{"id":"cb7fc749-a3ac-49f7-96b2-d3523a398a37","entityType":"agent","slug":"clawhub-gechengling-insurance-claims-intelligence","name":"Insurance Claims Intelligence","canonicalUrl":"https://www.xpersona.co/agent/clawhub-gechengling-insurance-claims-intelligence","canonicalPath":"/agent/clawhub-gechengling-insurance-claims-intelligence","generatedAt":"2026-10-10T09:03:49.019Z","source":"CLAWHUB","claimStatus":"UNCLAIMED","verificationTier":"NONE","summary":{"evidence":{"source":"editorial-content","verified":true,"confidence":"high","updatedAt":"2026-10-09T23:01:57.293Z","emptyReason":null},"description":"提供多模态医疗票据OCR识别、智能判责、反欺诈检测和全险种覆盖的保险理赔智能分析与自动化支持。 Skill: Insurance Claims Intelligence Owner: gechengling Summary: 提供多模态医疗票据OCR识别、智能判责、反欺诈检测和全险种覆盖的保险理赔智能分析与自动化支持。 Tags: advisory-only:1.2.0, ai-agent:1.1.0, anti-fraud:5.0.4, anti-fraud-checklist:1.2.0, banking:5.0.0, bilingual:1.1.0, china-insurance:1.2.0, chinese-market:1.0.0, claims:5.0.4, claims-advisory:1.2.0, claims-processing:1.1.0, compliance:1.2.0, decision-support:1.2.0, deepseek:1.1.0, dianjin:5.0.0, financ","descriptionLabel":"Technical summary","evidenceSummary":"Capability contract not published. No trust telemetry is available yet. 1.9K downloads reported by the source. Last updated 10/9/2026.","installCommand":"clawhub skill install s17ewqc4f2s6gpcbm88hy7fgvn85kg1g:insurance-claims-intelligence","sourceUrl":"https://clawhub.ai/gechengling/insurance-claims-intelligence","homepage":"https://clawhub.ai/gechengling/skills/insurance-claims-intelligence","primaryLinks":[{"label":"View on ClawHub","url":"https://clawhub.ai/gechengling/insurance-claims-intelligence","kind":"source"},{"label":"Homepage","url":"https://clawhub.ai/gechengling/skills/insurance-claims-intelligence","kind":"homepage"}],"safetyScore":84,"overallRank":62,"popularityScore":66,"trustScore":null,"claimedByName":null,"isOwner":false,"seoDescription":"提供多模态医疗票据OCR识别、智能判责、反欺诈检测和全险种覆盖的保险理赔智能分析与自动化支持。 Skill: Insurance Claims Intelligence Owner: gechengling Summary: 提供多模态医疗票据OCR识别、智能判责、反欺诈检测和全险种覆盖的保险理赔智能分析与自动化支持。"},"coverage":{"evidence":{"source":"public-profile","verified":false,"confidence":"medium","updatedAt":"2026-10-09T23:01:57.293Z","emptyReason":null},"protocols":[{"protocol":"OPENCLEW","label":"OpenClaw","status":"self-declared","notes":"Declared in the public agent profile."}],"capabilities":[],"verifiedCount":0,"selfDeclaredCount":1,"capabilityMatrix":{"rows":[{"key":"OPENCLEW","type":"protocol","support":"unknown","confidenceSource":"profile","notes":"Listed on profile"}],"flattenedTokens":"protocol:OPENCLEW|unknown|profile"}},"adoption":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-09T23:01:57.293Z","emptyReason":null},"stars":null,"forks":null,"downloads":1915,"packageName":null,"latestVersion":"5.0.5","tractionLabel":"1.9K downloads"},"release":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-09T23:01:57.293Z","emptyReason":null},"lastUpdatedAt":"2026-10-09T23:01:57.293Z","lastCrawledAt":"2026-10-09T23:01:57.293Z","lastIndexedAt":null,"nextCrawlAt":"2026-10-10T23:01:57.293Z","lastVerifiedAt":null,"highlights":[{"version":"5.0.5","createdAt":"2026-09-24T05:15:38.910Z","changelog":"新增OCR置信度分级、判责与反欺诈示例、报告填写示例；合规与监管表新增维度；更新至2026-09-24","fileCount":3,"zipByteSize":12163},{"version":"5.0.4","createdAt":"2026-08-29T14:05:47.236Z","changelog":"v5.0.4: Removed duplicated disclaimer block, encoding verified clean. Receipt-type table gained a risk-point / verification column plus 3 new document types (outpatient prescription, e-invoice / blockchain receipt, claim application and loss certificate). Liability rules extended with rules 6-8 (disclosure and health declaration, insurable interest and beneficiary, accident nature vs exclusions). Anti-fraud checks extended with items 5-7 (time-geo conflict, ring/mediator patterns, loss-vs-fact deviation). Compliance table gained a violation-consequence column. New regulatory section refreshed through 2026-08-28.","fileCount":3,"zipByteSize":8799},{"version":"1.2.1","createdAt":"2026-06-28T13:14:46.721Z","changelog":"- Added an \"Insurance Regulatory Updates\" section summarizing key regulatory changes as of 2026-06-28. - Removed the redundant \"skill-card.md\" file for clarity and maintainability. - No changes to core advisory frameworks or template content. - Updated version number to 1.2.1.","fileCount":7,"zipByteSize":29980},{"version":"5.0.3","createdAt":"2026-06-01T22:35:27.210Z","changelog":"Security compliance: fixed garbled text, added capability declarations and advisory-only disclaimers; enriched content with detailed steps, rules, and report templates","fileCount":7,"zipByteSize":29191},{"version":"5.0.2","createdAt":"2026-06-01T15:11:49.686Z","changelog":"Security compliance update: added capability declarations and advisory-only disclaimers to meet ClawHub security scan requirements","fileCount":7,"zipByteSize":32659},{"version":"5.0.1","createdAt":"2026-06-01T14:17:07.461Z","changelog":"insurance-claims-intelligence 5.0.1 - Documentation update: SKILL.md updated for clarity and version bump to 5.0.1. - Outdated or redundant file skill-card.md removed. - No functional or API-impacting changes—maintenance release focused on documentation cleanup.","fileCount":7,"zipByteSize":33551},{"version":"5.0.0","createdAt":"2026-05-31T02:11:05.984Z","changelog":"融合阿里点金（Dianjin）金融数字员工精髓，版本升级至5.0.0","fileCount":7,"zipByteSize":32458},{"version":"3.0.2","createdAt":"2026-05-27T06:07:26.623Z","changelog":"**v3.0.2 changelog:** - Updated skill documentation (SKILL.md). - Note: SKILL.md now contains invalid (mojibake) characters and corrupted Chinese text; previous version was intact. No functionality code changes in this release.","fileCount":7,"zipByteSize":29963}]},"execution":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No published capability contract is available yet."},"installCommand":"clawhub skill install s17ewqc4f2s6gpcbm88hy7fgvn85kg1g:insurance-claims-intelligence","setupComplexity":"low","setupSteps":["Setup complexity is classified as HIGH. You must provision dedicated cloud infrastructure or an isolated VM. Do not run this directly on your local workstation.","Final validation: Expose the agent to a mock request payload inside a sandbox and trace the network egress before allowing access to real customer data."],"contract":{"contractStatus":"missing","authModes":[],"requires":[],"forbidden":[],"supportsMcp":false,"supportsA2a":false,"supportsStreaming":false,"inputSchemaRef":null,"outputSchemaRef":null,"dataRegion":null,"contractUpdatedAt":null,"sourceUpdatedAt":null,"freshnessSeconds":null},"invocationGuide":{"preferredApi":{"snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-insurance-claims-intelligence/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-insurance-claims-intelligence/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-insurance-claims-intelligence/trust"},"curlExamples":["curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-insurance-claims-intelligence/snapshot\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-insurance-claims-intelligence/contract\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-insurance-claims-intelligence/trust\""],"jsonRequestTemplate":{"query":"summarize this repo","constraints":{"maxLatencyMs":2000,"protocolPreference":["OPENCLEW"]}},"jsonResponseTemplate":{"ok":true,"result":{"summary":"...","confidence":0.9},"meta":{"source":"CLAWHUB","generatedAt":"2026-10-10T09:03:49.016Z"}},"retryPolicy":{"maxAttempts":3,"backoffMs":[500,1500,3500],"retryableConditions":["HTTP_429","HTTP_503","NETWORK_TIMEOUT"]}},"endpoints":{"dossierUrl":"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-insurance-claims-intelligence/dossier","snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-insurance-claims-intelligence/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-insurance-claims-intelligence/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-insurance-claims-intelligence/trust"}},"reliability":{"evidence":{"source":"runtime-metrics","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No trust, reliability, or runtime telemetry is available."},"trust":{"status":"unavailable","handshakeStatus":"UNKNOWN","verificationFreshnessHours":null,"reputationScore":null,"p95LatencyMs":null,"successRate30d":null,"fallbackRate":null,"attempts30d":null,"trustUpdatedAt":null,"trustConfidence":"unknown","sourceUpdatedAt":null,"freshnessSeconds":null},"decisionGuardrails":{"doNotUseIf":["Contract metadata is missing or unavailable for deterministic execution."],"safeUseWhen":[],"riskFlags":["missing_or_unavailable_contract","trust_data_unavailable","schema_references_missing"],"operationalConfidence":"low"},"executionMetrics":{"observedLatencyMsP50":null,"observedLatencyMsP95":null,"estimatedCostUsd":null,"uptime30d":null,"rateLimitRpm":null,"rateLimitBurst":null,"lastVerifiedAt":null,"verificationSource":null},"runtimeMetrics":{"successRate":null,"avgLatencyMs":null,"avgCostUsd":null,"hallucinationRate":null,"retryRate":null,"disputeRate":null,"p50Latency":null,"p95Latency":null,"lastUpdated":null}},"benchmarks":{"evidence":{"source":"no-benchmark-data","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No benchmark suites or observed failure patterns are available."},"suites":[],"failurePatterns":[]},"artifacts":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"high","updatedAt":"2026-10-09T23:01:57.293Z","emptyReason":null},"readme":"Skill: Insurance Claims Intelligence\n\nOwner: gechengling\n\nSummary: 提供多模态医疗票据OCR识别、智能判责、反欺诈检测和全险种覆盖的保险理赔智能分析与自动化支持。\n\nTags: advisory-only:1.2.0, ai-agent:1.1.0, anti-fraud:5.0.4, anti-fraud-checklist:1.2.0, banking:5.0.0, bilingual:1.1.0, china-insurance:1.2.0, chinese-market:1.0.0, claims:5.0.4, claims-advisory:1.2.0, claims-processing:1.1.0, compliance:1.2.0, decision-support:1.2.0, deepseek:1.1.0, dianjin:5.0.0, finance:5.0.0, fraud-detection:1.0.0, human-review-required:1.2.0, insurance:5.0.4, insurance-claims:1.2.0, insurance-claims-intelligence:5.0.5, insurance-tech:1.2.0, intelligent-claims:1.1.0, latest:5.0.5, li-pei:1.0.0, medical-ocr:1.1.0, medical-ocr-guidance:1.2.0, ocr:5.0.4\n\nVersion history:\n\nv5.0.5 | 2026-09-24T05:15:38.910Z | user\n\n新增OCR置信度分级、判责与反欺诈示例、报告填写示例；合规与监管表新增维度；更新至2026-09-24\n\nv5.0.4 | 2026-08-29T14:05:47.236Z | user\n\nv5.0.4: Removed duplicated disclaimer block, encoding verified clean. Receipt-type table gained a risk-point / verification column plus 3 new document types (outpatient prescription, e-invoice / blockchain receipt, claim application and loss certificate). Liability rules extended with rules 6-8 (disclosure and health declaration, insurable interest and beneficiary, accident nature vs exclusions). Anti-fraud checks extended with items 5-7 (time-geo conflict, ring/mediator patterns, loss-vs-fact deviation). Compliance table gained a violation-consequence column. New regulatory section refreshed through 2026-08-28.\n\nv1.2.1 | 2026-06-28T13:14:46.721Z | auto\n\n- Added an \"Insurance Regulatory Updates\" section summarizing key regulatory changes as of 2026-06-28.\n- Removed the redundant \"skill-card.md\" file for clarity and maintainability.\n- No changes to core advisory frameworks or template content.\n- Updated version number to 1.2.1.\n\nv5.0.3 | 2026-06-01T22:35:27.210Z | user\n\nSecurity compliance: fixed garbled text, added capability declarations and advisory-only disclaimers; enriched content with detailed steps, rules, and report templates\n\nv5.0.2 | 2026-06-01T15:11:49.686Z | user\n\nSecurity compliance update: added capability declarations and advisory-only disclaimers to meet ClawHub security scan requirements\n\nv5.0.1 | 2026-06-01T14:17:07.461Z | auto\n\ninsurance-claims-intelligence 5.0.1\n\n- Documentation update: SKILL.md updated for clarity and version bump to 5.0.1.\n- Outdated or redundant file skill-card.md removed.\n- No functional or API-impacting changes—maintenance release focused on documentation cleanup.\n\nv5.0.0 | 2026-05-31T02:11:05.984Z | user\n\n融合阿里点金（Dianjin）金融数字员工精髓，版本升级至5.0.0\n\nv3.0.2 | 2026-05-27T06:07:26.623Z | auto\n\n**v3.0.2 changelog:**\n\n- Updated skill documentation (SKILL.md).\n- Note: SKILL.md now contains invalid (mojibake) characters and corrupted Chinese text; previous version was intact. No functionality code changes in this release.\n\nv3.0.1 | 2026-05-25T03:30:06.344Z | auto\n\n- Added a new section with recent insurance regulatory updates (as of 2026-05-25), including changes to auto insurance, expanded disease coverage in health insurance, and new standards for major disease and special drug insurance in China.\n- Expanded keywords and updated the skill's description to include new claims areas such as 秒赔, 理赔决策, and more detailed insurance product types.\n- Minor formatting improvements in documentation.\n- Version updated to 3.0.1 and the `slug` field was introduced.\n\nv2.0.0 | 2026-05-11T06:09:47.380Z | auto\n\nNo file changes detected for this version. No updates or modifications have been made to the skill.\n\nv1.2.0 | 2026-05-04T18:26:01.472Z | user\n\nv1.2.0 Security & Compliance Update: (1) Added prominent disclaimers (EN/CN) stating this is advisory-only with NO executable models; (2) All accuracy figures now labeled as literature benchmarks, NOT validated results; (3) Removed ALL auto-approval language (no more 'auto-approve' or amount thresholds); (4) Added data security notices (PII redaction, API key management, vendor data retention policy); (5) Added anti-fraud data governance section (retention limits, access control, correction workflow); (6) All outputs now clearly labeled as 'drafts requiring licensed professional review'. Files changed: SKILL.md, README.md, references/claims_ocr_tech.md, references/claims_liability_engine.md.\n\nv1.1.0 | 2026-05-04T15:24:29.989Z | user\n\nv1.1.0 Bilingual optimization: English metadata + summaries; Bilingual README.md; SEO title optimization for international users; Keywords: insurance claims, intelligent claims, medical OCR.\n\nv1.0.0 | 2026-05-04T14:44:18.977Z | user\n\n首个保险行业全流程智能理赔Skill！整合多模态医疗票据OCR识别（百度/腾讯云/阿里云/合合信息）、智能理赔判责引擎（规则引擎+ML双驱动）、反欺诈知识图谱（GNN图神经网络）、行业全险种覆盖（医疗/重疾/寿险/意外/车险/财产险/团险），基于平安111极速赔、中国人寿智能理赔、太保数字劳动力实验室最佳实践构建，支持端到端理赔Agent，OCR识别→判责→核赔→反欺诈→合规检查全自动处理\n\nArchive index:\n\nArchive v5.0.5: 3 files, 12163 bytes\n\nFiles: skill-card.md (2391b), SKILL.md (23722b), _meta.json (148b)\n\nFile v5.0.5:SKILL.md\n\n---\nname: Insurance Claims Intelligence Expert\ndescription: Advisory skill for insurance claims processing workflows — provides templates, checklists, and decision-support frameworks for medical OCR, liability determination, anti-fraud assessment, and claims reporting. Human review required for all claim decisions. Keywords: insurance claims, claims advisory, medical OCR, anti-fraud, insurance tech, China insurance, decision support, 智能理赔, 理赔风控, 医疗单据识别, 责任认定, 理赔报告, 秒赔, 理赔决策, 医疗险理赔, 重疾理赔, 车险理赔.\nslug: insurance-claims-intelligence\nversion: 5.0.5\n\ncapabilities:\n  - educational-reference\n  - advisory-only\n  - requires-human-review\n  - illustrative-code-samples\n---\n\n# Insurance Claims Intelligence Expert / 保险行业智能理赔专家\n\n> **⚠️ SECURITY NOTICE / 安全声明**\n> - **Type:** Educational reference / analytical framework ONLY\n> - **No runnable package is bundled:** the architecture sketches and short code\n  fragments below are illustrative reference material for you to adapt in your own\n  environment; this skill ships no installer, service, or executable payload\n> - **No persistent storage, network calls, background execution, or credential collection**\n> - **All outputs are for reference only and require human review before real-world application**\n> - **This skill does NOT provide financial, legal, or insurance advice**\n> - **Users must exercise their own judgment and consult qualified professionals**\n\n\n\n> **⚠️ DISCLAIMER / 免责声明**\n> - **English:** This skill provides advisory templates, checklists, and decision-support frameworks ONLY. It does NOT contain executable models, trained GNN weights, or production OCR integrations. All accuracy figures (e.g., \"92%-96%\") are literature-reported benchmarks or design targets, NOT validated results of this skill. ALL claim approvals, denials, payout amounts, and fraud labels MUST be reviewed and confirmed by a licensed insurance professional before use. This skill is NOT a substitute for human judgment or regulatory compliance review.