f8cf6",
"batchIndex": 11,
"title": "语言模型中的全局工作空间",
"url": "https://www.anthropic.com/research/global-workspace",
"urlClass": "web_page",
"receiptTemplate": {
"candidateId": "88970f6e-610e-40bb-b796-dbcc7f7f8cf6",
"batchIndex": 11,
"openedOriginalUrl": true,
"finalUrl": "https://www.anthropic.com/research/global-workspace",
"sourceConfidence": "<high|medium|low>",
"decisionReason": "<browser_evidence_summary>",
"suggestedOutcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"evidence": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:11"
],
"reviewedAt": "2026-09-19T07:17:37.700Z"
}
},
{
"candidateId": "33d0b4d1-7e8a-4413-bce1-27abc2500bfc",
"batchIndex": 12,
"title": "开辟新赛道：小米 YU7 GT 创全球首个纽北自动驾驶圈速纪录，官方圈速榜新增\"自动驾驶\"分类",
"url": "https://www.ithome.com/0/967/234.htm",
"urlClass": "news_article",
"receiptTemplate": {
"candidateId": "33d0b4d1-7e8a-4413-bce1-27abc2500bfc",
"batchIndex": 12,
"openedOriginalUrl": true,
"finalUrl": "https://www.ithome.com/0/967/234.htm",
"sourceConfidence": "<high|medium|low>",
"decisionReason": "<browser_evidence_summary>",
"suggestedOutcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"evidence": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:12"
],
"reviewedAt": "2026-09-19T07:17:37.700Z"
}
},
{
"candidateId": "31ca07fd-4885-4d32-96ad-2cc87779b01c",
"batchIndex": 13,
"title": "扎克伯格：建千兆瓦级AI集群，集中精英与资本",
"url": "https://x.com/rohanpaul_ai/status/2073834219659534816",
"urlClass": "social_discussion",
"receiptTemplate": {
"candidateId": "31ca07fd-4885-4d32-96ad-2cc87779b01c",
"batchIndex": 13,
"openedOriginalUrl": true,
"finalUrl": "https://x.com/rohanpaul_ai/status/2073834219659534816",
"sourceConfidence": "<high|medium|low>",
"decisionReason": "<browser_evidence_summary>",
"suggestedOutcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"evidence": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:13"
],
"reviewedAt": "2026-09-19T07:17:37.700Z"
}
},
{
"candidateId": "a74c7bd4-4d1f-4066-9d39-f277c42b29f9",
"batchIndex": 14,
"title": "Anthropic在Claude Code中植入隐写术代码识别中国用户",
"url": "https://mp.weixin.qq.com/s/yLb4T2UC16ebKHApdBbgWw",
"urlClass": "news_article",
"receiptTemplate": {
"candidateId": "a74c7bd4-4d1f-4066-9d39-f277c42b29f9",
"batchIndex": 14,
"openedOriginalUrl": true,
"finalUrl": "https://mp.weixin.qq.com/s/yLb4T2UC16ebKHApdBbgWw",
"sourceConfidence": "<high|medium|low>",
"decisionReason": "<browser_evidence_summary>",
"suggestedOutcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"evidence": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:14"
],
"reviewedAt": "2026-09-19T07:17:37.700Z"
}
},
{
"candidateId": "e662e047-7914-4356-8c32-6e54eedebd05",
"batchIndex": 15,
"title": "纽约时报修订诉讼，指控微软为OpenAI建造版权侵权超级计算机",
"url": "https://arstechnica.com/tech-policy/2026/06/microsoft-built-supercomputer-to-help-openai-infringe-copyrights-nyt-alleged",
"urlClass": "web_page",
"receiptTemplate": {
"candidateId": "e662e047-7914-4356-8c32-6e54eedebd05",
"batchIndex": 15,
"openedOriginalUrl": true,
"finalUrl": "https://arstechnica.com/tech-policy/2026/06/microsoft-built-supercomputer-to-help-openai-infringe-copyrights-nyt-alleged",
"sourceConfidence": "<high|medium|low>",
"decisionReason": "<browser_evidence_summary>",
"suggestedOutcome": "<converted_to_source|updated_existing_card|created_new_card|skipped|blocked|no_action>",
"evidence": [
"opened_original_url",
"source_confidence",
"decision_reason",
"candidate_batch_index:15"
],
"reviewedAt": "2026-09-19T07:17:37.700Z"
}
}
]
},
"prioritySlugs": [
"mcp-hermes-agent",
"sakana-fugu",
"zambo-mcp",
"cloudflare-security-audit-skill",
"github-mcp-server"
],
"priorityWorkItems": [
{
"slug": "mcp-hermes-agent",
"name": "Hermes Agent MCP Server",
"type": "mcp",
"priority": 0,
"score": 95,
"missing": [
"sandbox_failed"
],
"recommendedActions": [
"sandbox_verify"
],
"evidence": {
"sourceCount": 3,
"officialSourceCount": 1,
"hasExecutableInstallCommand": true,
"hasInstallUrl": true,
"installCandidateUrls": [
"https://hermes-agent.nousresearch.com/docs",
"https://github.com/nousresearch/hermes-agent",
"https://hermes-agent.nousresearch.com"
],
"sandboxStatus": "failed",
"wikiPageCount": 1,
"wikiLinkCount": 1,
"relationCount": 1,
"revisionCount": 3,
"hasChineseCardCopy": true,
"hasChineseWikiCopy": true
},
"sourcePolicy": {
"kind": "external_official",
"label": "外部官方源",
"reason": "卡片已有官网、官方仓库或安装链接，必须补齐官方 source、安装证据和真实 sandbox。",
"requiresInstallEvidence": true,
"requiresOfficialSource": true,
"requiresSandbox": true,
"allowsEditorialMaintenance": false
},
"previousOutcome": null,
"nextActionHint": "最近 sandbox 未跑通",
"actionCriteria": [
{
"action": "sandbox_verify",
"successCondition": "真实 sandbox 结果必须为 passed，并满足 sandboxPlan.successSignals；失败只能作为 raw source，并在 loop_report 标记 blocked/skipped。",
"requiredEvidence": [
"sandbox_result=passed",
"sandboxPlan.successSignals",
"input_notes",
"process_notes",
"output_notes"
],
"currentState": "sandbox failed",
"satisfied": false
}
],
"actionPolicy": [
{
"action": "sandbox_verify",
"status": "cooldown",
"reason": "最近 sandbox 已失败；先补新安装来源、改变 sandbox 策略，或由用户明确要求后再重跑。",
"canRunNow": false
}
],
"readyActions": [],
"cooldownActions": [
"sandbox_verify"
],
"blockedActions": [],
"sandboxPlan": {
"required": true,
"readiness": "ready",
"reason": "已有 installCommand；可以用 bounded sandbox 验证最小安装或启动路径。",
"mcpTool": "toolkoi_sandbox_verify",
"image": "python:3.12-alpine",
"sandboxCommand": "set -eux\npip install hermes-agent-mcp\nprintf '\\nTOOLKOI_SANDBOX_INSTALL_COMMAND_COMPLETED\\n'\nprintf 'Hermes must inspect command output and only mark passed when the tool starts or answers a minimal help/version/smoke probe.\\n'",
"inputNotes": [
"工具：mcp-hermes-agent",
"只使用公开包、官方仓库、demo 数据或本地临时文件。",
"不得注入生产凭据。"
],
"processNotes": [
"运行 installCommand 后，补充 --help、--version、stdio 初始化、HTTP health check 或最小查询中的一种真实 probe。",
"如果只完成安装但没有启动/查询证据，loop_report 只能写 blocked/skipped，不能写 completed。"
],
"outputNotes": [
"记录 exit code、关键 stdout/stderr、probe 结果和是否需要凭据。",
"passed 只能表示最小 probe 真实成功。"
],
"successSignals": [
"exit_code=0",
"输出包含 help/version/health/query 之一",
"raw source 中有 输入/处理/输出/验证结果"
],
"blockers": []
}
},
{