\n> - **中文：** 本Skill仅提供咨询模板、检查清单和决策支持框架，不含可执行模型、已训练GNN权重或生产级OCR集成。所有准确率数据（如\"92%-96%\"）均来自文献基准或设计目标，非本Skill实测结果。所有理赔核准、拒付、赔付金额及欺诈标签，**必须经持证保险专业人士审核确认后方可使用**。本Skill不可替代人工判断或监管合规审查。\n\n> **🔒 数据最小化前置声明 / Data Minimisation (apply before any OCR or analysis step)**\n> 1. 先问“这一条数据是否必需”：非理赔必需的字段（如完整身份证号、无关病史、家庭成员信息）一律不采集、不粘贴、不上传。\n> 2. 优先使用脱敏副本：姓名、证件号、联系电话默认以掩码形式处理（如 `张*`、`310***********1234`）。\n> 3. 单次任务单次授权：明确本次处理的用途与范围，任务结束后删除临时文件与对话中的原始影像描述。\n> 4. 任何需要写入文件或对外发送的结果，先在对话中完整展示给用户预览，经用户明确确认后再落盘/发送。\n\n> **🔒 DATA SECURITY / 数据安全**\n> - Medical invoices, diagnosis records, and claimant data are sensitive personal information under China's Personal Information Protection Law (PIPL). Before using OCR features, obtain user consent, redact/remove unnecessary PII, prefer on-prem/private deployment for production, and confirm the OCR vendor's data retention and cross-border transfer terms.\n> - API keys and credentials MUST be stored in environment variables or a secret manager. Never hardcode keys in production systems.\n> - 生产环境部署须具备等保/国密合规能力；OCR 与模型服务应优先私有化，避免医疗敏感数据出域。\n\n---\n\n## Artifact Type / 作品类型\n\n**This is a documentation-and-template skill.** It contains:\n- ✅ Workflow checklists and decision trees\n- ✅ Report templates and output formats\n- ✅ Reference architectures and integration guidance\n- ✅ Example Python code (requires your own API keys and data)\n\nIt does NOT contain:\n- ❌ Pre-trained ML/GNN models\n- ❌ Executable OCR or claims processing code\n- ❌ Bundled third-party API credentials\n\n---\n\n## Trigger Keywords / 触发关键词\n\n**English Triggers:** insurance claims advisory, claims workflow, claim analysis, medical OCR guidance, insurance fraud assessment, claim liability review, policy clause analysis, anti-fraud checklist, insurance tech reference, claims report template\n\n**中文触发词：** 保险理赔咨询 / 理赔流程指导 / 理赔分析 / 医疗发票识别指导 / 理赔判责参考 / 责任认定流程 / 医疗险理赔 / 重疾险理赔 / 寿险理赔 / 意外险理赔 / 车险理赔 / 财产险理赔 / 理赔反欺诈 / 欺诈检测参考 / 骗保识别指导 / 理赔风控参考 / 保险条款解读 / 责任免除说明 / 保障范围分析 / 赔付比例计算 / 产品对比参考 / 条款比对指导 / 合同解读参考\n\n---\n\n## Core Capabilities / 核心能力（咨询框架）\n\n### 1. Medical Receipt OCR — Guidance Framework / 医疗票据OCR识别（指导框架）\n\n**支持的票据类型（覆盖全场景）：**\n\n| Receipt Type / 票据类型 | Extracted Fields / 识别内容 | Insurance Types / 适用险种 | 风险点/校验要点 |\n|------------------------|-------------------|------------------|----------------|\n| 全国统一门诊发票 | 发票号、医院、金额、明细项目 | 医疗险、意外险 | 校验发票号码在税务平台真实性、医院等级与条款约定是否一致 |\n| 全国统一住院发票 | 入院/出院日期、总金额、自费比例 | 医疗险、重疾险 | 比对住院天数与出院小结、关注自费/自付比例是否超条款上限 |\n| 医疗费用明细清单 | 药品明细、检查项目、单价、数量 | 医疗险 | 核对医保目录内外用药、识别重复收费与超限价项目 |\n| 医保结算单 | 医保账户支付、自付金额、报销比例 | 医疗险 | 验证医保结算数据与发票金额勾稽关系 |\n| 出院小结 | 诊断、住院天数、治疗经过 | 重疾险、寿险 | 关注主诊断与重疾定义匹配、既往症时间线 |\n| 病历首页 | 主要诊断、手术名称、ICD编码 | 重疾险 | ICD编码与重疾/轻症定义映射校验 |\n| 检查报告单 | 影像报告、检验结果 | 重疾险 | 检验数值与诊断结论一致性、报告时间线 |\n| 费用结算单 | 分项金额、总计金额 | 财产险、责任险 | 损失金额第三方佐证、免赔与责任限额核对 |\n| 处方笺（门急诊） | 药品名称、剂量、用法、开方医师 | 医疗险、重疾险 | 处方与诊断相关性、超量开药识别 |\n| 电子发票/区块链票据 | 发票代码、校验码、开具平台 | 全险种 | 链上验真、防止重复理赔与克隆发票 |\n| 理赔申请书/出险证明 | 出险时间地点、事故经过、受益人 | 全险种 | 出险时间是否在保险期内、事故性质与免责比对 | 高 | 申请书与事故证明时间不一致，需第三方证明补强 |\n\n### OCR 字段置信度分级与人工复核策略（2026-09-24 新增）\n\n| 字段类别 | 示例字段 | 建议置信度阈值 | 低于阈值的处理 | 可否自动带入赔款计算 |\n|---------|---------|--------------|--------------|-------------------|\n| 金额类 | 总金额、自付金额、医保支付 | ≥ 0.98 | 必须人工核对原始票据 | 否，须人工确认 |\n| 日期类 | 就诊日、入院/出院日 | ≥ 0.95 | 人工比对病历与结算单 | 否 |\n| 编码类 | ICD 编码、药品编码 | ≥ 0.95 | 人工映射校验 | 否 |\n| 机构类 | 医院名称、医院等级 | ≥ 0.90 | 以官方名录为准 | 否 |\n| 文本描述类 | 诊断名称、手术名称 | ≥ 0.85 | 人工通读确认 | 否 |\n\n**OCR 落地两条经验**：① 置信度只用于排队，不用于自动结论——低置信度件优先送人工，而不是直接拒赔；\n② 同一张票据同时出现“金额改动痕迹”与“连号发票”时，直接升级为反欺诈调查件，不再走常规核赔。\n\n> **⚠️ OCR Data Handling / OCR数据处理提醒**\n> - Only send necessary fields to OCR providers; redact/unnecessary PII beforehand.\n> - Confirm the OCR vendor's data retention policy (Prefer: no storage / auto-delete within 24h).\n> - For production use, prefer private on-prem OCR deployment to avoid third-party data transfer.\n> - **中文：** 仅发送必要字段至OCR服务商；事前脱敏/删除非必要个人信息；确认OCR厂商数据留存策略（优先：不留存/24小时内自动删除）；生产环境优先使用私有化本地部署OCR，避免第三方数据传输。\n\n**参考技术架构（需自行集成）：**\n\n```text\n原始图像\n  ↓\n图像预处理（去噪/倾斜校正/二值化）\n  ↓\nCNN特征提取（ResNet50/EfficientNet）—— 需自行训练或调用云服务API\n  ↓\nRNN序列建模（BiLSTM）+ Attention机制\n  ↓\nCRF层解码 → 结构化文本输出\n  ↓\n字段标准化 → JSON/表格结构化结果\n```\n\n### 2. Liability Determination — Advisory Framework / 理赔判责引擎（咨询框架）\n\n**咨询级判责检查清单（需人工逐项确认）：**\n\n```text\n规则1：等待期检查（人工确认）\n  └─ 出险日期 - 保单生效日 < 等待期 → 建议拒付，需人工复核\n\n规则2：既往症筛查（人工确认）\n  └─ 既往症库匹配 → 责任免除 → 建议拒付/比例赔付，需人工复核\n\n规则3：免赔额校验（人工确认）\n  └─ 累计自付金额 < 免赔额 → 建议暂不赔付，需人工复核\n\n规则4：就诊机构核查（人工确认）\n  └─ 非二级及以上公立医院（需视条款）→ 提示确认，需人工复核\n\n规则5：险种责任匹配（人工确认）\n  └─ 就诊科室/诊断是否符合条款保障范围 → 建议全额/比例/拒付，需人工复核\n\n规则6：如实告知/健康告知核查（人工确认）\n  └─ 投保前未如实告知既往症/健康状况 → 依《保险法》第十六条评估解除合同权与拒赔风险，需人工复核\n\n规则7：保险利益与受益人核验（人工确认）\n  └─ 索赔申请人是否具备保险利益、受益人指定是否有效 → 身份与关系证明核验，需人工复核\n\n规则8：事故性质与免责比对（人工确认）\n  └─ 出险原因是否落入责任免除（如违法犯罪、酒驾、战争等）→ 命中免责建议拒付，需人工复核\n```\n\n**判责规则应用示例（2026-09-24 新增，均为示意，结论须人工复核）：**\n\n| 示例 | 事实要点 | 命中规则 | 建议结论（草稿） | 仍需人工确认的点 |\n|------|---------|---------|----------------|-----------------|\n| 示例一：等待期刚过即出险 | 保单 2026-03-01 生效，等待期 90 天，就诊日 2026-05-20，诊断甲状腺结节 | 规则1 等待期检查 | 就诊日在等待期内（第 80 天），建议按条款拒付并说明依据 | 条款对“等待期内就诊、等待期后确诊”的约定；是否存在续保无等待期情形 |\n| 示例二：免赔额未达线 | 年度免赔额 1 万元，本次自付 3,800 元，年内累计 6,200 元 | 规则3 免赔额校验 | 累计未达免赔额，建议暂不赔付并留存累计记录 | 是否为家庭共享免赔额；社保报销部分是否计入 |\n| 示例三：就诊机构不符 | 条款约定二级及以上公立医院，实际就诊于私立口腔诊所 | 规则4 就诊机构核查 | 建议比例赔付或拒付，需向客户说明条款依据 | 条款是否含“特定医疗机构清单”扩展；是否属急诊就近就医 |\n| 示例四：免责条款命中 | 事故认定书载明酒驾 | 规则8 事故性质与免责比对 | 建议拒付并出具书面拒赔通知，注明免责条款编号 | 免责条款是否已履行明确说明义务；是否存在条款解释争议 |\n\n> **⚠️ IMPORTANT / 重要提醒**\n> The liability determination output is a **decision-support suggestion ONLY**. Final approval/denial MUST be made by an authorized human reviewer. This skill does NOT auto-approve any claim amount.\n> **中文：** 判责输出**仅为决策支持建议**，最终核准/拒付**必须由授权人工审核员作出**。本Skill不对任何理赔金额进行自动审批。\n\n### 3. Anti-Fraud Assessment — Advisory Framework / 反欺诈评估（咨询框架）\n\n**反欺诈检查清单（咨询级）：**\n\n```text\n检查项1：就诊频率异常\n  └─ 同一被保人短期内多次就诊 → 标记，建议人工调查\n\n检查项2：票据真实性验证\n  └─ 发票号重复 / 医院不存在 / 金额异常 → 标记，建议人工调查\n\n检查项3：诊断与用药匹配性\n  └─ 诊断与开具药品明显不符 → 标记，建议人工调查\n\n检查项4：关系网络异常\n  └─ 同一医生/医院集中出现在多起理赔 → 标记，建议人工调查\n\n检查项5：时间-地理冲突\n  └─ 同一被保人短时间内异地（甚至跨国）连续就诊/出险 → 标记，建议人工调查\n\n检查项6：团伙/中介特征\n  └─ 多起理赔共享同一代理人、同一联系电话或同一银行账户 → 标记，建议人工调查\n\n检查项7：损失与事实背离\n  └─ 申报损失金额显著高于同类案件均值、缺乏第三方佐证 → 标记，建议人工调查\n```\n\n**欺诈风险分级与处置路径（2026-09-24 新增）：**\n\n| 风险等级 | 触发情形（示意） | 建议处置 | 时限 | 移送标准 |\n|---------|----------------|---------|------|---------|\n| 低（观察） | 单一可疑点，如票据影像清晰度低 | 常规核赔 + 记录留痕 | 正常时效内 | 不移送 |\n| 中（核查） | 两个可疑点，如就诊频繁 + 金额异常 | 补充材料 + 电话回访核实 | 3 个工作日内联系客户 | 不移送，内部登记 |\n| 高（调查） | 三个及以上可疑点，或命中团伙/中介特征 | 转欺诈调查岗，暂停赔付 | 24 小时内分派 | 涉嫌金额达标即按《反保险欺诈工作办法》移送 |\n| 极高（移送） | 伪造票据、冒名顶替、内外勾结 | 立即冻结流程并报案 | 即时 | 一律移送并保留完整证据链 |\n\n**反欺诈判定两个示例（示意）：**\n- **示例一（时间-地理冲突）**：被保人 09-05 在甲市门诊、09-06 在乙市住院，两地相距 1,200 公里且无转诊记录。\n  命中检查项 5，风险等级建议判为“高”，转调查岗核实是否冒名就医或票据挪用。\n- **示例二（团伙特征）**：近 3 个月 11 起理赔共享同一联系电话与同一收款账户，涉及 3 家分支机构。\n  命中检查项 6，风险等级建议判为“极高”，冻结相关案件并按程序移送，同时回溯同类历史赔案。\n\n> **🔒 Anti-Fraud Data Governance / 反欺诈数据治理**\n> - Retention limit / 留存期限：反欺诈图谱数据建议留存不超过 2 年，除非监管要求的更长留存期。\n> - Access control / 访问控制：图谱查询权限仅开放给授权欺诈调查员，禁止非授权人员访问。\n> - Data correction workflow / 数据更正流程：被保人有权请求更正错误数据，必须在 15 个工作日内处理。\n> - Poisoning safeguard / 污染防护：新案件数据进入图谱前，须经人工审核确认，防止恶意污染。\n\n### 4. Claims Report Templates / 理赔报告模板\n\n```markdown\n# 理赔分析报告（咨询草稿）\n**生成时间**: YYYY-MM-DD HH:mm\n**案件编号**: CL-XXXXXXXX\n**险种类别**: [险种名称]\n**处理状态**: [咨询草稿 — 需人工审核]\n**免责声明**: 本报告为AI辅助生成的咨询草稿，所有结论须经持证理赔师审核确认后方可生效。\n---\n## 一、票据识别结果（仅供参考）\n## 二、责任认定分析（仅供参考）\n## 三、赔付计算参考（仅供参考）\n## 四、反欺诈风险评估（仅供参考）\n## 五、建议下一步行动（需人工确认）\n```\n\n**填写示例（2026-09-24 新增，示意性草稿）：**\n\n```markdown\n# 理赔分析报告（咨询草稿）\n**生成时间**: 2026-09-24 10:20\n**案件编号**: CL-20260924-0087\n**险种类别**: 百万医疗险（含院外特药）\n**处理状态**: 咨询草稿 — 需人工审核\n---\n## 一、票据识别结果（仅供参考）\n- 住院发票 1 张：总金额 48,260.00 元，医保支付 26,140.00 元，自付 22,120.00 元（置信度 0.97，已建议人工复核）\n- 出院小结 1 份：入院 2026-08-11，出院 2026-08-19，共 8 天，主诊断编码 C50.9\n## 二、责任认定分析（仅供参考）\n- 等待期：已过（生效 2025-01-01，等待期 30 天）→ 未命中拒付情形\n- 就诊机构：三级甲等公立医院，符合条款约定\n- 免责比对：未见条款列明免责情形\n## 三、赔付计算参考（仅供参考）\n- 免赔额 10,000.00 元，本次可计入自付 22,120.00 元\n- 参考赔付区间 =（22,120.00 − 10,000.00）× 100% = 12,120.00 元\n- 院外特药 3,860.00 元需单独核对药品清单与特药目录\n## 四、反欺诈风险评估（仅供参考）\n- 风险等级：低（观察）。无可疑点叠加，建议常规核赔并留痕\n## 五、建议下一步行动（需人工确认）\n- 由持证理赔师复核发票原件与医保结算单勾稽关系\n- 确认院外特药是否在指定药店及目录内\n```\n\n---\n\n## 最新监管动态（截至 2026-09-24）/ Latest Regulatory Updates\n\n| 时间 | 监管动态 | 对理赔实务影响 | 主要涉及环节 | 可信度标注 |\n|------|----------|---------------|\n| 2026-08 | 金融监管总局发布保险理赔服务提质增效通知，要求简化小额理赔材料、推广\"秒赔/快赔\"与线上自助理赔 | 医疗险、车险线上自助理赔成为标配，OCR+规则引擎前置核赔加快落地 | 报案受理、材料收集 | 以官方最新发布为准 |\n| 2026-09 | 理赔服务数字化与\"高效办成一件事\"持续推进，电子票据验真、影像件归档与无纸化留存要求提高 | 理赔系统需具备电子发票验真与全流程影像归档能力，纸质件缺失不再当然构成拒赔理由 | 材料收集、理算、归档 | 以官方最新发布为准 |\n| 2026-09 | 保险消费者权益保护与理赔时效披露要求持续强化，结案率与平均结案时长等指标更多用于机构评价 | 时效数据需可量化、可追溯、可对账，超期案件需有预警与升级机制 | 核赔、结案、报送 | 以官方最新发布为准 |\n| 2026-07 | 反保险欺诈监管协作机制升级，行业欺诈线索共享平台常态化运行 | 跨机构关系网络图谱、团伙识别规则需前置嵌入理赔系统 | 反欺诈、核赔 | 以官方最新发布为准 |\n| 2026-06 | 个人保险实名制与医疗数据共享试点扩围，理赔可调用卫健/医保脱敏数据 | 出险真实性核验效率提升，但须严格控制 PIPL 授权与最小必要原则 | 报案受理、核赔 | 以官方最新发布为准 |\n| 2026-05 | 《保险消费投诉处理管理办法》修订，强化理赔纠纷首问负责与限时办结 | 理赔结论须附可解释依据，人工复核留痕要求提高 | 投诉处理、结案 | 以官方最新发布为准 |\n| 2026-03 | 监管推动\"应赔尽赔、能赔快赔\"，将理赔服务纳入消费者权益保护考核 | 理赔时效与满意度成为机构评价关键指标 | 结案、报送 | 以官方最新发布为准 |\n\n> **说明**：以上动态截至 2026-09-24，具体以监管机构官方发布为准；本 Skill 仅作方法论参考，不替代合规审查。\n\n---\n\n## Compliance & Human Review / 合规与人工审核要求\n\n| Compliance Item / 合规项 | Regulatory Basis / 监管依据 | Human Review Requirement / 人工审核要求 | 违规后果/处罚风险 | 建议留痕材料 |\n|--------------------|--------------------|----------------------|----------------|\n| 理赔时效 | 《保险法》第23条 | 核定结果须经人工确认后发出 | 超期核定可处监管通报、责令改正并赔偿迟延利息 | 报案时间戳、材料收齐时间戳、核定发出时间戳 |\n| 材料完整性 | 理赔管理办法 | 缺失材料列表由人工最终确认 | 材料缺失即拒赔易引发投诉与诉讼败诉 | 一次性告知清单及送达记录 |\n| 反欺诈合规 | 《反保险欺诈工作办法》2024 | 欺诈标记须经人工调查确认 | 应移送未移送涉嫌犯罪线索将追责 | 风险评分依据、调查记录、移送回执 |\n| 数据安全 | 《个人信息保护法》 | 医疗数据脱敏处理须经人工检查 | 泄露/非法提供个人信息可处高额罚款乃至刑事责任 | 授权书、脱敏前后对照、访问日志 |\n| 资金安全 | 反洗钱规定 | 大额理赔须人工复核 + 主管审批 | 未履行反洗钱义务可被处罚并冻结业务 | 身份识别记录、受益所有人核查、审批签批 |\n| 监管报送 | 保险监管报送与消费者权益保护评价要求 | 理赔数据报送须经人工复核 | 瞒报漏报将被监管处罚并影响评价评级 | 报送口径说明、复核人签署、报送回执 |\n\n**ALL outputs of this skill are drafts requiring licensed professional review. / 本Skill所有输出均为草稿，须经持证专业人士审核。**\n\n---\n\n## Output Format / 输出格式规范\n\nAll outputs must include the following disclaimer:\n\n```markdown\n> ⚠️ **免责声明 / Disclaimer**\n> 本输出为AI辅助咨询草稿，所有理赔决定、拒付结论、赔付金额及欺诈标签\n> 须经【持证保险理赔师】审核确认后方可生效。\n> This is an AI-assisted draft. All claim decisions must be reviewed by a\n> licensed insurance adjuster before taking effect.\n```\n\n---\n\n## 保存与外发前确认 / Save & Send Confirmation（2026-09-24 新增）\n\n1. 先在对话中完整展示理赔分析报告全文，供用户逐段预览；\n2. 用户明确确认后，才写入本地文件或对外发送；未经确认不落盘、不发送；\n3. 落盘时一并记录：生成时间、数据/票据截止日期、主要材料来源、审核人。\n\n---\n\n## 变更记录 / Changelog\n\n| 版本 | 日期 | 变更摘要 |\n|------|------|---------|\n| 5.0.5 | 2026-09-24 | 新增 OCR 字段置信度分级与复核策略、判责规则 4 则应用示例、欺诈风险分级与 2 则判定示例、报告模板填写示例；票据表新增复核优先级与高频差错列；合规表新增留痕材料列；监管动态更新至 2026-09-24 并新增 2 条；修正能力与代码样例不一致的声明 |\n| 5.0.4 | 2026-08-29 | 补充监管动态、合规与人工审核要求 |\n\n---\n\n## References / 参考文件\n\n| File / 文件 | Content / 内容说明 |\n|------|---------|\n| `references/claims_ocr_tech.md` | OCR技术架构参考 + 4家服务商对比 + Python示例代码（需自行配置API Key） |\n| `references/claims_liability_engine.md` | 判责规则参考 + 机器学习模型参考 + 3家公司实践参考 |\n| `references/claims_report_templates.md` | 报告模板 + 7种险种通知书模板参考 |\n\n> **⚠️ Reference files contain example code only. You must:**\n> - Provide your own API keys and store them in environment variables\n> - Provide your own training data and models\n> - Ensure human review of all outputs before use\n> - **中文：** 参考文件仅含示例代码，您必须：自行提供API密钥并存入环境变量；自行准备训练数据和模型；确保所有输出经人工审核后方可使用。\n\nFile v5.0.5:_meta.json\n\n{\n  \"ownerId\": \"kn74e704j3ygjcygnpf02rdvd185js13\",\n  \"slug\": \"insurance-claims-intelligence\",\n  \"version\": \"5.0.5\",\n  \"publishedAt\": 1790226938910\n}\n\nFile v5.0.5:skill-card.md\n\n## Description:\n\nAdvisory skill for insurance claims processing workflows that provides templates, checklists, and decision-support frameworks for medical OCR, liability determination, anti-fraud assessment, and claims reporting.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[gechengling](https://clawhub.ai/user/gechengling)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nExternal insurance operations teams, claims reviewers, and developers use this skill to draft claims workflow guidance, OCR review checklists, liability review notes, anti-fraud triage prompts, and claims report templates. All claim decisions, payout amounts, denial language, and fraud labels require qualified human review.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Claims guidance may affect sensitive insurance outcomes if used as a final decision.\n\nMitigation: Use outputs only as drafts and require a qualified insurance professional to review all claim decisions, payout amounts, denial language, and fraud labels before use.\n\nRisk: Claims workflows may involve personal, medical, or identifying data.\n\nMitigation: Do not paste unnecessary personal or medical data; redact identifiers where possible and follow data minimization practices before OCR or analysis.\n\nRisk: Regulatory references and compliance practices can change over time.\n\nMitigation: Verify regulatory references against current official sources before relying on them in claims handling or reporting.\n\n## Reference(s):\n\n- [ClawHub release page](https://clawhub.ai/gechengling/skills/insurance-claims-intelligence)\n- [Skill source artifact](artifact/SKILL.md)\n\n## Skill Output:\n\n**Output Type(s):** [Text, Markdown, Code, Configuration, Guidance]\n\n**Output Format:** [Markdown drafts with checklists, tables, report templates, and illustrative code snippets]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [Outputs are advisory drafts and must include a human-review disclaimer before operational use.]\n\n## Skill Version(s):\n\n5.0.5 (source: frontmatter, release metadata, changelog)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nArchive v5.0.4: 3 files, 8799 bytes\n\nFiles: skill-card.md (2524b), SKILL.md (15352b), _meta.json (148b)\n\nFile v5.0.4:SKILL.md\n\n---\nname: Insurance Claims Intelligence Expert\ndescription: Advisory skill for insurance claims processing workflows — provides templates, checklists, and decision-support frameworks for medical OCR, liability determination, anti-fraud assessment, and claims reporting. Human review required for all claim decisions. Keywords: insurance claims, claims advisory, medical OCR, anti-fraud, insurance tech, China insurance, decision support, 智能理赔, 理赔风控, 医疗单据识别, 责任认定, 理赔报告, 秒赔, 理赔决策, 医疗险理赔, 重疾理赔, 车险理赔.\nslug: insurance-claims-intelligence\nversion: 5.0.4\n\ncapabilities:\n  - educational-reference\n  - advisory-only\n  - requires-human-review\n  - no-executable-code\n---\n\n# Insurance Claims Intelligence Expert / 保险行业智能理赔专家\n\n> **⚠️ SECURITY NOTICE / 安全声明**\n> - **Type:** Educational reference / analytical framework ONLY\n> - **No executable code, scripts, or binaries are included in this skill**\n> - **No persistent storage, network calls, background execution, or credential collection**\n> - **All outputs are for reference only and require human review before real-world application**\n> - **This skill does NOT provide financial, legal, or insurance advice**\n> - **Users must exercise their own judgment and consult qualified professionals**\n\n\n\n> **⚠️ DISCLAIMER / 免责声明**\n> - **English:** This skill provides advisory templates, checklists, and decision-support frameworks ONLY. It does NOT contain executable models, trained GNN weights, or production OCR integrations. All accuracy figures (e.g., \"92%-96%\") are literature-reported benchmarks or design targets, NOT validated results of this skill. ALL claim approvals, denials, payout amounts, and fraud labels MUST be reviewed and confirmed by a licensed insurance professional before use. This skill is NOT a substitute for human judgment or regulatory compliance review.\n> - **中文：** 本Skill仅提供咨询模板、检查清单和决策支持框架，不含可执行模型、已训练GNN权重或生产级OCR集成。所有准确率数据（如\"92%-96%\"）均来自文献基准或设计目标，非本Skill实测结果。所有理赔核准、拒付、赔付金额及欺诈标签，**必须经持证保险专业人士审核确认后方可使用**。本Skill不可替代人工判断或监管合规审查。\n\n> **🔒 DATA SECURITY / 数据安全**\n> - Medical invoices, diagnosis records, and claimant data are sensitive personal information under China's Personal Information Protection Law (PIPL). Before using OCR features, obtain user consent, redact/remove unnecessary PII, prefer on-prem/private deployment for production, and confirm the OCR vendor's data retention and cross-border transfer terms.