"slug": "sakana-fugu",
"name": "Sakana Fugu",
"type": "mcp",
"priority": 0,
"score": 95,
"missing": [
"sandbox_failed"
],
"recommendedActions": [
"sandbox_verify"
],
"evidence": {
"sourceCount": 2,
"officialSourceCount": 1,
"hasExecutableInstallCommand": true,
"hasInstallUrl": true,
"installCandidateUrls": [
"https://sakana.ai/fugu",
"https://github.com/sakanaai/fugu",
"https://sakana.ai"
],
"sandboxStatus": "failed",
"wikiPageCount": 1,
"wikiLinkCount": 1,
"relationCount": 1,
"revisionCount": 3,
"hasChineseCardCopy": true,
"hasChineseWikiCopy": true
},
"sourcePolicy": {
"kind": "external_official",
"label": "外部官方源",
"reason": "卡片已有官网、官方仓库或安装链接，必须补齐官方 source、安装证据和真实 sandbox。",
"requiresInstallEvidence": true,
"requiresOfficialSource": true,
"requiresSandbox": true,
"allowsEditorialMaintenance": false
},
"previousOutcome": null,
"nextActionHint": "最近 sandbox 未跑通",
"actionCriteria": [
{
"action": "sandbox_verify",
"successCondition": "真实 sandbox 结果必须为 passed，并满足 sandboxPlan.successSignals；失败只能作为 raw source，并在 loop_report 标记 blocked/skipped。",
"requiredEvidence": [
"sandbox_result=passed",
"sandboxPlan.successSignals",
"input_notes",
"process_notes",
"output_notes"
],
"currentState": "sandbox failed",
"satisfied": false
}
],
"actionPolicy": [
{
"action": "sandbox_verify",
"status": "cooldown",
"reason": "最近 sandbox 已失败；先补新安装来源、改变 sandbox 策略，或由用户明确要求后再重跑。",
"canRunNow": false
}
],
"readyActions": [],
"cooldownActions": [
"sandbox_verify"
],
"blockedActions": [],
"sandboxPlan": {
"required": true,
"readiness": "ready",
"reason": "已有 installCommand；可以用 bounded sandbox 验证最小安装或启动路径。",
"mcpTool": "toolkoi_sandbox_verify",
"image": "python:3.12-alpine",
"sandboxCommand": "set -eux\npip install sakana-fugu\nprintf '\\nTOOLKOI_SANDBOX_INSTALL_COMMAND_COMPLETED\\n'\nprintf 'Hermes must inspect command output and only mark passed when the tool starts or answers a minimal help/version/smoke probe.\\n'",
"inputNotes": [
"工具：sakana-fugu",
"只使用公开包、官方仓库、demo 数据或本地临时文件。",
"不得注入生产凭据。"
],
"processNotes": [
"运行 installCommand 后，补充 --help、--version、stdio 初始化、HTTP health check 或最小查询中的一种真实 probe。",
"如果只完成安装但没有启动/查询证据，loop_report 只能写 blocked/skipped，不能写 completed。"
],
"outputNotes": [
"记录 exit code、关键 stdout/stderr、probe 结果和是否需要凭据。",
"passed 只能表示最小 probe 真实成功。"
],
"successSignals": [
"exit_code=0",
"输出包含 help/version/health/query 之一",
"raw source 中有 输入/处理/输出/验证结果"
],
"blockers": []
}
},
{
"slug": "zambo-mcp",
"name": "Zambo MCP",
"type": "mcp",
"priority": 0,
"score": 95,
"missing": [
"sandbox_failed"
],
"recommendedActions": [
"sandbox_verify"
],
"evidence": {
"sourceCount": 3,
"officialSourceCount": 1,
"hasExecutableInstallCommand": true,
"hasInstallUrl": true,
"installCandidateUrls": [
"https://zambo.dev",
"https://github.com/zambo-dev/zambo-mcp"
],
"sandboxStatus": "failed",
"wikiPageCount": 1,
"wikiLinkCount": 2,
"relationCount": 2,
"revisionCount": 3,
"hasChineseCardCopy": true,
"hasChineseWikiCopy": true
},
"sourcePolicy": {
"kind": "external_official",
"label": "外部官方源",
"reason": "卡片已有官网、官方仓库或安装链接，必须补齐官方 source、安装证据和真实 sandbox。",