\n> - API keys and credentials MUST be stored in environment variables or a secret manager. Never hardcode keys in production systems.\n> - 生产环境部署须具备等保/国密合规能力；OCR 与模型服务应优先私有化，避免医疗敏感数据出域。\n\n---\n\n## Artifact Type / 作品类型\n\n**This is a documentation-and-template skill.** It contains:\n- ✅ Workflow checklists and decision trees\n- ✅ Report templates and output formats\n- ✅ Reference architectures and integration guidance\n- ✅ Example Python code (requires your own API keys and data)\n\nIt does NOT contain:\n- ❌ Pre-trained ML/GNN models\n- ❌ Executable OCR or claims processing code\n- ❌ Bundled third-party API credentials\n\n---\n\n## Trigger Keywords / 触发关键词\n\n**English Triggers:** insurance claims advisory, claims workflow, claim analysis, medical OCR guidance, insurance fraud assessment, claim liability review, policy clause analysis, anti-fraud checklist, insurance tech reference, claims report template\n\n**中文触发词：** 保险理赔咨询 / 理赔流程指导 / 理赔分析 / 医疗发票识别指导 / 理赔判责参考 / 责任认定流程 / 医疗险理赔 / 重疾险理赔 / 寿险理赔 / 意外险理赔 / 车险理赔 / 财产险理赔 / 理赔反欺诈 / 欺诈检测参考 / 骗保识别指导 / 理赔风控参考 / 保险条款解读 / 责任免除说明 / 保障范围分析 / 赔付比例计算 / 产品对比参考 / 条款比对指导 / 合同解读参考\n\n---\n\n## Core Capabilities / 核心能力（咨询框架）\n\n### 1. Medical Receipt OCR — Guidance Framework / 医疗票据OCR识别（指导框架）\n\n**支持的票据类型（覆盖全场景）：**\n\n| Receipt Type / 票据类型 | Extracted Fields / 识别内容 | Insurance Types / 适用险种 | 风险点/校验要点 |\n|------------------------|-------------------|------------------|----------------|\n| 全国统一门诊发票 | 发票号、医院、金额、明细项目 | 医疗险、意外险 | 校验发票号码在税务平台真实性、医院等级与条款约定是否一致 |\n| 全国统一住院发票 | 入院/出院日期、总金额、自费比例 | 医疗险、重疾险 | 比对住院天数与出院小结、关注自费/自付比例是否超条款上限 |\n| 医疗费用明细清单 | 药品明细、检查项目、单价、数量 | 医疗险 | 核对医保目录内外用药、识别重复收费与超限价项目 |\n| 医保结算单 | 医保账户支付、自付金额、报销比例 | 医疗险 | 验证医保结算数据与发票金额勾稽关系 |\n| 出院小结 | 诊断、住院天数、治疗经过 | 重疾险、寿险 | 关注主诊断与重疾定义匹配、既往症时间线 |\n| 病历首页 | 主要诊断、手术名称、ICD编码 | 重疾险 | ICD编码与重疾/轻症定义映射校验 |\n| 检查报告单 | 影像报告、检验结果 | 重疾险 | 检验数值与诊断结论一致性、报告时间线 |\n| 费用结算单 | 分项金额、总计金额 | 财产险、责任险 | 损失金额第三方佐证、免赔与责任限额核对 |\n| 处方笺（门急诊） | 药品名称、剂量、用法、开方医师 | 医疗险、重疾险 | 处方与诊断相关性、超量开药识别 |\n| 电子发票/区块链票据 | 发票代码、校验码、开具平台 | 全险种 | 链上验真、防止重复理赔与克隆发票 |\n| 理赔申请书/出险证明 | 出险时间地点、事故经过、受益人 | 全险种 | 出险时间是否在保险期内、事故性质与免责比对 |\n\n> **⚠️ OCR Data Handling / OCR数据处理提醒**\n> - Only send necessary fields to OCR providers; redact/unnecessary PII beforehand.\n> - Confirm the OCR vendor's data retention policy (Prefer: no storage / auto-delete within 24h).\n> - For production use, prefer private on-prem OCR deployment to avoid third-party data transfer.\n> - **中文：** 仅发送必要字段至OCR服务商；事前脱敏/删除非必要个人信息；确认OCR厂商数据留存策略（优先：不留存/24小时内自动删除）；生产环境优先使用私有化本地部署OCR，避免第三方数据传输。\n\n**参考技术架构（需自行集成）：**\n\n```text\n原始图像\n  ↓\n图像预处理（去噪/倾斜校正/二值化）\n  ↓\nCNN特征提取（ResNet50/EfficientNet）—— 需自行训练或调用云服务API\n  ↓\nRNN序列建模（BiLSTM）+ Attention机制\n  ↓\nCRF层解码 → 结构化文本输出\n  ↓\n字段标准化 → JSON/表格结构化结果\n```\n\n### 2. Liability Determination — Advisory Framework / 理赔判责引擎（咨询框架）\n\n**咨询级判责检查清单（需人工逐项确认）：**\n\n```text\n规则1：等待期检查（人工确认）\n  └─ 出险日期 - 保单生效日 < 等待期 → 建议拒付，需人工复核\n\n规则2：既往症筛查（人工确认）\n  └─ 既往症库匹配 → 责任免除 → 建议拒付/比例赔付，需人工复核\n\n规则3：免赔额校验（人工确认）\n  └─ 累计自付金额 < 免赔额 → 建议暂不赔付，需人工复核\n\n规则4：就诊机构核查（人工确认）\n  └─ 非二级及以上公立医院（需视条款）→ 提示确认，需人工复核\n\n规则5：险种责任匹配（人工确认）\n  └─ 就诊科室/诊断是否符合条款保障范围 → 建议全额/比例/拒付，需人工复核\n\n规则6：如实告知/健康告知核查（人工确认）\n  └─ 投保前未如实告知既往症/健康状况 → 依《保险法》第十六条评估解除合同权与拒赔风险，需人工复核\n\n规则7：保险利益与受益人核验（人工确认）\n  └─ 索赔申请人是否具备保险利益、受益人指定是否有效 → 身份与关系证明核验，需人工复核\n\n规则8：事故性质与免责比对（人工确认）\n  └─ 出险原因是否落入责任免除（如违法犯罪、酒驾、战争等）→ 命中免责建议拒付，需人工复核\n```\n\n> **⚠️ IMPORTANT / 重要提醒**\n> The liability determination output is a **decision-support suggestion ONLY**. Final approval/denial MUST be made by an authorized human reviewer. This skill does NOT auto-approve any claim amount.\n> **中文：** 判责输出**仅为决策支持建议**，最终核准/拒付**必须由授权人工审核员作出**。本Skill不对任何理赔金额进行自动审批。\n\n### 3. Anti-Fraud Assessment — Advisory Framework / 反欺诈评估（咨询框架）\n\n**反欺诈检查清单（咨询级）：**\n\n```text\n检查项1：就诊频率异常\n  └─ 同一被保人短期内多次就诊 → 标记，建议人工调查\n\n检查项2：票据真实性验证\n  └─ 发票号重复 / 医院不存在 / 金额异常 → 标记，建议人工调查\n\n检查项3：诊断与用药匹配性\n  └─ 诊断与开具药品明显不符 → 标记，建议人工调查\n\n检查项4：关系网络异常\n  └─ 同一医生/医院集中出现在多起理赔 → 标记，建议人工调查\n\n检查项5：时间-地理冲突\n  └─ 同一被保人短时间内异地（甚至跨国）连续就诊/出险 → 标记，建议人工调查\n\n检查项6：团伙/中介特征\n  └─ 多起理赔共享同一代理人、同一联系电话或同一银行账户 → 标记，建议人工调查\n\n检查项7：损失与事实背离\n  └─ 申报损失金额显著高于同类案件均值、缺乏第三方佐证 → 标记，建议人工调查\n```\n\n> **🔒 Anti-Fraud Data Governance / 反欺诈数据治理**\n> - Retention limit / 留存期限：反欺诈图谱数据建议留存不超过 2 年，除非监管要求的更长留存期。\n> - Access control / 访问控制：图谱查询权限仅开放给授权欺诈调查员，禁止非授权人员访问。\n> - Data correction workflow / 数据更正流程：被保人有权请求更正错误数据，必须在 15 个工作日内处理。\n> - Poisoning safeguard / 污染防护：新案件数据进入图谱前，须经人工审核确认，防止恶意污染。\n\n### 4. Claims Report Templates / 理赔报告模板\n\n```markdown\n# 理赔分析报告（咨询草稿）\n**生成时间**: YYYY-MM-DD HH:mm\n**案件编号**: CL-XXXXXXXX\n**险种类别**: [险种名称]\n**处理状态**: [咨询草稿 — 需人工审核]\n**免责声明**: 本报告为AI辅助生成的咨询草稿，所有结论须经持证理赔师审核确认后方可生效。\n---\n## 一、票据识别结果（仅供参考）\n## 二、责任认定分析（仅供参考）\n## 三、赔付计算参考（仅供参考）\n## 四、反欺诈风险评估（仅供参考）\n## 五、建议下一步行动（需人工确认）\n```\n\n---\n\n## 最新监管动态（截至2026-08-28）/ Latest Regulatory Updates\n\n| 时间 | 监管动态 | 对理赔实务影响 |\n|------|----------|---------------|\n| 2026-08 | 金融监管总局发布保险理赔服务提质增效通知，要求简化小额理赔材料、推广\"秒赔/快赔\"与线上自助理赔 | 医疗险、车险线上自助理赔成为标配，OCR+规则引擎前置核赔加快落地 |\n| 2026-07 | 反保险欺诈监管协作机制升级，行业欺诈线索共享平台常态化运行 | 跨机构关系网络图谱、团伙识别规则需前置嵌入理赔系统 |\n| 2026-06 | 个人保险实名制与医疗数据共享试点扩围，理赔可调用卫健/医保脱敏数据 | 出险真实性核验效率提升，但须严格控制 PIPL 授权与最小必要原则 |\n| 2026-05 | 《保险消费投诉处理管理办法》修订，强化理赔纠纷首问负责与限时办结 | 理赔结论须附可解释依据，人工复核留痕要求提高 |\n| 2026-03 | 监管推动\"应赔尽赔、能赔快赔\"，将理赔服务纳入消费者权益保护考核 | 理赔时效与满意度成为机构评价关键指标 |\n\n> **说明**：以上动态截至 2026-08-28，具体以监管机构官方发布为准；本 Skill 仅作方法论参考，不替代合规审查。\n\n---\n\n## Compliance & Human Review / 合规与人工审核要求\n\n| Compliance Item / 合规项 | Regulatory Basis / 监管依据 | Human Review Requirement / 人工审核要求 | 违规后果/处罚风险 |\n|--------------------|--------------------|----------------------|----------------|\n| 理赔时效 | 《保险法》第23条 | 核定结果须经人工确认后发出 | 超期核定可处监管通报、责令改正并赔偿迟延利息 |\n| 材料完整性 | 理赔管理办法 | 缺失材料列表由人工最终确认 | 材料缺失即拒赔易引发投诉与诉讼败诉 |\n| 反欺诈合规 | 《反保险欺诈工作办法》2024 | 欺诈标记须经人工调查确认 | 应移送未移送涉嫌犯罪线索将追责 |\n| 数据安全 | 《个人信息保护法》 | 医疗数据脱敏处理须经人工检查 | 泄露/非法提供个人信息可处高额罚款乃至刑事责任 |\n| 资金安全 | 反洗钱规定 | 大额理赔须人工复核 + 主管审批 | 未履行反洗钱义务可被处罚并冻结业务 |\n| 监管报送 | 保险监管报送与消费者权益保护评价要求 | 理赔数据报送须经人工复核 | 瞒报漏报将被监管处罚并影响评价评级 |\n\n**ALL outputs of this skill are drafts requiring licensed professional review. / 本Skill所有输出均为草稿，须经持证专业人士审核。**\n\n---\n\n## Output Format / 输出格式规范\n\nAll outputs must include the following disclaimer:\n\n```markdown\n> ⚠️ **免责声明 / Disclaimer**\n> 本输出为AI辅助咨询草稿，所有理赔决定、拒付结论、赔付金额及欺诈标签\n> 须经【持证保险理赔师】审核确认后方可生效。\n> This is an AI-assisted draft. All claim decisions must be reviewed by a\n> licensed insurance adjuster before taking effect.\n```\n\n---\n\n## References / 参考文件\n\n| File / 文件 | Content / 内容说明 |\n|------|---------|\n| `references/claims_ocr_tech.md` | OCR技术架构参考 + 4家服务商对比 + Python示例代码（需自行配置API Key） |\n| `references/claims_liability_engine.md` | 判责规则参考 + 机器学习模型参考 + 3家公司实践参考 |\n| `references/claims_report_templates.md` | 报告模板 + 7种险种通知书模板参考 |\n\n> **⚠️ Reference files contain example code only. You must:**\n> - Provide your own API keys and store them in environment variables\n> - Provide your own training data and models\n> - Ensure human review of all outputs before use\n> - **中文：** 参考文件仅含示例代码，您必须：自行提供API密钥并存入环境变量；自行准备训练数据和模型；确保所有输出经人工审核后方可使用。\n\nFile v5.0.4:_meta.json\n\n{\n  \"ownerId\": \"kn74e704j3ygjcygnpf02rdvd185js13\",\n  \"slug\": \"insurance-claims-intelligence\",\n  \"version\": \"5.0.4\",\n  \"publishedAt\": 1788012347236\n}\n\nFile v5.0.4:skill-card.md\n\n## Description:\n\nAdvisory skill for insurance claims processing workflows that provides templates, checklists, and decision-support frameworks for medical OCR, liability review, anti-fraud assessment, and claims reporting, with human review required for claim decisions.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[gechengling](https://clawhub.ai/user/gechengling)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nClaims teams, insurance operations staff, and developers use this skill to draft OCR review guidance, liability checks, anti-fraud assessments, compliance reminders, and claims report templates. Outputs are advisory drafts that require licensed professional review before claim decisions or customer-facing use.\n\n### Deployment Geography for Use:\n\nGlobal, with China-focused examples and compliance references.\n\n## Known Risks and Mitigations:\n\nRisk: Insurance, legal, regulatory, or fraud outputs could be mistaken for final claims decisions.\n\nMitigation: Treat outputs as advisory drafts and require licensed insurance professional review before approvals, denials, payout amounts, or fraud labels are used.\n\nRisk: Medical invoices, diagnosis records, and claimant data can include sensitive personal information.\n\nMitigation: Use consent, data minimization, redaction, retention controls, and transfer review before sending data to OCR or model providers.\n\nRisk: Jurisdiction-specific insurance and regulatory guidance may be outdated or inapplicable.\n\nMitigation: Verify legal and regulatory claims against authoritative sources for the intended jurisdiction before relying on them.\n\n## Reference(s):\n\n- [ClawHub skill page](https://clawhub.ai/gechengling/skills/insurance-claims-intelligence)\n- [Source skill artifact](artifact/SKILL.md)\n\n## Skill Output:\n\n**Output Type(s):** [text, markdown, code, configuration, guidance]\n\n**Output Format:** [Markdown advisory drafts, checklists, report templates, structured field guidance, and example code snippets.]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [All claim decisions, payout amounts, and fraud labels require licensed professional review.]\n\n## Skill Version(s):\n\n5.0.4 (source: server release metadata and SKILL.md frontmatter)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nArchive v1.2.1: 7 files, 29980 bytes\n\nFiles: README.md (7594b), references/claims_liability_engine.md (18889b), references/claims_ocr_tech.md (14039b), references/claims_report_templates.md (11395b), skill-card.md (2890b), SKILL.md (13587b), _meta.json (148b)\n\nFile v1.2.1:SKILL.md\n\n---\r\nname: Insurance Claims Intelligence Expert\r\ndescription: Advisory skill for insurance claims processing workflows — provides templates, checklists, and decision-support frameworks for medical OCR, liability determination, anti-fraud assessment, and claims reporting. Human review required for all claim decisions. Keywords: insurance claims, claims advisory, medical OCR, anti-fraud, insurance tech, China insurance, decision support, 智能理赔, 理赔风控, 医疗单据识别, 责任认定, 理赔报告, 秒赔, 理赔决策, 医疗险理赔, 重疾理赔, 车险理赔.\r\nslug: insurance-claims-intelligence\r\nversion: 1.2.1\r\n\r\ncapabilities:\r\n  - educational-reference\r\n  - advisory-only\r\n  - requires-human-review\r\n  - no-executable-code\r\n---\r\n\r\n# Insurance Claims Intelligence Expert / 保险行业智能理赔专家\r\n\r\n> **⚠️ SECURITY NOTICE / 安全声明**\r\n> - **Type:** Educational reference / analytical framework ONLY\r\n> - **No executable code, scripts, or binaries are included in this skill**\r\n> - **No persistent storage, network calls, background execution, or credential collection**\r\n> - **All outputs are for reference only and require human review before real-world application**\r\n> - **This skill does NOT provide financial, legal, or insurance advice**\r\n> - **Users must exercise their own judgment and consult qualified professionals**\r\n\r\n\r\n\r\n> **⚠️ DISCLAIMER / 免责声明**\r\n> - **English:** This skill provides advisory templates, checklists, and decision-support frameworks ONLY. It does NOT contain executable models, trained GNN weights, or production OCR integrations. All accuracy figures (e.g., \"92%-96%\") are literature-reported benchmarks or design targets, NOT validated results of this skill. ALL claim approvals, denials, payout amounts, and fraud labels MUST be reviewed and confirmed by a licensed insurance professional before use. This skill is NOT a substitute for human judgment or regulatory compliance review.\r\n> - **中文：** 本Skill仅提供咨询模板、检查清单和决策支持框架，不含可执行模型、已训练GNN权重或生产级OCR集成。所有准确率数据（如\"92%-96%\"）均来自文献基准或设计目标，非本Skill实测结果。所有理赔核准、拒付、赔付金额及欺诈标签，**必须经持证保险专业人士审核确认后方可使用**。本Skill不可替代人工判断或监管合规审查。\r\n\r\n> **🔒 DATA SECURITY / 数据安全**\r\n> - Medical invoices, diagnosis records, and claimant data are sensitive personal information under China's Personal Information Protection Law (PIPL). Before using OCR features, obtain user consent, redact/remove unnecessary PII, prefer on-prem/private deployment for production, and confirm the OCR vendor's data retention and cross-border transfer terms.\r\n> - API keys and credentials MUST be stored in environment variables or a secret manager. Never hardcode keys in production systems.\r\n> - **English:** This skill provides advisory templates, checklists, and decision-support frameworks ONLY. It does NOT contain executable models, trained GNN weights, or production OCR integrations. All accuracy figures (e.g., \"92%-96%\") are literature-reported benchmarks or design targets, NOT validated results of this skill. ALL claim approvals, denials, payout amounts, and fraud labels MUST be reviewed and confirmed by a licensed insurance professional before use. This skill is NOT a substitute for human judgment or regulatory compliance review.\r\n> - **中文：** 本Skill仅提供咨询模板、检查清单和决策支持框架，不含可执行模型、已训练GNN权重或生产级OCR集成。所有准确率数据（如\"92%-96%\"）均来自文献基准或设计目标，非本Skill实测结果。所有理赔核准、拒付、赔付金额及欺诈标签，**必须经持证保险专业人士审核确认后方可使用**。本Skill不可替代人工判断或监管合规审查。\r\n\r\n---\r\n\r\n## Artifact Type / 作品类型\r\n\r\n**This is a documentation-and-template skill.** It contains:\r\n- ✅ Workflow checklists and decision trees\r\n- ✅ Report templates and output formats\r\n- ✅ Reference architectures and integration guidance\r\n- ✅ Example Python code (requires your own API keys and data)\r\n\r\nIt does NOT contain:\r\n- ❌ Pre-trained ML/GNN models\r\n- ❌ Executable OCR or claims processing code\r\n- ❌ Bundled third-party API credentials\r\n\r\n---\r\n\r\n## Trigger Keywords / 触发关键词\r\n\r\n**English Triggers:** insurance claims advisory, claims workflow, claim analysis, medical OCR guidance, insurance fraud assessment, claim liability review, policy clause analysis, anti-fraud checklist, insurance tech reference, claims report template\r\n\r\n**中文触发词：** 保险理赔咨询 / 理赔流程指导 / 理赔分析 / 医疗发票识别指导 / 理赔判责参考 / 责任认定流程 / 医疗险理赔 / 重疾险理赔 / 寿险理赔 / 意外险理赔 / 车险理赔 / 财产险理赔 / 理赔反欺诈 / 欺诈检测参考 / 骗保识别指导 / 理赔风控参考 / 保险条款解读 / 责任免除说明 / 保障范围分析 / 赔付比例计算 / 产品对比参考 / 条款比对指导 / 合同解读参考\r\n\r\n---\r\n\r\n\r\n### 保险监管最新动态 [2026-06-28更新]\r\n\r\n| 动态类型 | 内容摘要 | 发布时间 | 影响范围 |\r\n|---------|---------|---------|---------|\r\n| 监管发布 | 金融监管总局发布《关于银行业保险业人工智能安全开发应用的指导意见》，界定承保理赔、风险管理等为AI高风险应用场景 | 2026-06-18 | 保险AI应用合规与风险管控 |\r\n| 监管计划 | NFRA发布《2026年规章制定工作计划》：保险资金运用管理办法、偿付能力管理办法纳入修订，流动性风险、网络安全管理办法新制定 | 2026-06-23 | 保险监管全链条 |\r\n| 监管施行 | NFRA 2026年第2号令《银行保险机构许可证管理办法》6月1日起施行，取消保险许可证统一为金融许可证 | 2026-06-01 | 保险业务资质与合规管理 |\r\n\r\n> **数据截止**: 2026-06-28 | 来源：国家金融监督管理总局、行业公开信息\r\n> **声明**: 以上动态供参考，具体以官方最新发布为准\r\n\r\n## Core Capabilities / 核心能力（咨询框架）\r\n\r\n### 1. Medical Receipt OCR — Guidance Framework / 医疗票据OCR识别（指导框架）\r\n\r\n**支持的票据类型（覆盖全场景）：**\r\n\r\n| Receipt Type / 票据类型 | Extracted Fields / 识别内容 | Insurance Types / 适用险种 |\r\n|------------------------|-------------------|------------------|\r\n| 全国统一门诊发票 | 发票号、医院、金额、明细项目 | 医疗险、意外险 |\r\n| 全国统一住院发票 | 入院/出院日期、总金额、自费比例 | 医疗险、重疾险 |\r\n| 医疗费用明细清单 | 药品明细、检查项目、单价、数量 | 医疗险 |\r\n| 医保结算单 | 医保账户支付、自付金额、报销比例 | 医疗险 |\r\n| 出院小结 | 诊断、住院天数、治疗经过 | 重疾险、寿险 |\r\n| 病历首页 | 主要诊断、手术名称、ICD编码 | 重疾险 |\r\n| 检查报告单 | 影像报告、检验结果 | 重疾险 |\r\n| 费用结算单 | 分项金额、总计金额 | 财产险、责任险 |\r\n\r\n> **⚠️ OCR Data Handling / OCR数据处理提醒**\r\n> - Only send necessary fields to OCR providers; redact/unnecessary PII beforehand.\r\n> - Confirm the OCR vendor's data retention policy (Prefer: no storage / auto-delete within 24h).\r\n> - For production use, prefer private on-prem OCR deployment to avoid third-party data transfer.\r\n> - **中文：** 仅发送必要字段至OCR服务商；事前脱敏/删除非必要个人信息；确认OCR厂商数据留存策略（优先：不留存/24小时内自动删除）；生产环境优先使用私有化本地部署OCR，避免第三方数据传输。\r\n\r\n**参考技术架构（需自行集成）：**\r\n\r\n```text\r\n原始图像\r\n  ↓\r\n图像预处理（去噪/倾斜校正/二值化）\r\n  ↓\r\nCNN特征提取（ResNet50/EfficientNet）—— 需自行训练或调用云服务API\r\n  ↓\r\nRNN序列建模（BiLSTM）+ Attention机制\r\n  ↓\r\nCRF层解码 → 结构化文本输出\r\n  ↓\r\n字段标准化 → JSON/表格结构化结果\r\n```\r\n\r\n### 2. Liability Determination — Advisory Framework / 理赔判责引擎（咨询框架）\r\n\r\n**咨询级判责检查清单（需人工逐项确认）：**\r\n\r\n```text\r\n规则1：等待期检查（人工确认）\r\n  └─ 出险日期 - 保单生效日 < 等待期 → 建议拒付，需人工复核\r\n\r\n规则2：既往症筛查（人工确认）\r\n  └─ 既往症库匹配 → 责任免除 → 建议拒付/比例赔付，需人工复核\r\n\r\n规则3：免赔额校验（人工确认）\r\n  └─ 累计自付金额 < 免赔额 → 建议暂不赔付，需人工复核\r\n\r\n规则4：就诊机构核查（人工确认）\r\n  └─ 非二级及以上公立医院（需视条款）→ 提示确认，需人工复核\r\n\r\n规则5：险种责任匹配（人工确认）\r\n  └─ 就诊科室/诊断是否符合条款保障范围 → 建议全额/比例/拒付，需人工复核\r\n```\r\n\r\n> **⚠️ IMPORTANT / 重要提醒**\r\n> The liability determination output is a **decision-support suggestion ONLY**. Final approval/denial MUST be made by an authorized human reviewer. This skill does NOT auto-approve any claim amount.