"requiresInstallEvidence": true,
"requiresOfficialSource": true,
"requiresSandbox": true,
"allowsEditorialMaintenance": false
},
"previousOutcome": null,
"nextActionHint": "最近 sandbox 未跑通",
"actionCriteria": [
{
"action": "sandbox_verify",
"successCondition": "真实 sandbox 结果必须为 passed，并满足 sandboxPlan.successSignals；失败只能作为 raw source，并在 loop_report 标记 blocked/skipped。",
"requiredEvidence": [
"sandbox_result=passed",
"sandboxPlan.successSignals",
"input_notes",
"process_notes",
"output_notes"
],
"currentState": "sandbox failed",
"satisfied": false
}
],
"actionPolicy": [
{
"action": "sandbox_verify",
"status": "cooldown",
"reason": "最近 sandbox 已失败；先补新安装来源、改变 sandbox 策略，或由用户明确要求后再重跑。",
"canRunNow": false
}
],
"readyActions": [],
"cooldownActions": [
"sandbox_verify"
],
"blockedActions": [],
"sandboxPlan": {
"required": true,
"readiness": "ready",
"reason": "已有 installCommand；可以用 bounded sandbox 验证最小安装或启动路径。",
"mcpTool": "toolkoi_sandbox_verify",
"image": "python:3.12-alpine",
"sandboxCommand": "set -eux\npip install zambo-mcp\nprintf '\\nTOOLKOI_SANDBOX_INSTALL_COMMAND_COMPLETED\\n'\nprintf 'Hermes must inspect command output and only mark passed when the tool starts or answers a minimal help/version/smoke probe.\\n'",
"inputNotes": [
"工具：zambo-mcp",
"只使用公开包、官方仓库、demo 数据或本地临时文件。",
"不得注入生产凭据。"
],
"processNotes": [
"运行 installCommand 后，补充 --help、--version、stdio 初始化、HTTP health check 或最小查询中的一种真实 probe。",
"如果只完成安装但没有启动/查询证据，loop_report 只能写 blocked/skipped，不能写 completed。"
],
"outputNotes": [
"记录 exit code、关键 stdout/stderr、probe 结果和是否需要凭据。",
"passed 只能表示最小 probe 真实成功。"
],
"successSignals": [
"exit_code=0",
"输出包含 help/version/health/query 之一",
"raw source 中有 输入/处理/输出/验证结果"
],
"blockers": []
}
},
{
"slug": "cloudflare-security-audit-skill",
"name": "Cloudflare Security Audit Skill",
"type": "skill",
"priority": 1,
"score": 70,
"missing": [
"sandbox_verification"
],
"recommendedActions": [
"sandbox_verify"
],
"evidence": {
"sourceCount": 2,
"officialSourceCount": 2,
"hasExecutableInstallCommand": false,
"hasInstallUrl": true,
"installCandidateUrls": [
"https://github.com/cloudflare/security-audit-skill"
],
"sandboxStatus": "pending",
"wikiPageCount": 1,
"wikiLinkCount": 2,
"relationCount": 3,
"revisionCount": 1,
"hasChineseCardCopy": true,
"hasChineseWikiCopy": true
},
"sourcePolicy": {
"kind": "external_official",
"label": "外部官方源",
"reason": "卡片已有官网、官方仓库或安装链接，必须补齐官方 source、安装证据和真实 sandbox。",
"requiresInstallEvidence": true,
"requiresOfficialSource": true,
"requiresSandbox": true,
"allowsEditorialMaintenance": false
},
"previousOutcome": null,
"nextActionHint": "缺少真实 sandbox 验证",
"actionCriteria": [
{
"action": "sandbox_verify",
"successCondition": "真实 sandbox 结果必须为 passed，并满足 sandboxPlan.successSignals；失败只能作为 raw source，并在 loop_report 标记 blocked/skipped。",
"requiredEvidence": [
"sandbox_result=passed",
"sandboxPlan.successSignals",
"input_notes",
"process_notes",
"output_notes"
],
"currentState": "sandbox pending",
"satisfied": false
}
],
"actionPolicy": [
{
"action": "sandbox_verify",
"status": "blocked",