\r\n> **中文：** 判责输出**仅为决策支持建议**，最终核准/拒付**必须由授权人工审核员作出**。本Skill不对任何理赔金额进行自动审批。\r\n\r\n### 3. Anti-Fraud Assessment — Advisory Framework / 反欺诈评估（咨询框架）\r\n\r\n**反欺诈检查清单（咨询级）：**\r\n\r\n```text\r\n检查项1：就诊频率异常\r\n  └─ 同一被保人短期内多次就诊 → 标记，建议人工调查\r\n\r\n检查项2：票据真实性验证\r\n  └─ 发票号重复 / 医院不存在 / 金额异常 → 标记，建议人工调查\r\n\r\n检查项3：诊断与用药匹配性\r\n  └─ 诊断与开具药品明显不符 → 标记，建议人工调查\r\n\r\n检查项4：关系网络异常\r\n  └─ 同一医生/医院集中出现在多起理赔 → 标记，建议人工调查\r\n```\r\n\r\n> **🔒 Anti-Fraud Data Governance / 反欺诈数据治理**\r\n> - Retention limit / 留存期限：反欺诈图谱数据建议留存不超过 2 年，除非监管要求的更长留存期。\r\n> - Access control / 访问控制：图谱查询权限仅开放给授权欺诈调查员，禁止非授权人员访问。\r\n> - Data correction workflow / 数据更正流程：被保人有权请求更正错误数据，必须在 15 个工作日内处理。\r\n> - Poisoning safeguard / 污染防护：新案件数据进入图谱前，须经人工审核确认，防止恶意污染。\r\n\r\n### 4. Claims Report Templates / 理赔报告模板\r\n\r\n```markdown\r\n# 理赔分析报告（咨询草稿）\r\n**生成时间**: YYYY-MM-DD HH:mm\r\n**案件编号**: CL-XXXXXXXX\r\n**险种类别**: [险种名称]\r\n**处理状态**: [咨询草稿 — 需人工审核]\r\n**免责声明**: 本报告为AI辅助生成的咨询草稿，所有结论须经持证理赔师审核确认后方可生效。\r\n---\r\n## 一、票据识别结果（仅供参考）\r\n## 二、责任认定分析（仅供参考）\r\n## 三、赔付计算参考（仅供参考）\r\n## 四、反欺诈风险评估（仅供参考）\r\n## 五、建议下一步行动（需人工确认）\r\n```\r\n\r\n---\r\n\r\n## Compliance & Human Review / 合规与人工审核要求\r\n\r\n| Compliance Item / 合规项 | Regulatory Basis / 监管依据 | Human Review Requirement / 人工审核要求 |\r\n|--------------------|--------------------|----------------------|\r\n| 理赔时效 | 《保险法》第23条 | 核定结果须经人工确认后发出 |\r\n| 材料完整性 | 理赔管理办法 | 缺失材料列表由人工最终确认 |\r\n| 反欺诈合规 | 《反保险欺诈工作办法》2024 | 欺诈标记须经人工调查确认 |\r\n| 数据安全 | 《个人信息保护法》 | 医疗数据脱敏处理须经人工检查 |\r\n| 资金安全 | 反洗钱规定 | 大额理赔须人工复核 + 主管审批 |\r\n\r\n**ALL outputs of this skill are drafts requiring licensed professional review. / 本Skill所有输出均为草稿，须经持证专业人士审核。**\r\n\r\n---\r\n\r\n## Output Format / 输出格式规范\r\n\r\nAll outputs must include the following disclaimer:\r\n\r\n```markdown\r\n> ⚠️ **免责声明 / Disclaimer**\r\n> 本输出为AI辅助咨询草稿，所有理赔决定、拒付结论、赔付金额及欺诈标签\r\n> 须经【持证保险理赔师】审核确认后方可生效。\r\n> This is an AI-assisted draft. All claim decisions must be reviewed by a\r\n> licensed insurance adjuster before taking effect.\r\n```\r\n\r\n---\r\n\r\n## References / 参考文件\r\n\r\n| File / 文件 | Content / 内容说明 |\r\n|------|---------|\r\n| `references/claims_ocr_tech.md` | OCR技术架构参考 + 4家服务商对比 + Python示例代码（需自行配置API Key） |\r\n| `references/claims_liability_engine.md` | 判责规则参考 + 机器学习模型参考 + 3家公司实践参考 |\r\n| `references/claims_report_templates.md` | 报告模板 + 7种险种通知书模板参考 |\r\n\r\n> **⚠️ Reference files contain example code only. You must:**\r\n> - Provide your own API keys and store them in environment variables\r\n> - Provide your own training data and models\r\n> - Ensure human review of all outputs before use\r\n> - **中文：** 参考文件仅含示例代码，您必须：自行提供API密钥并存入环境变量；自行准备训练数据和模型；确保所有输出经人工审核后方可使用。\n\nFile v1.2.1:README.md\n\n# Insurance Claims Intelligence Expert / 保险行业智能理赔专家\r\n\r\n> **⚠️ DISCLAIMER / 免责声明**\r\n> - **English:** This is an **advisory and template-only skill**. It does NOT contain executable models, trained GNN weights, or production OCR integrations. All accuracy figures (e.g., \"92%-96%\") are literature-reported benchmarks or design targets, **NOT validated results** of this skill. ALL claim approvals, denials, payout amounts, and fraud labels **MUST be reviewed and confirmed by a licensed insurance professional** before use. This skill is NOT a substitute for human judgment or regulatory compliance review.\r\n> - **中文：** 本Skill**仅为咨询模板和参考框架**，不含可执行模型、已训练GNN权重或生产级OCR集成。所有准确率数据（如\"92%-96%\"）均来自文献基准或设计目标，**非本Skill实测结果**。所有理赔核准、拒付、赔付金额及欺诈标签，**必须经持证保险专业人士审核确认后方可使用**。本Skill不可替代人工判断或监管合规审查。\r\n\r\n> **🔒 DATA SECURITY NOTICE / 数据安全提醒**\r\n> - Medical invoices and claimant data are sensitive personal information. Before using OCR features, obtain user consent, redact unnecessary PII, and prefer on-prem/private deployment.\r\n> - API keys MUST be stored in environment variables or a secret manager. Never hardcode keys.\r\n> - **中文：** 医疗发票和理赔申请人数据属于敏感个人信息。使用OCR功能前，须获得用户同意，脱敏非必要个人信息，优先使用本地私有化部署。API密钥必须存入环境变量或密钥管理器，禁止硬编码。\r\n\r\n---\r\n\r\n## ✨ What This Skill Provides / 本Skill提供的内容\r\n\r\n| Type / 类型 | Description / 说明 |\r\n|-------------|---------------------|\r\n| 📋 Workflow checklists / 流程检查清单 | Step-by-step claims review checklists for human reviewers |\r\n| 📄 Report templates / 报告模板 | Standardized output formats for claims analysis reports |\r\n| 🏗️ Reference architectures / 参考架构 | Guidance on OCR integration, rules engines, and GNN design |\r\n| 💡 Example code / 示例代码 | Python examples (require your own API keys and data) |\r\n| 📚 Company practices reference / 公司实践参考 | Summaries of industry best practices (advisory only) |\r\n\r\n**This skill does NOT provide:** executable models, pre-trained weights, bundled API credentials, or automated claim approval.\r\n\r\n---\r\n\r\n## Core Features / 核心功能（咨询级）\r\n\r\n### 1. Medical Receipt OCR — Integration Guidance / 医疗票据OCR（集成指导）\r\n\r\n**Supported document types / 支持票据类型（8类）：**\r\n门诊发票 / 住院发票 / 医疗费用明细清单 / 医保结算单 / 出院小结 / 病历首页 / 检查报告单 / 费用结算单\r\n\r\n**Integration options / 集成方案参考：**\r\n- Baidu AI OCR / 百度AI开放平台\r\n- Tencent Cloud OCR / 腾讯云OCR\r\n- Ali Cloud OCR / 阿里云OCR\r\n- Infologic OCR / 合合信息OCR\r\n\r\n> ⚠️ **Data Handling / 数据处理：** Only send necessary fields. Redact/unnecessary PII beforehand. Confirm vendor's data retention policy.\r\n> **中文：** 仅发送必要字段，事前脱敏非必要个人信息，确认服务商数据留存策略。\r\n\r\n### 2. Liability Determination — Advisory Checklist / 理赔判责（咨询检查清单）\r\n\r\nProvides structured checklists for human reviewers:\r\n- ✅ Waiting period check / 等待期检查\r\n- ✅ Pre-existing condition screen / 既往症筛查\r\n- ✅ Deductible verification / 免赔额校验\r\n- ✅ Hospital level verification / 就诊机构核查\r\n- ✅ Policy coverage match / 险种责任匹配\r\n\r\n> ⚠️ **All results are suggestions only. Final decisions MUST be made by authorized human reviewers.**\r\n> **中文：** 所有结果仅为建议，最终决定必须由授权人工审核员作出。\r\n\r\n### 3. Anti-Fraud Assessment — Checklist / 反欺诈评估（检查清单）\r\n\r\nStructured red-flag checklist for fraud investigation:\r\n- 🚩 Unusual visit frequency / 就诊频率异常\r\n- 🚩 Invoice authenticity verification / 票据真实性验证\r\n- 🚩 Diagnosis-medication mismatch / 诊断与用药不匹配\r\n- 🚩 Provider-case network anomalies / 医疗机构-案件网络异常\r\n\r\n**Data governance requirements / 数据治理要求：**\r\n- Retention limit / 留存期限：≤ 2 years (or per regulatory requirement) / 不超过2年（或监管要求期限）\r\n- Access control / 访问控制：Authorized fraud investigators only / 仅授权欺诈调查员可访问\r\n- Correction workflow / 更正流程：Data subjects have the right to request correction / 数据主体有权请求更正\r\n\r\n### 4. Claims Report Templates / 理赔报告模板\r\n\r\nStandardized templates for 7 insurance types:\r\n- Health insurance / 医疗险\r\n- Critical illness insurance / 重疾险\r\n- Life insurance / 寿险\r\n- Accident insurance / 意外险\r\n- Auto insurance / 车险\r\n- Property insurance / 财产险\r\n- Group insurance / 团险\r\n\r\nAll templates include the required disclaimer: \"This is an AI-assisted draft requiring licensed professional review.\"\r\n\r\n---\r\n\r\n## 🚀 Quick Start / 快速上手\r\n\r\n```bash\r\n# Install this skill (installs advisory templates only)\r\nnpx clawhub install insurance-claims-intelligence\r\n\r\n# Use in WorkBuddy (advisory mode only)\r\n/insurance-claims-intelligence \"Generate a claims review checklist for this outpatient case\"\r\n/insurance-claims-intelligence \"Create a fraud risk assessment template for these 5 cases\"\r\n```\r\n\r\n> ⚠️ **All outputs are drafts. Human review is mandatory.**\r\n> **中文：** 所有输出均为草稿，人工审核是强制要求。\r\n\r\n---\r\n\r\n## 📖 What's Included / 包含内容\r\n\r\n| File / 文件 | Content / 内容说明 |\r\n|-------------|---------------------|\r\n| `SKILL.md` | Full skill definition, trigger keywords, advisory workflows |\r\n| `references/claims_ocr_tech.md` | OCR integration guidance + provider comparison + example Python code (API key required) |\r\n| `references/claims_liability_engine.md` | Liability checklist + model reference + 3 company practices (advisory) |\r\n| `references/claims_report_templates.md` | Report templates + 7 insurance type notice templates |\r\n\r\n---\r\n\r\n## Provenance / 来源说明\r\n\r\n- **Author / 作者：** @gechengling\r\n- **Skill type / Skill类型：** Advisory templates and reference frameworks only / 仅含咨询模板和参考框架\r\n- **Contains executable code:** NO / 不含可执行代码\r\n- **Contains pre-trained models:** NO / 不含预训练模型\r\n- **Requires API credentials:** YES — you must provide your own OCR/LLM API keys / 需要您自行提供OCR/LLM API密钥\r\n- **License / 开源协议：** MIT-0\r\n\r\n---\r\n\r\n## Required Human Review / 强制人工审核要求\r\n\r\n| Action / 操作 | Human Review Required? / 需人工审核？ |\r\n|---------------|----------------------------------------|\r\n| Claim approval / 理赔核准 | ✅ MANDATORY / 强制 |\r\n| Claim denial / 理赔拒付 | ✅ MANDATORY / 强制 |\r\n| Payout amount decision / 赔付金额决定 | ✅ MANDATORY / 强制 |\r\n| Fraud label assignment / 欺诈标签标记 | ✅ MANDATORY / 强制 |\r\n| Customer-facing notice generation / 客户通知书生成 | ✅ MANDATORY / 强制 |\r\n\r\n---\r\n\r\n## Contact / 联系方式\r\n\r\nIf you have questions about this skill, contact: [@gechengling on ClawHub](https://clawhub.ai/gechengling)\r\n\r\n---\r\n\r\n*Last updated: 2026-05-05 — v1.2.0 — Added comprehensive disclaimers, removed auto-approval language, added data governance requirements.*\n\nFile v1.2.1:_meta.json\n\n{\n  \"ownerId\": \"kn74e704j3ygjcygnpf02rdvd185js13\",\n  \"slug\": \"insurance-claims-intelligence\",\n  \"version\": \"1.2.1\",\n  \"publishedAt\": 1782652486721\n}\n\nFile v1.2.1:references/claims_liability_engine.md\n\n# 保险理赔判责引擎参考框架（advisory only / 咨询级）\r\n\r\n> ⚠️ **DISCLAIMER / 免责声明**\r\n> - **English:** This document provides advisory frameworks, checklists, and reference code ONLY. It does NOT contain production-ready models, pre-trained weights, or automated claim decision capabilities. All accuracy figures (e.g., \"93% cases\", \"60 seconds\") are literature-reported benchmarks or design targets, NOT validated results of your deployment. ALL claim approvals, denials, and payout amounts MUST be reviewed and confirmed by a licensed insurance professional before use.\r\n> - **中文：** 本文档仅提供咨询框架、检查清单和参考代码，不含生产级模型、预训练权重或自动化理赔决策能力。所有准确率数据（如\"93%案件\"、\"60秒\"）均来自文献基准或设计目标，非您部署后的实测结果。所有理赔核准、拒付及赔付金额**必须经持证保险专业人士审核确认后方可使用**。\r\n\r\n> 🔒 **Human Review Mandatory / 强制人工审核**\r\n> - **NO automatic approval** — this framework only generates **decision-support suggestions**.\r\n> - **严禁自动审批** — 本框架仅生成**决策支持建议**，不具有任何自动审批能力。\r\n> - All outputs are **drafts** requiring licensed adjuster review and regulatory compliance check.\r\n> - 所有输出均为**草稿**，须经持证理赔师审核及监管合规检查。\r\n\r\n---\r\n\r\n## 一、判责引擎参考架构（人工审核框架）\r\n\r\n### 架构说明（所有输出需人工确认）\r\n\r\n```\r\n输入层：OCR结构化数据 + 保单信息 + 被保险人档案\r\n  ↓\r\n预处理层：数据清洗 → 字段标准化 → 缺失值处理\r\n  ↓\r\n规则引擎层（粗筛）：确定型规则快速分流\r\n  ├─ 等待期检查 → 建议拒付/继续（须人工确认）\r\n  ├─ 既往症检查 → 建议拒付/比例赔付/继续（须人工确认）\r\n  ├─ 免赔额检查 → 建议暂不赔付/继续（须人工确认）\r\n  ├─ 医院级别检查 → 提示确认/继续（须人工确认）\r\n  └─ 险种责任匹配 → 建议全额/比例/拒付（须人工确认）\r\n  ↓\r\nML推理层（精审）：不确定案件深度分析（参考框架）\r\n  ├─ NLP诊断解析 → ICD编码映射（须人工确认）\r\n  ├─ 治疗合理性分析 → DRG/DIP对照（须人工确认）\r\n  ├─ 费用异常检测 → 孤立森林/LOF（须人工确认）\r\n  └─ 欺诈风险评分 → 知识图谱参考（须人工确认）\r\n  ↓\r\n赔付计算层：条款公式 → 核定金额参考（须人工确认）\r\n  ↓\r\n输出层：理赔核定建议 / 拒付建议 / 补充材料建议\r\n  ↓\r\n⚠️ 【强制环节】持证理赔师人工审核 → 合规检查 → 最终决定\r\n```\r\n\r\n> ⚠️ **重要：** 以上架构中的任何\"自动\"、\"秒级\"、\"全自动\"描述均已删除。本框架仅提供**建议**，所有决定须由授权人工审核员作出。\r\n\r\n---\r\n\r\n## 二、规则引擎参考清单（需人工逐项确认）\r\n\r\n### 2.1 等待期检查（参考清单）\r\n\r\n| 险种 | 标准等待期 | 特殊约定 | 判责参考逻辑（须人工确认） |\r\n|------|-----------|---------|--------------------------|\r\n| 医疗险 | 30天 | 意外无等待期 | 出险日期 - 保单生效日 < 等待期 → 建议拒付（意外除外），**须人工确认** |\r\n| 重疾险 | 90天/180天 | 意外无等待期 | 确诊日期 - 保单生效日 < 等待期 → 建议拒付，**须人工确认** |\r\n| 寿险 | 90天/180天 | 意外无等待期 | 身故日期 - 保单生效日 < 等待期 → 建议拒付，**须人工确认** |\r\n| 意外险 | 无等待期 | 次日生效 | 直接通过（仍须人工确认） |\r\n\r\n```python\r\n# ⚠️ 参考代码（须自行测试、验证，并经人工审核后使用）\r\ndef check_waiting_period(policy: dict, claim: dict) -> dict:\r\n    \"\"\"等待期检查（参考实现，须经人工审核）\"\"\"\r\n    waiting_days = policy.get(\"waiting_days\", 30)\r\n    effect_date = parse_date(policy[\"effect_date\"])\r\n    incident_date = parse_date(claim[\"incident_date\"])\r\n    is_accident = claim.get(\"is_accident\", False)\r\n\r\n    if is_accident:\r\n        return {\"pass\": True, \"reason\": \"意外事故无等待期（建议，须人工确认）\"}\r\n\r\n    days_diff = (incident_date - effect_date).days\r\n    if days_diff < waiting_days:\r\n        return {\r\n            \"pass\": False,\r\n            \"reason\": f\"等待期内（生效{days_diff}天，等待期{waiting_days}天）\",\r\n            \"action_suggestion\": \"REJECT\",  # ⚠️ 仅为建议，非自动决定\r\n            \"human_review_required\": True\r\n        }\r\n    return {\"pass\": True, \"reason\": f\"等待期已过（生效{days_diff}天）\", \"human_review_required\": True}\r\n```\r\n\r\n### 2.2 既往症筛查（参考清单）\r\n\r\n**既往症定义（监管标准，仅供参考）：**\r\n1. 保险合同生效前，医生已有明确诊断、长期治疗未间断\r\n2. 保险合同生效前，医生已有明确诊断、治疗后症状未完全消失、有间断用药\r\n3. 保险合同生效前，医生已有明确诊断、但未予治疗\r\n4. 保险合同生效前，已有体检异常、但未确诊\r\n\r\n```python\r\n# ⚠️ 参考代码（须自行准备患者病史数据，并经人工审核）\r\ndef check_preexisting(claim_diagnosis: str, patient_history: list) -> dict:\r\n    \"\"\"既往症筛查（参考实现，须经人工审核）\"\"\"\r\n    risk_score = 0\r\n    matched_conditions = []\r\n\r\n    for condition in patient_history:\r\n        if fuzzy_match(claim_diagnosis, condition[\"diagnosis\"]):\r\n            risk_score += condition.get(\"severity\", 1)\r\n            matched_conditions.append(condition[\"diagnosis\"])\r\n\r\n    if risk_score >= 2:\r\n        return {\r\n            \"is_preexisting_suspected\": True,  # ⚠️ 仅为疑似，非确诊\r\n            \"matched\": matched_conditions,\r\n            \"action_suggestion\": \"REJECT\",\r\n            \"human_review_required\": True,\r\n            \"note\": \"须由理赔师结合病历进一步确认\"\r\n        }\r\n    return {\"is_preexisting_suspected\": False, \"human_review_required\": True}\r\n```\r\n\r\n### 2.3 免赔额校验（参考清单）\r\n\r\n```python\r\n# ⚠️ 参考代码（须经人工审核）\r\ndef check_deductible(claim_amount: float, policy: dict, ytd_paid: float) -> dict:\r\n    \"\"\"免赔额校验（参考实现，须经人工审核）\"\"\"\r\n    deductible = policy.get(\"deductible\", 0)\r\n    self_pay = claim_amount\r\n\r\n    accumulated = ytd_paid + self_pay\r\n    if accumulated <= deductible:\r\n        return {\r\n            \"pass\": False,\r\n            \"reason\": f\"未达免赔额（累计{accumulated:.2f}元，免赔{deductible:.2f}元）\",\r\n            \"action_suggestion\": \"DEFER\",\r\n            \"human_review_required\": True\r\n        }\r\n\r\n    payable = accumulated - deductible\r\n    return {\r\n        \"pass\": True,\r\n        \"payable_suggestion\": payable,  # ⚠️ 仅为建议金额\r\n        \"reason\": f\"已达免赔额，建议赔付{payable:.2f}元\",\r\n        \"human_review_required\": True\r\n    }\r\n```\r\n\r\n### 2.4 医院级别核查（参考清单）\r\n\r\n| 医院等级 | 医疗险 | 重疾险 | 寿险 | 意外险 |\r\n|---------|--------|--------|------|--------|\r\n| 三甲 | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） |\r\n| 三乙/三丙 | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） |\r\n| 二甲 | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） |\r\n| 二乙 | 视条款（须人工确认） | 视条款（须人工确认） | ✅ 建议通过（须人工确认） | 视条款（须人工确认） |\r\n| 一级/社区 | 视条款（须人工确认） | 视条款（须人工确认） | ✅ 建议通过（须人工确认） | 视条款（须人工确认） |\r\n| 私立医院 | 视条款（通常排除，须人工确认） | 视条款（须人工确认） | 视条款（须人工确认） | 视条款（须人工确认） |\r\n\r\n### 2.5 险种责任匹配（参考清单）\r\n\r\n```python\r\n# ⚠️ 参考代码（须经人工审核）\r\ndef match_coverage(diagnosis: str, policy_coverage: dict) -> dict:\r\n    \"\"\"责任范围匹配（参考实现，须经人工审核）\"\"\"\r\n    icd_code = map_to_icd10(diagnosis)  # 须自行实现并验证\r\n\r\n    if policy_coverage[\"type\"] == \"critical_illness\":\r\n        if icd_code in policy_coverage.get(\"covered_icd\", []):\r\n            return {\"match\": True, \"payout_type_suggestion\": \"FULL\", \"human_review_required\": True}\r\n        else:\r\n            return {\"match\": False, \"action_suggestion\": \"REJECT\", \"human_review_required\": True}\r\n\r\n    if policy_coverage[\"type\"] == \"medical\":\r\n        if is_medical_necessary(diagnosis) and in_medical_catalog(icd_code):\r\n            return {\"match\": True, \"payout_type_suggestion\": \"REIMBURSE\", \"human_review_required\": True}\r\n        else:\r\n            return {\"match\": False, \"action_suggestion\": \"REJECT\", \"human_review_required\": True}\r\n\r\n    if policy_coverage[\"type\"] == \"accident\":\r\n        if claim.get(\"accident_proof\"):\r\n            return {\"match\": True, \"payout_type_suggestion\": \"FULL_OR_PARTIAL\", \"human_review_required\": True}\r\n        