"reason": "现有 installCommand 不是可执行 shell 命令；Hermes 需要先从官方来源补安装路径，再运行 sandbox。",
"canRunNow": false
}
],
"readyActions": [],
"cooldownActions": [],
"blockedActions": [
"sandbox_verify"
],
"sandboxPlan": {
"required": true,
"readiness": "needs_install_source",
"reason": "现有 installCommand 不是可执行 shell 命令；Hermes 需要先从官方来源补安装路径，再运行 sandbox。",
"mcpTool": "toolkoi_sandbox_verify",
"image": null,
"sandboxCommand": null,
"inputNotes": [
"先调用 source_upsert 写入官方安装文档或 GitHub README。"
],
"processNotes": [
"从官方来源提取可在允许镜像中运行的最小 install/start/help 命令。"
],
"outputNotes": [
"如果只能找到产品介绍而没有安装路径，本轮应标记 skipped 或 blocked。"
],
"successSignals": [
"官方安装命令",
"无凭据最小 smoke test",
"sandbox raw source 写入 passed"
],
"blockers": [
"non_executable_install_command"
]
}
},
{
"slug": "github-mcp-server",
"name": "GitHub MCP Server",
"type": "mcp",
"priority": 1,
"score": 70,
"missing": [
"sandbox_verification"
],
"recommendedActions": [
"sandbox_verify"
],
"evidence": {
"sourceCount": 3,
"officialSourceCount": 2,
"hasExecutableInstallCommand": false,
"hasInstallUrl": true,
"installCandidateUrls": [
"https://github.com/github/github-mcp-server"
],
"sandboxStatus": "pending",
"wikiPageCount": 1,
"wikiLinkCount": 2,
"relationCount": 6,
"revisionCount": 0,
"hasChineseCardCopy": true,
"hasChineseWikiCopy": true
},
"sourcePolicy": {
"kind": "external_official",
"label": "外部官方源",
"reason": "卡片已有官网、官方仓库或安装链接，必须补齐官方 source、安装证据和真实 sandbox。",
"requiresInstallEvidence": true,
"requiresOfficialSource": true,
"requiresSandbox": true,
"allowsEditorialMaintenance": false
},
"previousOutcome": null,
"nextActionHint": "缺少真实 sandbox 验证",
"actionCriteria": [
{
"action": "sandbox_verify",
"successCondition": "真实 sandbox 结果必须为 passed，并满足 sandboxPlan.successSignals；失败只能作为 raw source，并在 loop_report 标记 blocked/skipped。",
"requiredEvidence": [
"sandbox_result=passed",
"sandboxPlan.successSignals",
"input_notes",
"process_notes",
"output_notes"
],
"currentState": "sandbox pending",
"satisfied": false
}
],
"actionPolicy": [
{
"action": "sandbox_verify",
"status": "blocked",
"reason": "现有 installCommand 不是可执行 shell 命令；Hermes 需要先从官方来源补安装路径，再运行 sandbox。",
"canRunNow": false
}
],
"readyActions": [],
"cooldownActions": [],
"blockedActions": [
"sandbox_verify"
],
"sandboxPlan": {
"required": true,
"readiness": "needs_install_source",
"reason": "现有 installCommand 不是可执行 shell 命令；Hermes 需要先从官方来源补安装路径，再运行 sandbox。",
"mcpTool": "toolkoi_sandbox_verify",
"image": null,
"sandboxCommand": null,
"inputNotes": [
"先调用 source_upsert 写入官方安装文档或 GitHub README。"
],
"processNotes": [
"从官方来源提取可在允许镜像中运行的最小 install/start/help 命令。"
],
"outputNotes": [
"如果只能找到产品介绍而没有安装路径，本轮应标记 skipped 或 blocked。"
],
"successSignals": [
"官方安装命令",
"无凭据最小 smoke test",
"sandbox raw source 写入 passed"
],
"blockers": [
"non_executable_install_command"
]
}
}
],
"reportRepairActions": [],
"priorityOutcomes": [],
"requiredActions": [
"sandbox_verify",
"candidate_research"
],
"evidenceRequired": [
"sandbox_result",
"candidate_decision"
],
"publishPolicy": "publish_when_sources_are_strong",
"rollbackRequired": false,
"previousReport": null
}