else:\r\n            return {\"match\": False, \"action_suggestion\": \"REQUEST_DOC\", \"human_review_required\": True}\r\n```\r\n\r\n---\r\n\r\n## 三、机器学习精审参考框架（须人工确认所有输出）\r\n\r\n### 3.1 NLP诊断解析 + ICD编码映射（参考）\r\n\r\n```python\r\n# ⚠️ 参考代码（须自行准备训练数据和模型，输出须经人工确认）\r\nimport torch\r\nfrom transformers import BertTokenizer, BertModel\r\n\r\ndef diagnose_nlp_analysis(diagnosis_text: str) -> dict:\r\n    \"\"\"\r\n    NLP诊断解析参考实现\r\n    ⚠️ 须自行 fine-tune 模型，输出须经人工确认\r\n    \"\"\"\r\n    # ⚠️ 须自行准备训练好的模型和ICD-10数据库\r\n    # tokenizer = BertTokenizer.from_pretrained(\"your-fine-tuned-model\")\r\n    # model = BertModel.from_pretrained(\"your-fine-tuned-model\")\r\n    # icd_db = load_your_icd10_database()\r\n\r\n    raise NotImplementedError(\r\n        \"须自行准备fine-tuned模型和ICD-10数据库，并对所有输出进行人工审核\"\r\n    )\r\n```\r\n\r\n### 3.2 治疗合理性分析（DRG/DIP对照参考）\r\n\r\n```python\r\n# ⚠️ 参考代码（须经人工审核）\r\ndef check_treatment_reasonableness(diagnosis: str, treatments: list, total_fee: float) -> dict:\r\n    \"\"\"\r\n    治疗合理性分析参考实现\r\n    ⚠️ 须自行准备DRG标准数据库，输出须经人工确认\r\n    \"\"\"\r\n    # drg_group = map_to_drg(diagnosis)  # 须自行实现\r\n    # std_fee_range = drg_group[\"fee_range\"]\r\n\r\n    flags = []\r\n    # if total_fee > std_fee_range[1]:\r\n    #     flags.append(f\"总费用超出DRG标准上限\")\r\n    # if total_fee < std_fee_range[0] * 0.5:\r\n    #     flags.append(f\"总费用异常偏低\")\r\n\r\n    return {\r\n        \"drg_group_suggestion\": \"须自行实现\",\r\n        \"flags\": flags,\r\n        \"reasonableness_score_suggestion\": max(0, 100 - len(flags) * 20),\r\n        \"human_review_required\": True,\r\n        \"note\": \"所有标记须经人工调查确认，不可自动拒付\"\r\n    }\r\n```\r\n\r\n### 3.3 欺诈风险评分（参考框架，非自动标记）\r\n\r\n```python\r\n# ⚠️ 参考代码（所有风险评分须经人工调查确认，不可自动拒付）\r\ndef fraud_risk_scoring_reference(claim: dict, graph: \"nx.Graph | None\") -> dict:\r\n    \"\"\"\r\n    欺诈风险评分参考框架\r\n    ⚠️ 所有评分仅为调查优先级参考，不可作为自动拒付依据\r\n    \"\"\"\r\n    risk_score = 0\r\n    risk_factors = []\r\n\r\n    # 因子1：同一患者短期内多次理赔（须人工核实）\r\n    # if claim.get(\"recent_claim_count\", 0) >= 3:\r\n    #     risk_score += 30\r\n    #     risk_factors.append(\"同一患者短期内多次理赔，建议人工调查\")\r\n\r\n    # 因子2：多家保险公司同时索赔（须人工调查）\r\n    # if claim.get(\"multi_insurer\", False):\r\n    #     risk_score += 40\r\n    #     risk_factors.append(\"多家公司同时索赔，建议人工调查\")\r\n\r\n    # 所有评分仅为调查优先级参考\r\n    if risk_score >= 70:\r\n        action = \"PRIORITY_INVESTIGATE\"  # 优先调查（非自动拒付）\r\n    elif risk_score >= 40:\r\n        action = \"ENHANCED_REVIEW\"  # 加强审核\r\n    else:\r\n        action = \"ROUTINE_REVIEW\"  # 常规审核\r\n\r\n    return {\r\n        \"risk_score_suggestion\": risk_score,\r\n        \"risk_factors\": risk_factors,\r\n        \"suggested_action\": action,  # ⚠️ 仅为建议\r\n        \"human_review_required\": True,\r\n        \"note\": \"所有欺诈风险评估须经人工调查确认，不可作为自动拒付依据\"\r\n    }\r\n```\r\n\r\n### 3.4 赔付金额计算（参考公式，须经人工确认）\r\n\r\n```python\r\n# ⚠️ 参考代码（所有赔付金额须经人工确认）\r\ndef calculate_payout_reference(claim_data: dict, policy: dict) -> dict:\r\n    \"\"\"赔付金额计算参考公式（须经人工确认）\"\"\"\r\n    coverage_type = policy[\"coverage_type\"]\r\n\r\n    if coverage_type == \"reimbursement\":  # 报销型\r\n        total_fee = claim_data[\"total_fee\"]\r\n        medical_insurance_pay = claim_data.get(\"medical_insurance_pay\", 0)\r\n        personal_pay = total_fee - medical_insurance_pay\r\n        deductible = policy.get(\"deductible\", 0)\r\n        reimbursement_ratio = policy.get(\"reimbursement_ratio\", 1.0)\r\n        payable = max(0, (personal_pay - deductible)) * reimbursement_ratio\r\n        return {\r\n            \"payout_suggestion\": round(payable, 2),  # ⚠️ 仅为建议\r\n            \"formula\": f\"({personal_pay} - {deductible}) × {reimbursement_ratio}\",\r\n            \"human_review_required\": True\r\n        }\r\n\r\n    elif coverage_type == \"fixed\":  # 定额给付\r\n        return {\r\n            \"payout_suggestion\": policy[\"sum_insured\"],\r\n            \"formula\": \"保额全额给付（须人工确认条款条件）\",\r\n            \"human_review_required\": True\r\n        }\r\n\r\n    # ... 其他险种类似，所有输出均须人工确认\r\n```\r\n\r\n---\r\n\r\n## 四、行业实践参考（文献综述，非本Skill实测结果）\r\n\r\n> ⚠️ **重要：** 以下公司实践描述为公开文献报道的参考信息，非本Skill的实测性能。部署效果取决于您的数据、模型和配置。\r\n\r\n### 4.1 平安\"111极速赔\"（公开报道参考）\r\n\r\n```\r\n公开报道的技术方向参考：\r\n  用户上传材料（拍照/PDF）\r\n    ↓\r\n  大模型解析（材料理解）\r\n    ↓\r\n  规则引擎判责\r\n    ↓\r\n  大模型复核（边缘case处理）\r\n    ↓\r\n  结果：公开报道称93%案件60秒内完成（⚠️ 此为平安报道数据，非本Skill承诺）\r\n```\r\n\r\n**创新点（公开报道摘录）：**\r\n- 大模型理解非结构化医疗文本\r\n- 端到端处理流程优化\r\n\r\n### 4.2 中国人寿智能理赔（公开报道参考）\r\n\r\n```\r\n公开报道的理赔金额分层处理方式（⚠️ 仅供参考，须按您的合规要求配置）：\r\n  理赔金额 ≤ 20000元 → 系统建议 → 人工审核确认 → 到账\r\n  20000元 < 金额 ≤ 50000元 → AI建议 + 人工复核 → 当天到账\r\n  金额 > 50000元 → 完整调查流程 → 3个工作日内\r\n```\r\n\r\n### 4.3 太保\"数字劳动力实验室\"（公开报道参考）\r\n\r\n```\r\n公开报道的Agent工作流方向（⚠️ 仅供参考）：\r\n  [接收案件] → [OCR识别] → [条款解析] → [判责决策建议]\r\n    → IF 简单案件 → 建议 → 人工确认\r\n    → IF 复杂案件 → 生成调查清单建议 → 派单调查员（人工）\r\n    → IF 高风险标记（建议） → 转反欺诈团队（人工调查）\r\n```\r\n\r\n---\r\n\r\n## 五、判责决策参考树（所有路径须人工确认）\r\n\r\n```\r\n案件提交\r\n  │\r\n  ├─ 规则检查（参考清单）\r\n  │   ├─ 等待期未过 → 建议拒付（须人工确认）\r\n  │   ├─ 既往症命中 → 建议拒付/比例赔付（须人工确认）\r\n  │   ├─ 未达免赔额 → 建议暂不赔付（须人工确认）\r\n  │   ├─ 医院不合规 → 提示确认/建议拒付（须人工确认）\r\n  │   └─ 险种不匹配 → 建议拒付（须人工确认）\r\n  │\r\n  ├─ IF 规则检查全部通过（建议）：\r\n  │   ├─ 简单案件 → 生成赔付建议（⚠️ 须人工确认）\r\n  │   ├─ 复杂案件 → 生成复核建议（⚠️ 须人工确认）\r\n  │   ├─ 大额案件 → 生成调查建议（⚠️ 须人工确认）\r\n  │   └─ 高风险标记（建议） → 进入反欺诈调查流程（人工）\r\n  │\r\n  └─ ⚠️【强制环节】持证理赔师人工审核 → 合规检查 → 最终决定\r\n```\r\n\r\n---\r\n\r\n## 六、反欺诈数据治理规范（必读）\r\n\r\n> 🔒 **数据治理要求（合规必备）**\r\n\r\n### 6.1 数据留存期限\r\n- 反欺诈图谱数据建议留存不超过 **2 年**，除非监管要求更长留存期\r\n- 已结案件的数据须定期归档或匿名化处理\r\n\r\n### 6.2 访问控制\r\n- 反欺诈图谱查询权限仅开放给 **授权欺诈调查员**\r\n- 禁止非授权人员（如客服、销售）访问欺诈评分和图谱数据\r\n- 所有查询须留下审计日志（谁、何时、查询了哪个案件）\r\n\r\n### 6.3 数据更正流程\r\n- 被保险人和投诉方有权请求更正错误数据\r\n- 数据更正请求须在 **15 个工作日内** 处理完毕\r\n- 更正后须通知所有曾收到该错误数据的决策环节\r\n\r\n### 6.4 图谱污染防护\r\n- 新案件数据进入图谱前，须经 **人工审核确认** 数据质量\r\n- 禁止将未经验证的第三方提供数据直接写入生产图谱\r\n- 定期（建议每季度）对图谱数据进行质量审计\r\n\r\n---\r\n\r\n## 七、模型效果评估指标（文献参考值，非承诺）\r\n\r\n| 指标 | 文献参考值 | 说明 |\r\n|------|------------|------|\r\n| 自动通过率 | ~70% | 文献报道行业水平，非本Skill承诺 |\r\n| 误拒率（False Reject） | ≤ 2% | 文献报道行业水平，非本Skill承诺 |\r\n| 漏检率（Fraud Escape） | ≤ 1% | 文献报道行业水平，非本Skill承诺 |\r\n| 平均处理时长 | ≤ 1分钟（自动案件） | 取决于部署配置，非本Skill承诺 |\r\n| 客户满意度 | ≥ 85% | 文献报道行业水平，非本Skill承诺 |\r\n\r\n> ⚠️ **重要：** 以上指标均为文献报道的行业参考值，非本Skill的性能承诺。实际效果取决于您的数据质量、模型训练、规则配置和人工审核流程。\r\n\r\n---\r\n\r\n*Last updated: 2026-05-05 — 删除所有自动审批描述；准确率/性能数据标注为文献基准（非本Skill实测）；所有输出增加\"须人工确认\"标注；增加反欺诈数据治理规范；增加强制人工审核环节说明。*\n\nFile v1.2.1:references/claims_ocr_tech.md\n\n# 保险理赔多模态医疗票据 OCR 技术参考（ advisory only / 咨询级）\r\n\r\n> ⚠️ **DISCLAIMER / 免责声明**\r\n> - **English:** This document provides technical reference and example code ONLY. It does NOT provide production-ready models, pre-trained weights, or bundled API credentials. All accuracy figures (e.g., \"~95%\") are vendor-published benchmarks under ideal conditions, NOT validated results of your deployment. You must provide your own API keys, training data, and models. All OCR results must be reviewed by human staff before use in claim decisions.\r\n> - **中文：** 本文档仅提供技术参考和示例代码，不含生产级模型、预训练权重或绑定的API凭证。所有准确率数据（如\"~95%\"）均为厂商发布的理想条件下基准结果，非您部署后的实测结果。您必须自行提供API密钥、训练数据和模型。所有OCR结果在用于理赔决定前，必须经人工审核。\r\n\r\n> 🔒 **DATA SECURITY / 数据安全**\r\n> - Medical invoices contain sensitive personal information (name, ID number, diagnosis, hospital). Before sending to ANY cloud OCR provider, you MUST: (1) obtain user consent, (2) redact unnecessary PII fields, (3) confirm the vendor's data retention policy (prefer: no storage / auto-delete within 24h), (4) consider private on-prem deployment for production use.\r\n> - API keys and credentials MUST be stored in environment variables or a secret manager. NEVER hardcode keys in production code.\r\n\r\n---\r\n\r\n## 一、主流 OCR 服务横向对比（厂商公开基准数据）\r\n\r\n| 服务商 | 支持票据类型 | 厂商公开准确率基准 | 调用限频 | 参考价格（元/千次） | 特色功能 |\r\n|---------|--------------|----------------|----------|----------------|----------|\r\n| 百度 AI | 发票+7类病历+2类报告 | ~95%（厂商基准） | 50次/秒 | 0.15 | 覆盖最全，EasyDL 自训练 |\r\n| 腾讯云 | 全国门诊/住院发票 | ~93%（厂商基准） | 5次/秒 | 0.12 | 与腾讯云生态打通，理赔场景优化 |\r\n| 阿里云 + 深智恒际 | 各省市门诊发票 | ~92%（厂商基准） | 20次/秒 | 0.18 | 支持定制化训练，票据类型覆盖深度好 |\r\n| 合合信息 TextIn | 医疗票据全品类 | ~94%（厂商基准） | 100次/秒 | 0.20 | 深度学习融合，表格还原精度高 |\r\n\r\n> ⚠️ **重要：** 以上准确率为厂商在理想条件下发布的基准数据，实际准确率取决于图像质量、票据类型、训练数据质量等因素。部署前须自行测试验证。\r\n\r\n---\r\n\r\n## 二、技术架构参考（需自行实现）\r\n\r\n### 完整处理流水线（参考架构）\r\n\r\n```\r\n原始图像输入\r\n  ↓\r\n[图像预处理]  — 需自行实现或调用云服务API\r\n  - 去噪（高斯滤波 / 中值滤波）\r\n  - 倾斜校正（霍夫直线检测 + 仿射变换）\r\n  - 二值化（Otsu 自适应阈值）\r\n  - 透视校正（四点变换）\r\n  ↓\r\n[版面分析]  — 需自行实现\r\n  - 关键点检测（角点 / 印章 / 表格线）\r\n  - 区域分割（标题 / 明细 / 印章区）\r\n  ↓\r\n[文本检测]  — 可选：CTPN / PSENet / DB-Net（需自行训练）\r\n  - 文字行定位\r\n  - 弯曲文本矫正\r\n  ↓\r\n[文本识别]  — 可选：CRNN + CTC / ViT-OCR（需自行训练）\r\n  - CNN 特征提取（ResNet50 / EfficientNet-B4）\r\n  - RNN 序列建模（BiLSTM × 2 层）\r\n  - Attention 对齐（Coverage Attention）\r\n  - CTC 解码 → 字符序列\r\n  ↓\r\n[结构化提取]  — 需自行实现规则 + NER 模型\r\n  - 正则表达式匹配（金额 / 日期 / 发票号）\r\n  - 命名实体识别 BIOES 标注\r\n  - 字段归一化（金额统一到 `float`，日期统一到 `YYYY-MM-DD`）\r\n  ↓\r\n[输出]  JSON 结构化结果（需人工审核）\r\n```\r\n\r\n> ⚠️ **重要：** 以上架构需要您自行准备训练数据、训练模型或购买云服务API。本Skill不提供任何预训练模型权重。\r\n\r\n---\r\n\r\n## 三、Python 调用示例（需自行配置API密钥）\r\n\r\n### ⚠️ 安全提醒（使用任何OCR服务前必读）\r\n\r\n1. **API Key 管理：** 将密钥存入环境变量，禁止硬编码\r\n2. **数据脱敏：** 发送图像前，用遮盖方式隐去姓名、身份证号等敏感字段\r\n3. **数据留存：** 确认服务商数据处理政策，优先选择\"不留存\"或\"24小时内自动删除\"\r\n4. **生产环境：** 建议使用私有化部署方案，医疗数据不出域\r\n\r\n### 3.1 百度 AI 医疗票据识别（示例）\r\n\r\n```python\r\nimport requests\r\nimport base64\r\nimport json\r\nimport os\r\n\r\n# ⚠️ 安全做法：从环境变量读取密钥，禁止硬编码\r\nAPI_KEY = os.environ.get(\"BAIDU_OCR_API_KEY\")\r\nSECRET_KEY = os.environ.get(\"BAIDU_OCR_SECRET_KEY\")\r\n\r\nif not API_KEY or not SECRET_KEY:\r\n    raise ValueError(\"请设置环境变量 BAIDU_OCR_API_KEY 和 BAIDU_OCR_SECRET_KEY\")\r\n\r\n# 获取 access_token\r\ndef get_access_token():\r\n    url = f\"https://aip.baidu.com/oauth/2.0/token?grant_type=client_credentials&client_id={API_KEY}&client_secret={SECRET_KEY}\"\r\n    return requests.get(url).json()[\"access_token\"]\r\n\r\n# 医疗票据识别（⚠️ 建议：事前对图像做脱敏处理）\r\ndef recognize_medical_invoice(image_path: str, token: str) -> dict:\r\n    with open(image_path, \"rb\") as f:\r\n        img_b64 = base64.b64encode(f.read()).decode(\"utf-8\")\r\n\r\n    url = f\"https://aip.baidu.com/rest/2.0/ocr/v1/medical_invoice?access_token={token}\"\r\n    payload = {\r\n        \"image\": img_b64,\r\n        \"detect_direction\": \"true\",\r\n        \"probability\": \"true\",\r\n    }\r\n    headers = {\"Content-Type\": \"application/x-www-form-urlencoded\"}\r\n    resp = requests.post(url, data=payload, headers=headers)\r\n    return resp.json()\r\n\r\n# 使用示例（⚠️ 结果须经人工审核）\r\n# token = get_access_token()\r\n# result = recognize_medical_invoice(\"invoice.jpg\", token)\r\n# print(json.dumps(result, ensure_ascii=False, indent=2))\r\n```\r\n\r\n### 3.2 腾讯云医疗发票识别（示例）\r\n\r\n```python\r\n# ⚠️ 安全做法：从环境变量读取密钥\r\nimport os\r\nfrom tencentcloud.common import credential\r\nfrom tencentcloud.common.profile.client_profile import ClientProfile\r\nfrom tencentcloud.common.profile.http_profile import HttpProfile\r\nfrom tencentcloud.ocr.v20181119 import ocr_client, models\r\n\r\ndef recognize_tencent_medical(image_path: str):\r\n    cred = credential.Credential(\r\n        os.environ.get(\"TENCENTCLOUD_SECRET_ID\"),\r\n        os.environ.get(\"TENCENTCLOUD_SECRET_KEY\")\r\n    )\r\n    http_profile = HttpProfile()\r\n    http_profile.endpoint = \"ocr.tencentcloudapi.com\"\r\n    client_profile = ClientProfile()\r\n    client_profile.httpProfile = http_profile\r\n    client = ocr_client.OcrClient(cred, \"ap-guangzhou\", client_profile)\r\n\r\n    with open(image_path, \"rb\") as f:\r\n        img_b64 = base64.b64encode(f.read()).decode(\"utf-8\")\r\n\r\n    req = models.MedicalInvoiceOCRRequest()\r\n    req.ImageBase64 = img_b64\r\n    resp = client.MedicalInvoiceOCR(req)\r\n    return resp.to_json_string(indent=2)\r\n\r\n# print(recognize_tencent_medical(\"invoice.jpg\"))\r\n```\r\n\r\n### 3.3 本地自训练 OCR（PyTorch + CRNN）— 参考代码\r\n\r\n```python\r\n# ⚠️ 注意：以下为参考架构代码，需要您自行准备训练数据和标注\r\nimport torch\r\nimport torch.nn as nn\r\nimport torchvision.models as models\r\nfrom torch.nn.utils.rnn import pack_padded_sequence\r\n\r\nclass CRNN(nn.Module):\r\n    \"\"\"\r\n    CRNN 参考模型架构：\r\n    CNN（特征提取）→ BiLSTM（序列建模）→ CTC（解码）\r\n    需要自行准备训练数据和训练脚本。\r\n    \"\"\"\r\n    def __init__(self, num_chars: int, hidden_size: int = 256):\r\n        super().__init__()\r\n        # CNN 主干：EfficientNet-B0（需自行预训练或下载权重）\r\n        backbone = models.efficientnet_b0(weights=None)  # 需自行提供权重\r\n        self.cnn = nn.Sequential(*list(backbone.children())[:-2])\r\n        self.cnn.add_module(\"adaptive_pool\", nn.AdaptiveAvgPool2d((None, 1)))\r\n\r\n        # BiLSTM × 2\r\n        self.lstm = nn.LSTM(\r\n            input_size=1280,\r\n            hidden_size=hidden_size,\r\n            num_layers=2,\r\n            bidirectional=True,\r\n            batch_first=True\r\n        )\r\n        self.fc = nn.Linear(hidden_size * 2, num_chars)\r\n\r\n    def forward(self, x):\r\n        conv = self.cnn(x)\r\n        conv = conv.squeeze(2)\r\n        conv = conv.permute(0, 2, 1)\r\n        lstm_out, _ = self.lstm(conv)\r\n        logits = self.fc(lstm_out)\r\n        return logits\r\n\r\n# 训练需自行准备：\r\n# - 训练数据（标注好的医疗票据图像）\r\n# - CTC Loss 配置\r\n# - 训练脚本\r\n# model = CRNN(num_chars=6624)\r\n# ctc_loss = nn.CTCLoss(zero_infinity=True)\r\n# optimizer = torch.optim.AdamW(model.parameters(), lr=1e-4)\r\n```\r\n\r\n---\r\n\r\n## 四、票据结构化字段规范（参考格式）\r\n\r\n### 4.1 全国医疗门诊发票标准字段（参考）\r\n\r\n```yaml\r\nbasic_info:\r\n  invoice_code: str        # 票据代码\r\n  invoice_number: str      # 票据号码\r\n  invoice_date: date      # 开票日期\r\n  verify_code: str        # 校验码\r\n  hospital_name: str     # 医院名称\r\n  hospital_level: str    # 医院等级（三级/二级/一级）\r\n\r\namount_info:\r\n  total: float          # 总金额（含税）\r\n  medical_insurance: float  # 医保统筹支付\r\n  personal_account: float  # 个人账户支付\r\n  self_pay: float       # 个人自付（分类自负）\r\n  self_fund: float      # 个人自费\r\n  deductible: float     # 起付线（免赔额部分）\r\n\r\ndetail_line_items:        # 费用明细行（数组）\r\n  - name: str           # 药品名称/诊疗项目名称\r\n    specification: str   # 规格\r\n    unit: str           # 单位\r\n    quantity: int       # 数量\r\n    unit_price: float   # 单价\r\n    total_price: float  # 该项总价\r\n    category: str       # 甲类/乙类/丙类/自费\r\n    insurance_code: str  # 医保目录编码\r\n\r\ndiagnosis:\r\n  icd10_code: str      # ICD-10 诊断编码\r\n  diagnosis_name: str  # 诊断名称（中文）\r\n  department: str      # 就诊科室\r\n```\r\n\r\n### 4.2 全国医疗住院发票附加字段（参考）\r\n\r\n```yaml\r\nhospitalization:\r\n  admission_date: date  # 入院日期\r\n  discharge_date: date # 出院日期\r\n  stay_days: int       # 住院天数\r\n  bed_number: str     # 床位号\r\n  total_prescriptions: int  # 总处方数\r\n  surgery_name: str   # 手术名称（如有）\r\n  drg_code: str       # DRG 分组编码（如有）\r\n```\r\n\r\n---\r\n\r\n## 五、多模态融合识别方案（兜底方案参考）\r\n\r\n当单一 OCR 效果不佳时（如盖章覆盖文字、票据折叠），可启用多模态大模型作为兜底：\r\n\r\n```python\r\n# ⚠️ 注意：多模态大模型也会接收图像数据，须同样遵守数据安全规定\r\ndef multimodal_fallback(image_path: str, ocr_result: dict) -> dict:\r\n    \"\"\"\r\n    OCR 识别置信度低于阈值时，可触发多模态识别作为参考。\r\n    ⚠️ 结果须经人工审核确认。\r\n    \"\"\"\r\n    import base64, os\r\n\r\n    # ⚠️ 建议：发送前对敏感字段做模糊化处理\r\n    with open(image_path, \"rb\") as f:\r\n        img_b64 = base64.b64encode(f.read()).decode()\r\n\r\n    prompt = \"\"\"\r\n    这是一张医疗发票照片，OCR 识别结果置信度较低。\r\n    请你根据图片内容，提取以下字段（JSON 格式）：\r\n    - 发票号码\r\n    - 开票日期\r\n    - 医院名称\r\n    - 总金额\r\n    - 医保统筹支付\r\n    - 个人自付\r\n    - 临床诊断\r\n    如果图片中看不清楚，则对应字段填 null。\r\n    \"\"\"\r\n\r\n    # 调用多模态 API（须自行配置密钥并遵守数据安全规定）\r\n    # response = call_multimodal_api(\r\n    #     model=\"qwen-vl-max\",\r\n    #     messages=[{\r\n    #         \"role\": \"user\",\r\n    #         \"content\": [\r\n    #             {\"image\": f\"data:image/jpeg;base64,{img_b64}\"},\r\n    #             {\"text\": prompt}\r\n    #         ]\r\n    #     }]\r\n    # )\r\n    # return parse_json_from_response(response)\r\n    raise NotImplementedError(\"须自行实现多模态API调用，并配置数据安全保护\")\r\n```\r\n\r\n**触发多模态兜底的条件（参考）：**\r\n- OCR 置信度均值 < 0.75\r\n- 关键字段缺失 ≥ 2 个（发票号 / 金额 / 日期）\r\n- 印章严重遮挡文字区域（通过图像分割检测到大面积红色区域）\r\n\r\n---\r\n\r\n## 六、部署建议（参考）\r\n\r\n| 部署方式 | 适用场景 | 推荐技术栈 | 数据合规建议 |\r\n|-----------|-----------|-----------|--------------|\r\n| **云端 API 调用** | 中小规模（< 10万张/月） | 百度 AI / 腾讯云 SDK | 确认服务商数据留存政策；建议签署数据处理协议 |\r\n| **私有化部署** | 大型保险公司 / 数据不出域 | PyTorchServe + CRNN 自训练模型 | 数据全程不出域，符合《个人信息保护法》要求 |\r\n| **混合架构** | 高可用要求 + 数据合规 | 云端 OCR 预处理 + 本地规则引擎核验 | 云端仅传输必要字段，敏感字段本地处理 |\r\n| **边缘端部署** | 移动端 / 小程序拍照即识别 | NCNN + Int8 量化 CRNN 模型 | 数据在设备端处理，不上传云端 |\r\n\r\n---\r\n\r\n## 七、数据留存与访问控制（必读）\r\n\r\n> ⚠️ **法律合规要求（中国《个人信息保护法》）**\r\n\r\n1. **数据最小化原则：** 仅收集和传输理赔处理所必需的最少字段\r\n2. **用户同意：** 将医疗票据发送至OCR服务前，须获得用户明确同意\r\n3. **数据留存期限：** 建议 OCR 服务商不留存数据，或设置 24 小时内自动删除；本地图谱数据留存不超过 2 年\r\n4. **访问控制：** 医疗票据图像和识别结果仅限授权理赔人员访问，禁止非授权人员查看\r\n5. **数据更正权利：** 被保人有权请求更正错误数据，必须在 15 个工作日内处理\r\n6. **跨境数据传输：** 如使用境外OCR服务，须进行数据出境安全评估\r\n\r\n---\r\n\r\n*Last updated: 2026-05-05 — 添加数据安全声明、API密钥安全提醒、准确率数据标注为厂商基准（非实测）、删除自动审批描述、增加数据留存和访问控制规范。*\n\nFile v1.2.1:references/claims_report_templates.md\n\n# 保险理赔报告模板与输出规范\r\n\r\n> 标准化的理赔分析报告模板，适用于各险种理赔场景。\r\n> 包含：报告结构、快捷输出指令、七项检查清单、FAQ模板。\r\n\r\n---\r\n\r\n## 一、标准理赔分析报告（通用模板）\r\n\r\n```markdown\r\n# 保险理赔智能分析报告\r\n\r\n**生成时间**: {{YYYY-MM-DD HH:mm}}\r\n**案件编号**: CL-{{YYYYMMDD}}-{{XXXX}}\r\n**保单号**: {{policy_no}}\r\n**被保险人**: {{name}}（{{gender}}，{{age}}岁）\r\n**险种类别**: {{coverage_type}}\r\n**理赔类型**: {{claim_type}}（门诊/住院/重疾/身故/意外/车险）\r\n**处理状态**: {{AUTO_PASS | MANUAL_REVIEW | HIGH_RISK | REJECT}}\r\n\r\n---\r\n\r\n## 一、票据识别结果\r\n\r\n### 1.1 医疗票据OCR识别\r\n\r\n| 票据类型 | 票据代码/号 | 开票日期 | 金额（元） | 识别状态 |\r\n|---------|------------|---------|-----------|---------|\r\n| {{ticket_type}} | {{ticket_code}} | {{date}} | ¥{{amount}} | {{status}} |\r\n\r\n### 1.2 关键字段提取\r\n\r\n| 字段 | 提取结果 | 置信度 |\r\n|------|---------|---------|\r\n| 医院名称 | {{hospital}} | {{confidence}} |\r\n| 诊断信息 | {{diagnosis}}（ICD-10: {{icd10}}） | {{confidence}} |\r\n| 总费用 | ¥{{total}} | {{confidence}} |\r\n| 医保支付 | ¥{{medical_pay}} | {{confidence}} |\r\n| 个人自付 | ¥{{self_pay}} | {{confidence}} |\r\n| 住院天数 | {{days}}天 | {{confidence}} |\r\n\r\n---\r\n\r\n## 二、责任认定结果\r\n\r\n| 审核维度 | 判定结果 | 说明 |\r\n|---------|---------|------|\r\n| 等待期 | {{PASS/FAIL}} | {{reason}} |\r\n| 既往症 | {{PASS/FAIL}} | {{reason}} |\r\n| 免赔额 | {{PASS/FAIL}} | 本年度累计¥{{accumulated}}，免赔额¥{{deductible}} |\r\n| 医院级别 | {{COMPLIANT/NON_COMPLIANT}} | {{hospital_level}} |\r\n| 保障范围 | {{COVERED/NOT_COVERED}} | {{reason}} |\r\n| 事故性质 | {{CONFIRMED/PENDING}} | {{reason}} |\r\n\r\n**综合判责结论**：{{CONCLUSION}}\r\n\r\n---\r\n\r\n## 三、赔付计算\r\n\r\n### 3.1 计算公式\r\n\r\n```\r\n可赔付金额 = （个人自付 - 免赔额）× 赔付比例\r\n           = （¥{{self_pay}} - ¥{{deductible}}）× {{ratio}}%\r\n           = ¥{{payable}}\r\n```\r\n\r\n### 3.2 赔付明细\r\n\r\n| 项目 | 金额（元） | 说明 |\r\n|------|-----------|------|\r\n| 发票总金额 | ¥{{total}} | OCR识别 |\r\n| 医保统筹支付 | ¥{{medical_pay}} | 医保基金支付部分 |\r\n| 个人账户支付 | ¥{{account_pay}} | 医保个人账户 |\r\n| **个人自付** | **¥{{self_pay}}** | 分类自负+自负 |\r\n| - 免赔额 | ¥{{deductible}} | 年免赔额 |\r\n| × 赔付比例 | {{ratio}}% | 条款约定 |\r\n| **核定赔付金额** | **¥{{payable}}** | 最终赔付 |\r\n\r\n---\r\n\r\n## 四、反欺诈风险评估\r\n\r\n| 风险维度 | 评分（0-100） | 风险等级 | 说明 |\r\n|---------|--------------|---------|------|\r\n| 票据真实性 | {{score}} | {{LOW/MEDIUM/HIGH}} | {{reason}} |\r\n| 行为异常 | {{score}} | {{level}} | {{reason}} |\r\n| 关系图谱 | {{score}} | {{level}} | {{reason}} |\r\n| 医疗合理性 | {{score}} | {{level}} | {{reason}} |\r\n| **综合风险** | **{{total_score}}** | **{{FINAL_LEVEL}}** | {{action}} |\r\n\r\n---\r\n\r\n## 五、最终结论\r\n\r\n- **理赔核定**：{{FULL_PAY | PARTIAL_PAY | REJECT}}\r\n- **赔付金额**：¥{{final_pay}}（大写：{{capitalized}}）\r\n- **到账时间**：{{days}}个工作日内\r\n- **所需材料**：{{status}}（已收齐 / 需补充：{{list}}）\r\n- **案件状态**：{{status}}\r\n\r\n---\r\n\r\n## 六、补充材料通知（如需要）\r\n\r\n请补充以下材料：\r\n1. {{doc1}}\r\n2. {{doc2}}\r\n3. {{doc3}}\r\n\r\n---\r\n\r\n> ⚠️ 本报告由 AI 理赔智能系统自动生成，核定金额仅供参考，\r\n> 最终以保险公司理赔部门审核结果为准。\r\n> 如有异议，请在收到通知书之日起60日内提出复议申请。\r\n```\r\n\r\n---\r\n\r\n## 二、分险种报告模板\r\n\r\n### 2.1 医疗险理赔报告（简版）\r\n\r\n```markdown\r\n# 医疗险理赔核定通知书\r\n\r\n**被保险人**：{{name}}\r\n**保单号**：{{policy_no}}\r\n**理赔金额**：¥{{payable}}\r\n\r\n## 费用明细\r\n- 门诊/住院总费用：¥{{total}}\r\n- 医保统筹支付：¥{{medical_pay}}\r\n- 个人自付：¥{{self_pay}}\r\n- 减：年免赔额：¥{{deductible}}\r\n- 乘：赔付比例：{{ratio}}%\r\n- **应付赔款：¥{{payable}}**\r\n\r\n## 赔付说明\r\n本次理赔已通过智能审核，赔款将于3个工作日内转入您指定的银行账户。\r\n\r\n---\r\n案件编号：{{case_id}}  |  客服热线：955XX\r\n```\r\n\r\n### 2.2 重疾险理赔报告\r\n\r\n```markdown\r\n# 重大疾病保险金给付通知书\r\n\r\n**被保险人**：{{name}}\r\n**保单号**：{{policy_no}}\r\n**重大疾病**：{{diagnosis}}（ICD-10：{{icd10}}）\r\n**给付金额**：¥{{sum_insured}}（大写：{{capitalized}}）\r\n\r\n## 核定依据\r\n- 病理报告已确认疾病符合条款约定\r\n- 等待期已满（{{days}}天）\r\n- 无责任免除情形\r\n\r\n## 给付说明\r\n重大疾病保险金已全额给付，本合同继续有效（如含身故责任）。\r\n\r\n---\r\n案件编号：{{case_id}}  |  给付日期：{{date}}\r\n```\r\n\r\n### 2.3 车险理赔计算书\r\n\r\n```markdown\r\n# 车辆损失险理赔计算书\r\n\r\n**被保险人**：{{name}}\r\n**保单号**：{{policy_no}}\r\n**事故编号**：{{accident_id}}\r\n**事故责任**：{{responsibility}}（全责/主责/同责/次责/无责）\r\n\r\n## 定损明细\r\n| 项目 | 金额（元） |\r\n|------|-----------|\r\n| 配件费用 | ¥{{parts}} |\r\n| 工时费用 | ¥{{labor}} |\r\n| 施救费用 | ¥{{rescue}} |\r\n| 减：残值扣减 | ¥{{salvage}} |\r\n| **核定损失** | **¥{{assessed_loss}}** |\r\n\r\n## 赔付计算\r\n- 核定损失：¥{{assessed_loss}}\r\n- 事故责任比例：{{resp_ratio}}%\r\n- 减：免赔额：¥{{deductible}}\r\n- **应付赔款：¥{{payable}}**\r\n\r\n---\r\n定损员：{{adjuster}}  |  审核日期：{{date}}\r\n```\r\n\r\n---\r\n\r\n## 三、快捷输出指令\r\n\r\n| 指令 | 输入示例 | 输出内容 |\r\n|------|---------|---------|\r\n| `/理赔` | `/理赔 门诊发票213.5元 诊断上呼吸道感染` | 完整理赔分析报告 |\r\n| `/识别` | `/识别 上传发票图片` | OCR结构化结果 |\r\n| `/判责` | `/判责 保单P123 诊断J06.9` | 责任认定结果 |\r\n| `/计算` | `/计算 发票5000元 免赔额1000元 比例80%` | 赔付金额 |\r\n| `/拒付` | `/拒付 案件CL20260504001` | 拒付通知书 |\r\n| `/反欺诈` | `/反欺诈 患者张三 近期5次理赔` | 欺诈风险报告 |\r\n| `/车险` | `/车险 事故认定书同责 定损2960元` | 车险理赔计算书 |\r\n| `/重疾` | `/重疾 病理报告肺腺癌T2N0M0` | 重疾险给付通知书 |\r\n\r\n---\r\n\r\n## 四、七项检查清单\r\n\r\n每次输出理赔报告前，必须检查以下7项：\r\n\r\n```markdown\r\n## ✅ 七项检查清单\r\n\r\n- [ ] 1. 被保险人身份信息是否完整（姓名/身份证/保单号）\r\n- [ ] 2. 票据识别关键字段是否齐全（金额/日期/医院/诊断）\r\n- [ ] 3. 责任免除条款是否已核查（等待期/既往症/医院级别）\r\n- [ ] 4. 赔付计算公式是否正确（免赔额/赔付比例/给付限额）\r\n- [ ] 5. 反欺诈风险评估是否已执行（综合风险评分）\r\n- [ ] 6. 金额大小写是否一致（阿拉伯数字与中文大写）\r\n- [ ] 7. 法律合规条款是否已标注（异议期/客服电话/监管依据）\r\n```\r\n\r\n---\r\n\r\n## 五、拒付通知书模板\r\n\r\n```markdown\r\n# 理赔拒付通知书\r\n\r\n**案件编号**：{{case_id}}\r\n**被保险人**：{{name}}\r\n**保单号**：{{policy_no}}\r\n**拒付日期**：{{date}}\r\n\r\n## 拒付原因\r\n\r\n经审核，您的理赔申请不符合保险条款约定，具体原因：\r\n\r\n**{{reason_code}}**：{{reason_detail}}\r\n\r\n{{#if 等待期未过}}\r\n等待期条款约定：本合同生效后{{waiting_days}}天为等待期，等待期内确诊的疾病，本公司不承担保险责任。\r\n您的确诊日期距离保单生效日仅{{actual_days}}天，在等待期内。\r\n{{/if}}\r\n\r\n{{#if 既往症}}\r\n根据条款\"责任免除\"第{{clause_no}}条，被保险人在本合同生效前已患有的疾病或症状，属于责任免除范围。\r\n{{/if}}\r\n\r\n{{#if 非保障范围}}\r\n本次就诊诊断（{{diagnosis}}）不属于本合同约定的保障范围。\r\n本产品保障范围详见条款第{{clause_no}}条。\r\n{{/if}}\r\n\r\n## 您的权利\r\n\r\n如您对本决定有异议，可在收到本通知书之日起60日内，向本公司申请复议，或向{{regulator}}投诉，或依法提起诉讼。\r\n\r\n---\r\n客服热线：{{hotline}}  |  投诉邮箱：{{email}}\r\n```\r\n\r\n---\r\n\r\n## 六、补充材料通知书模板\r\n\r\n```markdown\r\n# 补充理赔材料通知书\r\n\r\n**案件编号**：{{case_id}}\r\n**被保险人**：{{name}}\r\n**当前状态**：材料不齐，待补充\r\n\r\n## 需补充材料清单\r\n\r\n请您在{{deadline}}前补充以下材料，以便我们继续处理您的理赔申请：\r\n\r\n| 序号 | 材料名称 | 要求 | 备注 |\r\n|------|---------|------|------|\r\n| 1 | {{doc1}} | {{requirement}} | {{note}} |\r\n| 2 | {{doc2}} | {{requirement}} | {{note}} |\r\n| 3 | {{doc3}} | {{requirement}} | {{note}} |\r\n\r\n## 补充方式\r\n\r\n1. **线上上传**：登录{{app_name}} APP → 理赔服务 → 案件查询 → 补充材料\r\n2. **邮件发送**：将材料照片发送至 {{email}}\r\n3. **线下递交**：前往本公司客户服务中心（地址：{{address}}）\r\n\r\n---\r\n客服热线：{{hotline}}  |  案件处理员：{{handler}}\r\n```\r\n\r\n---\r\n\r\n## 七、FAQ回复模板\r\n\r\n### Q1：理赔需要多长时间？\r\n\r\n> 根据《保险法》第23条规定，保险人收到理赔申请后，应当及时作出核定；情形复杂的，应当在30日内作出核定。\r\n> 本公司承诺：简单案件（3000元以下）当日完成；普通案件（3000-20000元）3个工作日内完成；复杂案件30日内完成。\r\n\r\n### Q2：为什么我的理赔被拒付了？\r\n\r\n> 拒付原因通常包括：①等待期内出险；②既往症；③非保障范围；④医院不符合要求；⑤免赔额未达。\r\n> 详细拒付原因请查看《理赔拒付通知书》，如有异议可申请复议。\r\n\r\n### Q3：理赔款什么时候到账？\r\n\r\n> 理赔款在理赔决定作出后10日内支付（法律规定）。本公司实际到账时间：自动理赔案件实时到账；人工审核案件3个工作日内到账。\r\n\r\n### Q4：我对理赔决定不满意怎么办？\r\n\r\n> 您有以下途径：\r\n> 1. 向本公司申请复议（60日内）\r\n> 2. 向国家金融监督管理总局投诉\r\n> 3. 依法提起诉讼\r\n\r\n### Q5：医疗发票OCR识别不准确怎么办？\r\n\r\n> 您可以提供更清晰的发票照片重新识别，或选择人工审核通道。OCR识别结果仅供参考，最终以人工审核为准。\r\n\r\n---\r\n\r\n## 八、输出格式规范总结\r\n\r\n| 格式类型 | 适用场景 | 输出方式 |\r\n|---------|---------|---------|\r\n| 完整报告 | 综合理赔分析 | Markdown 表格 + 分段 |\r\n| 核定通知书 | 赔付决定 | 正式文书格式（含印章位置） |\r\n| 拒付通知书 | 拒付决定 | 正式文书格式 + 法律条款引用 |\r\n| 补充材料通知 | 材料不齐 | 清单格式 + 补充方式说明 |\r\n| 欺诈风险报告 | 高风险案件 | 评分表格 + 风险因子列表 |\r\n| 计算书 | 车险/财产险 | 定损明细表格 + 计算公式 |\r\n\r\n---\r\n\r\n> 📌 **重要声明**：所有输出模板均为参考格式，实际使用时须根据各保险公司具体条款和监管要求调整。AI生成内容须经持牌理赔师审核确认后，方可作为正式理赔文书使用。\n\nFile v1.2.1:skill-card.md\n\n## Description: <br>\nInsurance Claims Intelligence provides advisory templates, checklists, and decision-support frameworks for medical OCR, liability review, anti-fraud assessment, and claims reporting. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[gechengling](https://clawhub.ai/user/gechengling) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nInsurance claims teams, reviewers, and developers use this skill to draft claim review checklists, OCR integration guidance, fraud-risk assessment templates, and report formats. All outputs are decision-support drafts and require licensed human review before any real claim action. <br>\n\n### Deployment Geography for Use: <br>\nGlobal, with China-focused insurance workflow and regulatory references. <br>\n\n## Known Risks and Mitigations: <br>\nRisk: Draft outputs could be mistaken for final claim approvals, denials, payout amounts, fraud labels, or customer-facing notices. <br>\nMitigation: Use the skill only for advisory drafts and require licensed insurance reviewer sign-off before any claim decision, payment, fraud label, or notice is issued. <br>\nRisk: Medical invoices, diagnosis records, and claimant details may expose sensitive personal data when used with OCR or multimodal examples. <br>\nMitigation: Obtain consent, redact unnecessary personal data, prefer private or on-prem OCR for production, and confirm vendor retention and transfer terms before processing claim data. <br>\nRisk: Example integrations require external API credentials and user-provided data. <br>\nMitigation: Store credentials in environment variables or a secret manager, never hardcode keys, and test any example code with approved data before operational use. <br>\n\n\n## Reference(s): <br>\n- [ClawHub skill page](https://clawhub.ai/gechengling/skills/insurance-claims-intelligence) <br>\n- [Claims OCR technology reference](references/claims_ocr_tech.md) <br>\n- [Claims liability engine reference](references/claims_liability_engine.md) <br>\n- [Claims report templates](references/claims_report_templates.md) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [text, markdown, code, shell commands, configuration, guidance] <br>\n**Output Format:** [Markdown with tables, checklists, report templates, JSON-style examples, and Python snippets] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [Outputs are advisory drafts and must include human review before claim decisions or customer-facing use.] <br>\n\n## Skill Version(s): <br>\n1.2.1 (source: server release and SKILL.md frontmatter) <br>\n\n## Ethical Considerations: <br>\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment. <br>\n\nArchive v5.0.3: 7 files, 29191 bytes\n\nFiles: README.md (7594b), references/claims_liability_engine.md (18889b), references/claims_ocr_tech.md (14039b), references/claims_report_templates.md (11395b), skill-card.md (2176b), SKILL.md (12571b), _meta.json (148b)\n\nFile v5.0.3:SKILL.md\n\n---\r\nname: Insurance Claims Intelligence Expert\r\ndescription: Advisory skill for insurance claims processing workflows — provides templates, checklists, and decision-support frameworks for medical OCR, liability determination, anti-fraud assessment, and claims reporting. Human review required for all claim decisions. Keywords: insurance claims, claims advisory, medical OCR, anti-fraud, insurance tech, China insurance, decision support, 智能理赔, 理赔风控, 医疗单据识别, 责任认定, 理赔报告, 秒赔, 理赔决策, 医疗险理赔, 重疾理赔, 车险理赔.\r\nslug: insurance-claims-intelligence\r\nversion: 1.2.0\r\n\ncapabilities:\n  - educational-reference\n  - advisory-only\n  - requires-human-review\n  - no-executable-code\n---\r\n\r\n# Insurance Claims Intelligence Expert / 保险行业智能理赔专家\n\n> **⚠️ SECURITY NOTICE / 安全声明**\n> - **Type:** Educational reference / analytical framework ONLY\n> - **No executable code, scripts, or binaries are included in this skill**\n> - **No persistent storage, network calls, background execution, or credential collection**\n> - **All outputs are for reference only and require human review before real-world application**\n> - **This skill does NOT provide financial, legal, or insurance advice**\n> - **Users must exercise their own judgment and consult qualified professionals**\n\n\r\n\r\n> **⚠️ DISCLAIMER / 免责声明**\r\n> - **English:** This skill provides advisory templates, checklists, and decision-support frameworks ONLY. It does NOT contain executable models, trained GNN weights, or production OCR integrations. All accuracy figures (e.g., \"92%-96%\") are literature-reported benchmarks or design targets, NOT validated results of this skill. ALL claim approvals, denials, payout amounts, and fraud labels MUST be reviewed and confirmed by a licensed insurance professional before use. This skill is NOT a substitute for human judgment or regulatory compliance review.\r\n> - **中文：** 本Skill仅提供咨询模板、检查清单和决策支持框架，不含可执行模型、已训练GNN权重或生产级OCR集成。所有准确率数据（如\"92%-96%\"）均来自文献基准或设计目标，非本Skill实测结果。所有理赔核准、拒付、赔付金额及欺诈标签，**必须经持证保险专业人士审核确认后方可使用**。本Skill不可替代人工判断或监管合规审查。\r\n\r\n> **🔒 DATA SECURITY / 数据安全**\r\n> - Medical invoices, diagnosis records, and claimant data are sensitive personal information under China's Personal Information Protection Law (PIPL). Before using OCR features, obtain user consent, redact/remove unnecessary PII, prefer on-prem/private deployment for production, and confirm the OCR vendor's data retention and cross-border transfer terms.\r\n> - API keys and credentials MUST be stored in environment variables or a secret manager. Never hardcode keys in production systems.\r\n> - **English:** This skill provides advisory templates, checklists, and decision-support frameworks ONLY. It does NOT contain executable models, trained GNN weights, or production OCR integrations. All accuracy figures (e.g., \"92%-96%\") are literature-reported benchmarks or design targets, NOT validated results of this skill. ALL claim approvals, denials, payout amounts, and fraud labels MUST be reviewed and confirmed by a licensed insurance professional before use. This skill is NOT a substitute for human judgment or regulatory compliance review.\r\n> - **中文：** 本Skill仅提供咨询模板、检查清单和决策支持框架，不含可执行模型、已训练GNN权重或生产级OCR集成。所有准确率数据（如\"92%-96%\"）均来自文献基准或设计目标，非本Skill实测结果。所有理赔核准、拒付、赔付金额及欺诈标签，**必须经持证保险专业人士审核确认后方可使用**。本Skill不可替代人工判断或监管合规审查。\r\n\r\n---\r\n\r\n## Artifact Type / 作品类型\r\n\r\n**This is a documentation-and-template skill.** It contains:\r\n- ✅ Workflow checklists and decision trees\r\n- ✅ Report templates and output formats\r\n- ✅ Reference architectures and integration guidance\r\n- ✅ Example Python code (requires your own API keys and data)\r\n\r\nIt does NOT contain:\r\n- ❌ Pre-trained ML/GNN models\r\n- ❌ Executable OCR or claims processing code\r\n- ❌ Bundled third-party API credentials\r\n\r\n---\r\n\r\n## Trigger Keywords / 触发关键词\r\n\r\n**English Triggers:** insurance claims advisory, claims workflow, claim analysis, medical OCR guidance, insurance fraud assessment, claim liability review, policy clause analysis, anti-fraud checklist, insurance tech reference, claims report template\r\n\r\n**中文触发词：** 保险理赔咨询 / 理赔流程指导 / 理赔分析 / 医疗发票识别指导 / 理赔判责参考 / 责任认定流程 / 医疗险理赔 / 重疾险理赔 / 寿险理赔 / 意外险理赔 / 车险理赔 / 财产险理赔 / 理赔反欺诈 / 欺诈检测参考 / 骗保识别指导 / 理赔风控参考 / 保险条款解读 / 责任免除说明 / 保障范围分析 / 赔付比例计算 / 产品对比参考 / 条款比对指导 / 合同解读参考\r\n\r\n---\r\n\r\n## Core Capabilities / 核心能力（咨询框架）\r\n\r\n### 1. Medical Receipt OCR — Guidance Framework / 医疗票据OCR识别（指导框架）\r\n\r\n**支持的票据类型（覆盖全场景）：**\r\n\r\n| Receipt Type / 票据类型 | Extracted Fields / 识别内容 | Insurance Types / 适用险种 |\r\n|------------------------|-------------------|------------------|\r\n| 全国统一门诊发票 | 发票号、医院、金额、明细项目 | 医疗险、意外险 |\r\n| 全国统一住院发票 | 入院/出院日期、总金额、自费比例 | 医疗险、重疾险 |\r\n| 医疗费用明细清单 | 药品明细、检查项目、单价、数量 | 医疗险 |\r\n| 医保结算单 | 医保账户支付、自付金额、报销比例 | 医疗险 |\r\n| 出院小结 | 诊断、住院天数、治疗经过 | 重疾险、寿险 |\r\n| 病历首页 | 主要诊断、手术名称、ICD编码 | 重疾险 |\r\n| 检查报告单 | 影像报告、检验结果 | 重疾险 |\r\n| 费用结算单 | 分项金额、总计金额 | 财产险、责任险 |\r\n\r\n> **⚠️ OCR Data Handling / OCR数据处理提醒**\r\n> - Only send necessary fields to OCR providers; redact/unnecessary PII beforehand.\r\n> - Confirm the OCR vendor's data retention policy (Prefer: no storage / auto-delete within 24h).\r\n> - For production use, prefer private on-prem OCR deployment to avoid third-party data transfer.\r\n> - **中文：** 仅发送必要字段至OCR服务商；事前脱敏/删除非必要个人信息；确认OCR厂商数据留存策略（优先：不留存/24小时内自动删除）；生产环境优先使用私有化本地部署OCR，避免第三方数据传输。\r\n\r\n**参考技术架构（需自行集成）：**\r\n\r\n```text\r\n原始图像\r\n  ↓\r\n图像预处理（去噪/倾斜校正/二值化）\r\n  ↓\r\nCNN特征提取（ResNet50/EfficientNet）—— 需自行训练或调用云服务API\r\n  ↓\r\nRNN序列建模（BiLSTM）+ Attention机制\r\n  ↓\r\nCRF层解码 → 结构化文本输出\r\n  ↓\r\n字段标准化 → JSON/表格结构化结果\r\n```\r\n\r\n### 2. Liability Determination — Advisory Framework / 理赔判责引擎（咨询框架）\r\n\r\n**咨询级判责检查清单（需人工逐项确认）：**\r\n\r\n```text\r\n规则1：等待期检查（人工确认）\r\n  └─ 出险日期 - 保单生效日 < 等待期 → 建议拒付，需人工复核\r\n\r\n规则2：既往症筛查（人工确认）\r\n  └─ 既往症库匹配 → 责任免除 → 建议拒付/比例赔付，需人工复核\r\n\r\n规则3：免赔额校验（人工确认）\r\n  └─ 累计自付金额 < 免赔额 → 建议暂不赔付，需人工复核\r\n\r\n规则4：就诊机构核查（人工确认）\r\n  └─ 非二级及以上公立医院（需视条款）→ 提示确认，需人工复核\r\n\r\n规则5：险种责任匹配（人工确认）\r\n  └─ 就诊科室/诊断是否符合条款保障范围 → 建议全额/比例/拒付，需人工复核\r\n```\r\n\r\n> **⚠️ IMPORTANT / 重要提醒**\r\n> The liability determination output is a **decision-support suggestion ONLY**. Final approval/denial MUST be made by an authorized human reviewer. This skill does NOT auto-approve any claim amount.\r\n> **中文：** 判责输出**仅为决策支持建议**，最终核准/拒付**必须由授权人工审核员作出**。本Skill不对任何理赔金额进行自动审批。\r\n\r\n### 3. Anti-Fraud Assessment — Advisory Framework / 反欺诈评估（咨询框架）\r\n\r\n**反欺诈检查清单（咨询级）：**\r\n\r\n```text\r\n检查项1：就诊频率异常\r\n  └─ 同一被保人短期内多次就诊 → 标记，建议人工调查\r\n\r\n检查项2：票据真实性验证\r\n  └─ 发票号重复 / 医院不存在 / 金额异常 → 标记，建议人工调查\r\n\r\n检查项3：诊断与用药匹配性\r\n  └─ 诊断与开具药品明显不符 → 标记，建议人工调查\r\n\r\n检查项4：关系网络异常\r\n  └─ 同一医生/医院集中出现在多起理赔 → 标记，建议人工调查\r\n```\r\n\r\n> **🔒 Anti-Fraud Data Governance / 反欺诈数据治理**\r\n> - Retention limit / 留存期限：反欺诈图谱数据建议留存不超过 2 年，除非监管要求的更长留存期。\r\n> - Access control / 访问控制：图谱查询权限仅开放给授权欺诈调查员，禁止非授权人员访问。\r\n> - Data correction workflow / 数据更正流程：被保人有权请求更正错误数据，必须在 15 个工作日内处理。\r\n> - Poisoning safeguard / 污染防护：新案件数据进入图谱前，须经人工审核确认，防止恶意污染。\r\n\r\n### 4. Claims Report Templates / 理赔报告模板\r\n\r\n```markdown\r\n# 理赔分析报告（咨询草稿）\r\n**生成时间**: YYYY-MM-DD HH:mm\r\n**案件编号**: CL-XXXXXXXX\r\n**险种类别**: [险种名称]\r\n**处理状态**: [咨询草稿 — 需人工审核]\r\n**免责声明**: 本报告为AI辅助生成的咨询草稿，所有结论须经持证理赔师审核确认后方可生效。\r\n---\r\n## 一、票据识别结果（仅供参考）\r\n## 二、责任认定分析（仅供参考）\r\n## 三、赔付计算参考（仅供参考）\r\n## 四、反欺诈风险评估（仅供参考）\r\n## 五、建议下一步行动（需人工确认）\r\n```\r\n\r\n---\r\n\r\n## Compliance & Human Review / 合规与人工审核要求\r\n\r\n| Compliance Item / 合规项 | Regulatory Basis / 监管依据 | Human Review Requirement / 人工审核要求 |\r\n|--------------------|--------------------|----------------------|\r\n| 理赔时效 | 《保险法》第23条 | 核定结果须经人工确认后发出 |\r\n| 材料完整性 | 理赔管理办法 | 缺失材料列表由人工最终确认 |\r\n| 反欺诈合规 | 《反保险欺诈工作办法》2024 | 欺诈标记须经人工调查确认 |\r\n| 数据安全 | 《个人信息保护法》 | 医疗数据脱敏处理须经人工检查 |\r\n| 资金安全 | 反洗钱规定 | 大额理赔须人工复核 + 主管审批 |\r\n\r\n**ALL outputs of this skill are drafts requiring licensed professional review. / 本Skill所有输出均为草稿，须经持证专业人士审核。**\r\n\r\n---\r\n\r\n## Output Format / 输出格式规范\r\n\r\nAll outputs must include the following disclaimer:\r\n\r\n```markdown\r\n> ⚠️ **免责声明 / Disclaimer**\r\n> 本输出为AI辅助咨询草稿，所有理赔决定、拒付结论、赔付金额及欺诈标签\r\n> 须经【持证保险理赔师】审核确认后方可生效。\r\n> This is an AI-assisted draft. All claim decisions must be reviewed by a\r\n> licensed insurance adjuster before taking effect.\r\n```\r\n\r\n---\r\n\r\n## References / 参考文件\r\n\r\n| File / 文件 | Content / 内容说明 |\r\n|------|---------|\r\n| `references/claims_ocr_tech.md` | OCR技术架构参考 + 4家服务商对比 + Python示例代码（需自行配置API Key） |\r\n| `references/claims_liability_engine.md` | 判责规则参考 + 机器学习模型参考 + 3家公司实践参考 |\r\n| `references/claims_report_templates.md` | 报告模板 + 7种险种通知书模板参考 |\r\n\r\n> **⚠️ Reference files contain example code only. You must:**\r\n> - Provide your own API keys and store them in environment variables\r\n> - Provide your own training data and models\r\n> - Ensure human review of all outputs before use\r\n> - **中文：** 参考文件仅含示例代码，您必须：自行提供API密钥并存入环境变量；自行准备训练数据和模型；确保所有输出经人工审核后方可使用。\n\nFile v5.0.3:README.md\n\n# Insurance Claims Intelligence Expert / 保险行业智能理赔专家\r\n\r\n> **⚠️ DISCLAIMER / 免责声明**\r\n> - **English:** This is an **advisory and template-only skill**. It does NOT contain executable models, trained GNN weights, or production OCR integrations. All accuracy figures (e.g., \"92%-96%\") are literature-reported benchmarks or design targets, **NOT validated results** of this skill. ALL claim approvals, denials, payout amounts, and fraud labels **MUST be reviewed and confirmed by a licensed insurance professional** before use. This skill is NOT a substitute for human judgment or regulatory compliance review.\r\n> - **中文：** 本Skill**仅为咨询模板和参考框架**，不含可执行模型、已训练GNN权重或生产级OCR集成。所有准确率数据（如\"92%-96%\"）均来自文献基准或设计目标，**非本Skill实测结果**。所有理赔核准、拒付、赔付金额及欺诈标签，**必须经持证保险专业人士审核确认后方可使用**。本Skill不可替代人工判断或监管合规审查。\r\n\r\n> **🔒 DATA SECURITY NOTICE / 数据安全提醒**\r\n> - Medical invoices and claimant data are sensitive personal information. Before using OCR features, obtain user consent, redact unnecessary PII, and prefer on-prem/private deployment.\r\n> - API keys MUST be stored in environment variables or a secret manager. Never hardcode keys.\r\n> - **中文：** 医疗发票和理赔申请人数据属于敏感个人信息。使用OCR功能前，须获得用户同意，脱敏非必要个人信息，优先使用本地私有化部署。API密钥必须存入环境变量或密钥管理器，禁止硬编码。\r\n\r\n---\r\n\r\n## ✨ What This Skill Provides / 本Skill提供的内容\r\n\r\n| Type / 类型 | Description / 说明 |\r\n|-------------|---------------------|\r\n| 📋 Workflow checklists / 流程检查清单 | Step-by-step claims review checklists for human reviewers |\r\n| 📄 Report templates / 报告模板 | Standardized output formats for claims analysis reports |\r\n| 🏗️ Reference architectures / 参考架构 | Guidance on OCR integration, rules engines, and GNN design |\r\n| 💡 Example code / 示例代码 | Python examples (require your own API keys and data) |\r\n| 📚 Company practices reference / 公司实践参考 | Summaries of industry best practices (advisory only) |\r\n\r\n**This skill does NOT provide:** executable models, pre-trained weights, bundled API credentials, or automated claim approval.\r\n\r\n---\r\n\r\n## Core Features / 核心功能（咨询级）\r\n\r\n### 1. Medical Receipt OCR — Integration Guidance / 医疗票据OCR（集成指导）\r\n\r\n**Supported document types / 支持票据类型（8类）：**\r\n门诊发票 / 住院发票 / 医疗费用明细清单 / 医保结算单 / 出院小结 / 病历首页 / 检查报告单 / 费用结算单\r\n\r\n**Integration options / 集成方案参考：**\r\n- Baidu AI OCR / 百度AI开放平台\r\n- Tencent Cloud OCR / 腾讯云OCR\r\n- Ali Cloud OCR / 阿里云OCR\r\n- Infologic OCR / 合合信息OCR\r\n\r\n> ⚠️ **Data Handling / 数据处理：** Only send necessary fields. Redact/unnecessary PII beforehand. Confirm vendor's data retention policy.\r\n> **中文：** 仅发送必要字段，事前脱敏非必要个人信息，确认服务商数据留存策略。\r\n\r\n### 2. Liability Determination — Advisory Checklist / 理赔判责（咨询检查清单）\r\n\r\nProvides structured checklists for human reviewers:\r\n- ✅ Waiting period check / 等待期检查\r\n- ✅ Pre-existing condition screen / 既往症筛查\r\n- ✅ Deductible verification / 免赔额校验\r\n- ✅ Hospital level verification / 就诊机构核查\r\n- ✅ Policy coverage match / 险种责任匹配\r\n\r\n> ⚠️ **All results are suggestions only. Final decisions MUST be made by authorized human reviewers.**\r\n> **中文：** 所有结果仅为建议，最终决定必须由授权人工审核员作出。\r\n\r\n### 3. Anti-Fraud Assessment — Checklist / 反欺诈评估（检查清单）\r\n\r\nStructured red-flag checklist for fraud investigation:\r\n- 🚩 Unusual visit frequency / 就诊频率异常\r\n- 🚩 Invoice authenticity verification / 票据真实性验证\r\n- 🚩 Diagnosis-medication mismatch / 诊断与用药不匹配\r\n- 🚩 Provider-case network anomalies / 医疗机构-案件网络异常\r\n\r\n**Data governance requirements / 数据治理要求：**\r\n- Retention limit / 留存期限：≤ 2 years (or per regulatory requirement) / 不超过2年（或监管要求期限）\r\n- Access control / 访问控制：Authorized fraud investigators only / 仅授权欺诈调查员可访问\r\n- Correction workflow / 更正流程：Data subjects have the right to request correction / 数据主体有权请求更正\r\n\r\n### 4. Claims Report Templates / 理赔报告模板\r\n\r\nStandardized templates for 7 insurance types:\r\n- Health insurance / 医疗险\r\n- Critical illness insurance / 重疾险\r\n- Life insurance / 寿险\r\n- Accident insurance / 意外险\r\n- Auto insurance / 车险\r\n- Property insurance / 财产险\r\n- Group insurance / 团险\r\n\r\nAll templates include the required disclaimer: \"This is an AI-assisted draft requiring licensed professional review.\"\r\n\r\n---\r\n\r\n## 🚀 Quick Start / 快速上手\r\n\r\n```bash\r\n# Install this skill (installs advisory templates only)\r\nnpx clawhub install insurance-claims-intelligence\r\n\r\n# Use in WorkBuddy (advisory mode only)\r\n/insurance-claims-intelligence \"Generate a claims review checklist for this outpatient case\"\r\n/insurance-claims-intelligence \"Create a fraud risk assessment template for these 5 cases\"\r\n```\r\n\r\n> ⚠️ **All outputs are drafts. Human review is mandatory.**\r\n> **中文：** 所有输出均为草稿，人工审核是强制要求。\r\n\r\n---\r\n\r\n## 📖 What's Included / 包含内容\r\n\r\n| File / 文件 | Content / 内容说明 |\r\n|-------------|---------------------|\r\n| `SKILL.md` | Full skill definition, trigger keywords, advisory workflows |\r\n| `references/claims_ocr_tech.md` | OCR integration guidance + provider comparison + example Python code (API key required) |\r\n| `references/claims_liability_engine.md` | Liability checklist + model reference + 3 company practices (advisory) |\r\n| `references/claims_report_templates.md` | Report templates + 7 insurance type notice templates |\r\n\r\n---\r\n\r\n## Provenance / 来源说明\r\n\r\n- **Author / 作者：** @gechengling\r\n- **Skill type / Skill类型：** Advisory templates and reference frameworks only / 仅含咨询模板和参考框架\r\n- **Contains executable code:** NO / 不含可执行代码\r\n- **Contains pre-trained models:** NO / 不含预训练模型\r\n- **Requires API credentials:** YES — you must provide your own OCR/LLM API keys / 需要您自行提供OCR/LLM API密钥\r\n- **License / 开源协议：** MIT-0\r\n\r\n---\r\n\r\n## Required Human Review / 强制人工审核要求\r\n\r\n| Action / 操作 | Human Review Required? / 需人工审核？ |\r\n|---------------|----------------------------------------|\r\n| Claim approval / 理赔核准 | ✅ MANDATORY / 强制 |\r\n| Claim denial / 理赔拒付 | ✅ MANDATORY / 强制 |\r\n| Payout amount decision / 赔付金额决定 | ✅ MANDATORY / 强制 |\r\n| Fraud label assignment / 欺诈标签标记 | ✅ MANDATORY / 强制 |\r\n| Customer-facing notice generation / 客户通知书生成 | ✅ MANDATORY / 强制 |\r\n\r\n---\r\n\r\n## Contact / 联系方式\r\n\r\nIf you have questions about this skill, contact: [@gechengling on ClawHub](https://clawhub.ai/gechengling)\r\n\r\n---\r\n\r\n*Last updated: 2026-05-05 — v1.2.0 — Added comprehensive disclaimers, removed auto-approval language, added data governance requirements.*\n\nFile v5.0.3:_meta.json\n\n{\n  \"ownerId\": \"kn74e704j3ygjcygnpf02rdvd185js13\",\n  \"slug\": \"insurance-claims-intelligence\",\n  \"version\": \"5.0.3\",\n  \"publishedAt\": 1780353327210\n}\n\nFile v5.0.3:references/claims_liability_engine.md\n\n# 保险理赔判责引擎参考框架（advisory only / 咨询级）\r\n\r\n> ⚠️ **DISCLAIMER / 免责声明**\r\n> - **English:** This document provides advisory frameworks, checklists, and reference code ONLY. It does NOT contain production-ready models, pre-trained weights, or automated claim decision capabilities. All accuracy figures (e.g., \"93% cases\", \"60 seconds\") are literature-reported benchmarks or design targets, NOT validated results of your deployment. ALL claim approvals, denials, and payout amounts MUST be reviewed and confirmed by a licensed insurance professional before use.\r\n> - **中文：** 本文档仅提供咨询框架、检查清单和参考代码，不含生产级模型、预训练权重或自动化理赔决策能力。所有准确率数据（如\"93%案件\"、\"60秒\"）均来自文献基准或设计目标，非您部署后的实测结果。所有理赔核准、拒付及赔付金额**必须经持证保险专业人士审核确认后方可使用**。\r\n\r\n> 🔒 **Human Review Mandatory / 强制人工审核**\r\n> - **NO automatic approval** — this framework only generates **decision-support suggestions**.\r\n> - **严禁自动审批** — 本框架仅生成**决策支持建议**，不具有任何自动审批能力。\r\n> - All outputs are **drafts** requiring licensed adjuster review and regulatory compliance check.\r\n> - 所有输出均为**草稿**，须经持证理赔师审核及监管合规检查。\r\n\r\n---\r\n\r\n## 一、判责引擎参考架构（人工审核框架）\r\n\r\n### 架构说明（所有输出需人工确认）\r\n\r\n```\r\n输入层：OCR结构化数据 + 保单信息 + 被保险人档案\r\n  ↓\r\n预处理层：数据清洗 → 字段标准化 → 缺失值处理\r\n  ↓\r\n规则引擎层（粗筛）：确定型规则快速分流\r\n  ├─ 等待期检查 → 建议拒付/继续（须人工确认）\r\n  ├─ 既往症检查 → 建议拒付/比例赔付/继续（须人工确认）\r\n  ├─ 免赔额检查 → 建议暂不赔付/继续（须人工确认）\r\n  ├─ 医院级别检查 → 提示确认/继续（须人工确认）\r\n  └─ 险种责任匹配 → 建议全额/比例/拒付（须人工确认）\r\n  ↓\r\nML推理层（精审）：不确定案件深度分析（参考框架）\r\n  ├─ NLP诊断解析 → ICD编码映射（须人工确认）\r\n  ├─ 治疗合理性分析 → DRG/DIP对照（须人工确认）\r\n  ├─ 费用异常检测 → 孤立森林/LOF（须人工确认）\r\n  └─ 欺诈风险评分 → 知识图谱参考（须人工确认）\r\n  ↓\r\n赔付计算层：条款公式 → 核定金额参考（须人工确认）\r\n  ↓\r\n输出层：理赔核定建议 / 拒付建议 / 补充材料建议\r\n  ↓\r\n⚠️ 【强制环节】持证理赔师人工审核 → 合规检查 → 最终决定\r\n```\r\n\r\n> ⚠️ **重要：** 以上架构中的任何\"自动\"、\"秒级\"、\"全自动\"描述均已删除。本框架仅提供**建议**，所有决定须由授权人工审核员作出。\r\n\r\n---\r\n\r\n## 二、规则引擎参考清单（需人工逐项确认）\r\n\r\n### 2.1 等待期检查（参考清单）\r\n\r\n| 险种 | 标准等待期 | 特殊约定 | 判责参考逻辑（须人工确认） |\r\n|------|-----------|---------|--------------------------|\r\n| 医疗险 | 30天 | 意外无等待期 | 出险日期 - 保单生效日 < 等待期 → 建议拒付（意外除外），**须人工确认** |\r\n| 重疾险 | 90天/180天 | 意外无等待期 | 确诊日期 - 保单生效日 < 等待期 → 建议拒付，**须人工确认** |\r\n| 寿险 | 90天/180天 | 意外无等待期 | 身故日期 - 保单生效日 < 等待期 → 建议拒付，**须人工确认** |\r\n| 意外险 | 无等待期 | 次日生效 | 直接通过（仍须人工确认） |\r\n\r\n```python\r\n# ⚠️ 参考代码（须自行测试、验证，并经人工审核后使用）\r\ndef check_waiting_period(policy: dict, claim: dict) -> dict:\r\n    \"\"\"等待期检查（参考实现，须经人工审核）\"\"\"\r\n    waiting_days = policy.get(\"waiting_days\", 30)\r\n    effect_date = parse_date(policy[\"effect_date\"])\r\n    incident_date = parse_date(claim[\"incident_date\"])\r\n    is_accident = claim.get(\"is_accident\", False)\r\n\r\n    if is_accident:\r\n        return {\"pass\": True, \"reason\": \"意外事故无等待期（建议，须人工确认）\"}\r\n\r\n    days_diff = (incident_date - effect_date).days\r\n    if days_diff < waiting_days:\r\n        return {\r\n            \"pass\": False,\r\n            \"reason\": f\"等待期内（生效{days_diff}天，等待期{waiting_days}天）\",\r\n            \"action_suggestion\": \"REJECT\",  # ⚠️ 仅为建议，非自动决定\r\n            \"human_review_required\": True\r\n        }\r\n    return {\"pass\": True, \"reason\": f\"等待期已过（生效{days_diff}天）\", \"human_review_required\": True}\r\n```\r\n\r\n### 2.2 既往症筛查（参考清单）\r\n\r\n**既往症定义（监管标准，仅供参考）：**\r\n1. 保险合同生效前，医生已有明确诊断、长期治疗未间断\r\n2. 保险合同生效前，医生已有明确诊断、治疗后症状未完全消失、有间断用药\r\n3. 保险合同生效前，医生已有明确诊断、但未予治疗\r\n4. 保险合同生效前，已有体检异常、但未确诊\r\n\r\n```python\r\n# ⚠️ 参考代码（须自行准备患者病史数据，并经人工审核）\r\ndef check_preexisting(claim_diagnosis: str, patient_history: list) -> dict:\r\n    \"\"\"既往症筛查（参考实现，须经人工审核）\"\"\"\r\n    risk_score = 0\r\n    matched_conditions = []\r\n\r\n    for condition in patient_history:\r\n        if fuzzy_match(claim_diagnosis, condition[\"diagnosis\"]):\r\n            risk_score += condition.get(\"severity\", 1)\r\n            matched_conditions.append(condition[\"diagnosis\"])\r\n\r\n    if risk_score >= 2:\r\n        return {\r\n            \"is_preexisting_suspected\": True,  # ⚠️ 仅为疑似，非确诊\r\n            \"matched\": matched_conditions,\r\n            \"action_suggestion\": \"REJECT\",\r\n            \"human_review_required\": True,\r\n            \"note\": \"须由理赔师结合病历进一步确认\"\r\n        }\r\n    return {\"is_preexisting_suspected\": False, \"human_review_required\": True}\r\n```\r\n\r\n### 2.3 免赔额校验（参考清单）\r\n\r\n```python\r\n# ⚠️ 参考代码（须经人工审核）\r\ndef check_deductible(claim_amount: float, policy: dict, ytd_paid: float) -> dict:\r\n    \"\"\"免赔额校验（参考实现，须经人工审核）\"\"\"\r\n    deductible = policy.get(\"deductible\", 0)\r\n    self_pay = claim_amount\r\n\r\n    accumulated = ytd_paid + self_pay\r\n    if accumulated <= deductible:\r\n        return {\r\n            \"pass\": False,\r\n            \"reason\": f\"未达免赔额（累计{accumulated:.2f}元，免赔{deductible:.2f}元）\",\r\n            \"action_suggestion\": \"DEFER\",\r\n            \"human_review_required\": True\r\n        }\r\n\r\n    payable = accumulated - deductible\r\n    return {\r\n        \"pass\": True,\r\n        \"payable_suggestion\": payable,  # ⚠️ 仅为建议金额\r\n        \"reason\": f\"已达免赔额，建议赔付{payable:.2f}元\",\r\n        \"human_review_required\": True\r\n    }\r\n```\r\n\r\n### 2.4 医院级别核查（参考清单）\r\n\r\n| 医院等级 | 医疗险 | 重疾险 | 寿险 | 意外险 |\r\n|---------|--------|--------|------|--------|\r\n| 三甲 | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） |\r\n| 三乙/三丙 | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） |\r\n| 二甲 | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） | ✅ 建议通过（须人工确认） |\r\n| 二乙 | 视条款（须人工确认） | 视条款（须人工确认） | ✅ 建议通过（须人工确认） | 视条款（须人工确认） |\r\n| 一级/社区 | 视条款（须人工确认） | 视条款（须人工确认） | ✅ 建议通过（须人工确认） | 视条款（须人工确认） |\r\n| 私立医院 | 视条款（通常排除，须人工确认） | 视条款（须人工确认） | 视条款（须人工确认） | 视条款（须人工确认） |\r\n\r\n### 2.5 险种责任匹配（参考清单）\r\n\r\n```python\r\n# ⚠️ 参考代码（须经人工审核）\r\ndef match_coverage(diagnosis: str, policy_coverage: dict) -> dict:\r\n    \"\"\"责任范围匹配（参考实现，须经人工审核）\"\"\"\r\n    icd_code = map_to_icd10(diagnosis)  # 须自行实现并验证\r\n\r\n    if policy_coverage[\"type\"] == \"critical_illness\":\r\n        if icd_code in policy_coverage.get(\"covered_icd\", []):\r\n            return {\"match\": True, \"payout_type_suggestion\": \"FULL\", \"human_review_required\": True}\r\n        else:\r\n            return {\"match\": False, \"action_suggestion\": \"REJECT\", \"human_review_required\": True}\r\n\r\n    if policy_coverage[\"type\"] == \"medical\":\r\n        if is_medical_necessary(diagnosis) and in_medical_catalog(icd_code):\r\n            return {\"match\": True, \"payout_type_suggestion\": \"REIMBURSE\", \"human_review_required\": True}\r\n        else:\r\n            return {\"match\": False, \"action_suggestion\": \"REJECT\", \"human_review_required\": True}\r\n\r\n    if policy_coverage[\"type\"] == \"accident\":\r\n        if claim.get(\"accident_proof\"):\r\n            return {\"match\": True, \"payout_type_suggestion\": \"FULL_OR_PARTIAL\", \"human_review_required\": True}\r\n        else:\r\n            return {\"match\": False, \"action_suggestion\": \"REQUEST_DOC\", \"human_review_required\": True}\r\n```\r\n\r\n---\r\n\r\n## 三、机器学习精审参考框架（须人工确认所有输出）\r\n\r\n### 3.1 NLP诊断解析 + ICD编码映射（参考）\r\n\r\n```python\r\n# ⚠️ 参考代码（须自行准备训练数据和模型，输出须经人工确认）\r\nimport torch\r\nfrom transformers import BertTokenizer, BertModel\r\n\r\ndef diagnose_nlp_analysis(diagnosis_text: str) -> dict:\r\n    \"\"\"\r\n    NLP诊断解析参考实现\r\n    ⚠️ 须自行 fine-tune 模型，输出须经人工确认\r\n    \"\"\"\r\n    # ⚠️ 须自行准备训练好的模型和ICD-10数据库\r\n    # tokenizer = BertTokenizer.from_pretrained(\"your-fine-tuned-model\")\r\n    # model = BertModel.from_pretrained(\"your-fine-tuned-model\")\r\n    # icd_db = load_your_icd10_database()\r\n\r\n    raise NotImplementedError(\r\n        \"须自行准备fine-tuned模型和ICD-10数据库，并对所有输出进行人工审核\"\r\n    )\r\n```\r\n\r\n### 3.2 治疗合理性分析（DRG/DIP对照参考）\r\n\r\n```python\r\n# ⚠️ 参考代码（须经人工审核）\r\ndef check_treatment_reasonableness(diagnosis: str, treatments: list, total_fee: float) -> dict:\r\n    \"\"\"\r\n    治疗合理性分析参考实现\r\n    ⚠️ 须自行准备DRG标准数据库，输出须经人工确认\r\n    \"\"\"\r\n    # drg_group = map_to_drg(diagnosis)  # 须自行实现\r\n    # std_fee_range = drg_group[\"fee_range\"]\r\n\r\n    flags = []\r\n    # if total_fee > std_fee_range[1]:\r\n    #     flags.append(f\"总费用超出DRG标准上限\")\r\n    # if total_fee < std_fee_range[0] * 0.5:\r\n    #     flags.append(f\"总费用异常偏低\")\r\n\r\n    return {\r\n        \"drg_group_suggestion\": \"须自行实现\",\r\n        \"flags\": flags,\r\n        \"reasonableness_score_suggestion\": max(0, 100 - len(flags) * 20),\r\n        \"human_review_required\": True,\r\n        \"note\": \"所有标记须经人工调查确认，不可自动拒付\"\r\n    }\r\n```\r\n\r\n### 3.3 欺诈风险评分（参考框架，非自动标记）\r\n\r\n```python\r\n# ⚠️ 参考代码（所有风险评分须经人工调查确认，不可自动拒付）\r\ndef fraud_risk_scoring_reference(claim: dict, graph: \"nx.Graph | None\") -> dict:\r\n    \"\"\"\r\n    欺诈风险评分参考框架\r\n    ⚠️ 所有评分仅为调查优先级参考，不可作为自动拒付依据\r\n    \"\"\"\r\n    risk_score = 0\r\n    risk_factors = []\r\n\r\n    # 因子1：同一患者短期内多次理赔（须人工核实）\r\n    # if claim.get(\"recent_claim_count\", 0) >= 3:\r\n    #     risk_score += 30\r\n    #     risk_factors.append(\"同一患者短期内多次理赔，建议人工调查\")\r\n\r\n    # 因子2：多家保险公司同时索赔（须人工调查）\r\n    # if claim.get(\"multi_insurer\", False):\r\n    #     risk_score += 40\r\n    #     risk_factors.append(\"多家公司同时索赔，建议人工调查\")\r\n\r\n    # 所有评分仅为调查优先级参考\r\n    if risk_score >= 70:\r\n        action = \"PRIORITY_INVESTIGATE\"  # 优先调查（非自动拒付）\r\n    elif risk_score >= 40:\r\n        action = \"ENHANCED_REVIEW\"  # 加强审核\r\n    else:\r\n        action = \"ROUTINE_REVIEW\"  # 常规审核\r\n\r\n    return {\r\n        \"risk_score_suggestion\": risk_score,\r\n        \"risk_factors\": risk_factors,\r\n        \"suggested_action\": action,  # ⚠️ 仅为建议\r\n        \"human_review_required\": True,\r\n        \"note\": \"所有欺诈风险评估须经人工调查确认，不可作为自动拒付依据\"\r\n    }\r\n```\r\n\r\n### 3.4 赔付金额计算（参考公式，须经人工确认）\r\n\r\n```python\r\n# ⚠️ 参考代码（所有赔付金额须经人工确认）\r\ndef calculate_payout_reference(claim_data: dict, policy: dict) -> dict:\r\n    \"\"\"赔付金额计算参考公式（须经人工确认）\"\"\"\r\n    coverage_type = policy[\"coverage_type\"]\r\n\r\n    if coverage_type == \"reimbursement\":  # 报销型\r\n        total_fee = claim_data[\"total_fee\"]\r\n        medical_insurance_pay = claim_data.get(\"medical_insurance_pay\", 0)\r\n        personal_pay = total_fee - medical_insurance_pay\r\n        deductible = policy.get(\"deductible\", 0)\r\n        reimbursement_ratio = policy.get(\"reimbursement_ratio\", 1.0)\r\n        payable = max(0, (personal_pay - deductible)) * reimbursement_ratio\r\n        return {\r\n            \"payout_suggestion\": round(payable, 2),  # ⚠️ 仅为建议\r\n            \"formula\": f\"({personal_pay} - {deductible}) × {reimbursement_ratio}\",\r\n            \"human_review_required\": True\r\n        }\r\n\r\n    elif coverage_type == \"fixed\":  # 定额给付\r\n        return {\r\n            \"payout_suggestion\": policy[\"sum_insured\"],\r\n            \"formula\": \"保额全额给付（须人工确认条款条件）\",\r\n            \"human_review_required\": True\r\n        }\r\n\r\n    # ... 其他险种类似，所有输出均须人工确认\r\n```\r\n\r\n---\r\n\r\n## 四、行业实践参考（文献综述，非本Skill实测结果）\r\n\r\n> ⚠️ **重要：** 以下公司实践描述为公开文献报道的参考信息，非本Skill的实测性能。部署效果取决于您的数据、模型和配置。\r\n\r\n### 4.1 平安\"111极速赔\"（公开报道参考）\r\n\r\n```\r\n公开报道的技术方向参考：\r\n  用户上传材料（拍照/PDF）\r\n    ↓\r\n  大模型解析（材料理解）\r\n    ↓\r\n  规则引擎判责\r\n    ↓\r\n  大模型复核（边缘case处理）\r\n    ↓\r\n  结果：公开报道称93%案件60秒内完成（⚠️ 此为平安报道数据，非本Skill承诺）\r\n```\r\n\r\n**创新点（公开报道摘录）：**\r\n- 大模型理解非结构化医疗文本\r\n- 端到端处理流程优化\r\n\r\n### 4.2 中国人寿智能理赔（公开报道参考）\r\n\r\n```\r\n公开报道的理赔金额分层处理方式（⚠️ 仅供参考，须按您的合规要求配置）：\r\n  理赔金额 ≤ 20000元 → 系统建议 → 人工审核确认 → 到账\r\n  20000元 < 金额 ≤ 50000元 → AI建议 + 人工复核 → 当天到账\r\n  金额 > 50000元 → 完整调查流程 → 3个工作日内\r\n```\r\n\r\n### 4.3 太保\"数字劳动力实验室\"（公开报道参考）\r\n\r\n```\r\n公开报道的Agent工作流方向（⚠️ 仅供参考）：\r\n  [接收案件] → [OCR识别] → [条款解析] → [判责决策建议]\r\n    → IF 简单案件 → 建议 → 人工确认\r\n    → IF 复杂案件 → 生成调查清单建议 → 派单调查员（人工）\r\n    → IF 高风险标记（建议） → 转反欺诈团队（人工调查）\r\n```\r\n\r\n---\r\n\r\n## 五、判责决策参考树（所有路径须人工确认）\r\n\r\n```\r\n案件提交\r\n  │\r\n  ├─ 规则检查（参考清单）\r\n  │   ├─ 等待期未过 → 建议拒付（须人工确认）\r\n  │   ├─ 既往症命中 → 建议拒付/比例赔付（须人工确认）\r\n  │   ├─ 未达免赔额 → 建议暂不赔付（须人工确认）\r\n  │   ├─ 医院不合规 → 提示确认/建议拒付（须人工确认）\r\n  │   └─ 险种不匹配 → 建议拒付（须人工确认）\r\n  │\r\n  ├─ IF 规则检查全部通过（建议）：\r\n  │   ├─ 简单案件 → 生成赔付建议（⚠️ 须人工确认）\r\n  │   ├─ 复杂案件 → 生成复核建议（⚠️ 须人工确认）\r\n  │   ├─ 大额案件 → 生成调查建议（⚠️ 须人工确认）\r\n  │   └─ 高风险标记（建议） → 进入反欺诈调查流程（人工）\r\n  │\r\n  └─ ⚠️【强制环节】持证理赔师人工审核 → 合规检查 → 最终决定\r\n```\r\n\r\n---\r\n\r\n## 六、反欺诈数据治理规范（必读）\r\n\r\n> 🔒 **数据治理要求（合规必备）**\r\n\r\n### 6.1 数据留存期限\r\n- 反欺诈图谱数据建议留存不超过 **2 年**，除非监管要求更长留存期\r\n- 已结案件的数据须定期归档或匿名化处理\r\n\r\n### 6.2 访问控制\r\n- 反欺诈图谱查询权限仅开放给 **授权欺诈调查员**\r\n- 禁止非授权人员（如客服、销售）访问欺诈评分和图谱数据\r\n- 所有查询须留下审计日志（谁、何时、查询了哪个案件）\r\n\r\n### 6.3 数据更正流程\r\n- 被保险人和投诉方有权请求更正错误数据\r\n- 数据更正请求须在 **15 个工作日内** 处理完毕\r\n- 更正后须通知所有曾收到该错误数据的决策环节\r\n\r\n### 6.4 图谱污染防护\r\n- 新案件数据进入图谱前，须经 **人工审核确认** 数据质量\r\n- 禁止将未经验证的第三方提供数据直接写入生产图谱\r\n- 定期（建议每季度）对图谱数据进行质量审计\r\n\r\n---\r\n\r\n## 七、模型效果评估指标（文献参考值，非承诺）\r\n\r\n| 指标 | 文献参考值 | 说明 |\r\n|------|------------|------|\r\n| 自动通过率 | ~70% | 文献报道行业水平，非本Skill承诺 |\r\n| 误拒率（False Reject） | ≤ 2% | 文献报道行业水平，非本Skill承诺 |\r\n| 漏检率（Fraud Escape） | ≤ 1% | 文献报道行业水平，非本Skill承诺 |\r\n| 平均处理时长 | ≤ 1分钟（自动案件） | 取决于部署配置，非本Skill承诺 |\r\n| 客户满意度 | ≥ 85% | 文献报道行业水平，非本Skill承诺 |\r\n\r\n> ⚠️ **重要：** 以上指标均为文献报道的行业参考值，非本Skill的性能承诺。实际效果取决于您的数据质量、模型训练、规则配置和人工审核流程。\r\n\r\n---\r\n\r\n*Last updated: 2026-05-05 — 删除所有自动审批描述；准确率/性能数据标注为文献基准（非本Skill实测）；所有输出增加\"须人工确认\"标注；增加反欺诈数据治理规范；增加强制人工审核环节说明。*\n\nFile v5.0.3:references/claims_ocr_tech.md\n\n# 保险理赔多模态医疗票据 OCR 技术参考（ advisory only / 咨询级）\r\n\r\n> ⚠️ **DISCLAIMER / 免责声明**\r\n> - **English:** This document provides technical reference and example code ONLY. It does NOT provide production-ready models, pre-trained weights, or bundled API credentials. All accuracy figures (e.g., \"~95%\") are vendor-published benchmarks under ideal conditions, NOT validated results of your deployment. You must provide your own API keys, training data, and models. All OCR results must be reviewed by human staff before use in claim decisions.\r\n> - **中文：** 本文档仅提供技术参考和示例代码，不含生产级模型、预训练权重或绑定的API凭证。所有准确率数据（如\"~95%\"）均为厂商发布的理想条件下基准结果，非您部署后的实测结果。您必须自行提供API密钥、训练数据和模型。所有OCR结果在用于理赔决定前，必须经人工审核。\r\n\r\n> 🔒 **DATA SECURITY / 数据安全**\r\n> - Medical invoices contain sensitive personal information (name, ID number, diagnosis, hospital). Before sending to ANY cloud OCR provider, you MUST: (1) obtain user consent, (2) redact unnecessary PII fields, (3) confirm the vendor's data retention policy (prefer: no storage / auto-delete within 24h), (4) consider private on-prem deployment for production use.\r\n> - API keys and credentials MUST be stored in environment variables or a secret manager. NEVER hardcode keys in production code.\r\n\r\n---\r\n\r\n## 一、主流 OCR 服务横向对比（厂商公开基准数据）\r\n\r\n| 服务商 | 支持票据类型 | 厂商公开准确率基准 | 调用限频 | 参考价格（元/千次） | 特色功能 |\r\n|---------|--------------|----------------|----------|----------------|----------|\r\n| 百度 AI | 发票+7类病历+2类报告 | ~95%（厂商基准） | 50次/秒 | 0.15 | 覆盖最全，EasyDL 自训练 |\r\n| 腾讯云 | 全国门诊/住院发票 | ~93%（厂商基准） | 5次/秒 | 0.12 | 与腾讯云生态打通，理赔场景优化 |\r\n| 阿里云 + 深智恒际 | 各省市门诊发票 | ~92%（厂商基准） | 20次/秒 | 0.18 | 支持定制化训练，票据类型覆盖深度好 |\r\n| 合合信息 TextIn | 医疗票据全品类 | ~94%（厂商基准） | 100次/秒 | 0.20 | 深度学习融合，表格还原精度高 |\r\n\r\n> ⚠️ **重要：** 以上准确率为厂商在理想条件下发布的基准数据，实际准确率取决于图像质量、票据类型、训练数据质量等因素。部署前须自行测试验证。\r\n\r\n---\r\n\r\n## 二、技术架构参考（需自行实现）\r\n\r\n### 完整处理流水线（参考架构）\r\n\r\n```\r\n原始图像输入\r\n  ↓\r\n[图像预处理]  — 需自行实现或调用云服务API\r\n  - 去噪（高斯滤波 / 中值滤波）\r\n  - 倾斜校正（霍夫直线检测 + 仿射变换）\r\n  - 二值化（Otsu 自适应阈值）\r\n  - 透视校正（四点变换）\r\n  ↓\r\n[版面分析]  — 需自行实现\r\n  - 关键点检测（角点 / 印章 / 表格线）\r\n  - 区域分割（标题 / 明细 / 印章区）\r\n  ↓\r\n[文本检测]  — 可选：CTPN / PSENet / DB-Net（需自行训练）\r\n  - 文字行定位\r\n  - 弯曲文本矫正\r\n  ↓\r\n[文本识别]  — 可选：CRNN + CTC / ViT-OCR（需自行训练）\r\n  - CNN 特征提取（ResNet50 / EfficientNet-B4）\r\n  - RNN 序列建模（BiLSTM × 2 层）\r\n  - Attention 对齐（Coverage Attention）\r\n  - CTC 解码 → 字符序列\r\n  ↓\r\n[结构化提取]  — 需自行实现规则 + NER 模型\r\n  - 正则表达式匹配（金额 / 日期 / 发票号）\r\n  - 命名实体识别 BIOES 标注\r\n  - 字段归一化（金额统一到 `float`，日期统一到 `YYYY-MM-DD`）\r\n  ↓\r\n[输出]  JSON 结构化结果（需人工审核）\r\n```\r\n\r\n> ⚠️ **重要：** 以上架构需要您自行准备训练数据、训练模型或购买云服务API。本Skill不提供任何预训练模型权重。\r\n\r\n---\r\n\r\n## 三、Python 调用示例（需自行配置API密钥）\r\n\r\n### ⚠️ 安全提醒（使用任何OCR服务前必读）\r\n\r\n1. **API Key 管理：** 将密钥存入环境变量，禁止硬编码\r\n2. **数据脱敏：** 发送图像前，用遮盖方式隐去姓名、身份证号等敏感字段\r\n3. **数据留存：** 确认服务商数据处理政策，优先选择\"不留存\"或\"24小时内自动删除\"\r\n4. **生产环境：** 建议使用私有化部署方案，医疗数据不出域\r\n\r\n### 3.1 百度 AI 医疗票据识别（示例）\r\n\r\n```python\r\nimport requests\r\nimport base64\r\nimport json\r\nimport os\r\n\r\n# ⚠️ 安全做法：从环境变量读取密钥，禁止硬编码\r\nAPI_KEY = os.environ.get(\"BAIDU_OCR_API_KEY\")\r\nSECRET_KEY = os.environ.get(\"BAIDU_OCR_SECRET_KEY\")\r\n\r\nif not API_KEY or not SECRET_KEY:\r\n    raise ValueError(\"请设置环境变量 BAIDU_OCR_API_KEY 和 BAIDU_OCR_SECRET_KEY\")\r\n\r\n# 获取 access_token\r\ndef get_access_token():\r\n    url = f\"https://aip.baidu.com/oauth/2.0/token?grant_type=client_credentials&client_id={API_KEY}&client_secret={SECRET_KEY}\"\r\n    return requests.get(url).json()[\"access_token\"]\r\n\r\n# 医疗票据识别（⚠️ 建议：事前对图像做脱敏处理）\r\ndef recognize_medical_invoice(image_path: str, token: str) -> dict:\r\n    with open(image_path, \"rb\") as f:\r\n        img_b64 = base64.b64encode(f.read()).decode(\"utf-8\")\r\n\r\n    url = f\"https://aip.baidu.com/rest/2.0/ocr/v1/medical_invoice?access_token={token}\"\r\n    payload = {\r\n        \"image\": img_b64,\r\n        \"detect_direction\": \"true\",\r\n        \"probability\": \"true\",\r\n    }\r\n    headers = {\"Content-Type\": \"application/x-www-form-urlencoded\"}\r\n    resp = requests.post(url, data=payload, headers=headers)\r\n    return resp.json()\r\n\r\n# 使用示例（⚠️ 结果须经人工审核）\r\n# token = get_access_token()\r\n# result = recognize_medical_invoice(\"invoice.jpg\", token)\r\n# print(json.dumps(result, ensure_ascii=False, indent=2))\r\n```\r\n\r\n### 3.2 腾讯云医疗发票识别（示例）\r\n\r\n```python\r\n# ⚠️ 安全做法：从环境变量读取密钥\r\nimport os\r\nfrom tencentcloud.common import credential\r\nfrom tencentcloud.common.profile.client_profile import ClientProfile\r\nfrom tencentcloud.common.profile.http_profile import HttpProfile\r\nfrom tencentcloud.ocr.v20181119 import ocr_client, models\r\n\r\ndef recognize_tencent_medical(image_path: str):\r\n    cred = credential.Credential(\r\n        os.environ.get(\"TENCENTCLOUD_SECRET_ID\"),\r\n        os.environ.get(\"TENCENTCLOUD_SECRET_KEY\")\r\n    )\r\n    http_profile = HttpProfile()\r\n    http_profile.endpoint = \"ocr.tencentcloudapi.com\"\r\n    client_profile = ClientProfile()\r\n    client_profile.httpProfile = http_profile\r\n    client = ocr_client.OcrClient(cred, \"ap-guangzhou\", client_profile)\r\n\r\n    with open(image_path, \"rb\") as f:\r\n        img_b64 = base64.b64encode(f.read()).decode(\"utf-8\")\r\n\r\n    req = models.MedicalInvoiceOCRRequest()\r\n    req.ImageBase64 = img_b64\r\n    resp = client.MedicalInvoiceOCR(req)\r\n    return resp.to_json_string(indent=2)\r\n\r\n# print(recognize_tencent_medical(\"invoice.jpg\"))\r\n```\r\n\r\n### 3.3 本地自训练 OCR（PyTorch + CRNN）— 参考代码\r\n\r\n```python\r\n# ⚠️ 注意：以下为参考架构代码，需要您自行准备训练数据和标注\r\nimport torch\r\nimport torch.nn as nn\r\nimport torchvision.models as models\r\nfrom torch.nn.utils.rnn import pack_padded_sequence\r\n\r\nclass CRNN(nn.Module):\r\n    \"\"\"\r\n    CRNN 参考模型架构：\r\n    CNN（特征提取）→ BiLSTM（序列建模）→ CTC（解码）\r\n    需要自行准备训练数据和训练脚本。\r\n    \"\"\"\r\n    def __init__(self, num_chars: int, hidden_size: int = 256):\r\n        super().__init__()\r\n        # CNN 主干：EfficientNet-B0（需自行预训练或下载权重）\r\n        backbone = models.efficientnet_b0(weights=None)  # 需自行提供权重\r\n        self.cnn = nn.Sequential(*list(backbone.children())[:-2])\r\n        self.cnn.add_module(\"adaptive_pool\", nn.AdaptiveAvgPool2d((None, 1)))\r\n\r\n        # BiLSTM × 2\r\n        self.lstm = nn.LSTM(\r\n            input_size=1280,\r\n            hidden_size=hidden_size,\r\n            num_layers=2,\r\n            bidirectional=True,\r\n            batch_first=True\r\n        )\r\n        self.fc = nn.Linear(hidden_size * 2, num_chars)\r\n\r\n    def forward(self, x):\r\n        conv = self.cnn(x)\r\n        conv = conv.squeeze(2)\r\n        conv = conv.permute(0, 2, 1)\r\n        lstm_out, _ = self.lstm(conv)\r\n        logits = self.fc(lstm_out)\r\n        return logits\r\n\r\n# 训练需自行准备：\r\n# - 训练数据（标注好的医疗票据图像）\r\n# - CTC Loss 配置\r\n# - 训练脚本\r\n# model = CRNN(num_chars=6624)\r\n# ctc_loss = nn.CTCLoss(zero_infinity=True)\r\n# optimizer = torch.optim.AdamW(model.parameters(), lr=1e-4)\r\n```\r\n\r\n---\r\n\r\n## 四、票据结构化字段规范（参考格式）\r\n\r\n### 4.1 全国医疗门诊发票标准字段（参考）\r\n\r\n```yaml\r\nbasic_info:\r\n  invoice_code: str        # 票据代码\r\n  invoice_number: str      # 票据号码\r\n  invoice_date: date      # 开票日期\r\n  verify_code: str        # 校验码\r\n  hospital_name: str     # 医院名称\r\n  hospital_level: str    # 医院等级（三级/二级/一级）\r\n\r\namount_info:\r\n  total: float          # 总金额（含税）\r\n  medical_insurance: float  # 医保统筹支付\r\n  personal_account: float  # 个人账户支付\r\n  self_pay: float       # 个人自付（分类自负）\r\n  self_fund: float      # 个人自费\r\n  deductible: float     # 起付线（免赔额部分）\r\n\r\ndetail_line_items:        # 费用明细行（数组）\r\n  - name: str           # 药品名称/诊疗项目名称\r\n    specification: str   # 规格\r\n    unit: str           # 单位\r\n    quantity: int       # 数量\r\n    unit_price: float   # 单价\r\n    total_price: float  # 该项总价\r\n    category: str       # 甲类/乙类/丙类/自费\r\n    insurance_code: str  # 医保目录编码\r\n\r\ndiagnosis:\r\n  icd10_code: str      # ICD-10 诊断编码\r\n  diagnosis_name: str  # 诊断名称（中文）\r\n  department: str      # 就诊科室\r\n```\r\n\r\n### 4.2 全国医疗住院发票附加字段（参考）\r\n\r\n```yaml\r\nhospitalization:\r\n  admission_date: date  # 入院日期\r\n  discharge_date: date # 出院日期\r\n  stay_days: int       # 住院天数\r\n  bed_number: str     # 床位号\r\n  total_prescriptions: int  # 总处方数\r\n  surgery_name: str   # 手术名称（如有）\r\n  drg_code: str       # DRG 分组编码（如有）\r\n```\r\n\r\n---\r\n\r\n## 五、多模态融合识别方案（兜底方案参考）\r\n\r\n当单一 OCR 效果不佳时（如盖章覆盖文字、票据折叠），可启用多模态大模型作为兜底：\r\n\r\n```python\r\n# ⚠️ 注意：多模态大模型也会接收图像数据，须同样遵守数据安全规定\r\ndef multimodal_fallback(image_path: str, ocr_result: dict) -> dict:\r\n    \"\"\"\r\n    OCR 识别置信度低于阈值时，可触发多模态识别作为参考。\r\n    ⚠️ 结果须经人工审核确认。\r\n    \"\"\"\r\n    import base64, os\r\n\r\n    # ⚠️ 建议：发送前对敏感字段做模糊化处理\r\n    with open(image_path, \"rb\") as f:\r\n        img_b64 = base64.b64encode(f.read()).decode()\r\n\r\n    prompt = \"\"\"\r\n    这是一张医疗发票照片，OCR 识别结果置信度较低。\r\n    请你根据图片内容，提取以下字段（JSON 格式）：\r\n    - 发票号码\r\n    - 开票日期\r\n    - 医院名称\r\n    - 总金额\r\n    - 医保统筹支付\r\n    - 个人自付\r\n    - 临床诊断\r\n    如果图片中看不清楚，则对应字段填 null。\r\n    \"\"\"\r\n\r\n    # 调用多模态 API（须自行配置密钥并遵守数据安全规定）\r\n    # response = call_multimodal_api(\r\n    #     model=\"qwen-vl-max\",\r\n    #     messages=[{\r\n    #         \"role\": \"user\",\r\n    #         \"content\": [\r\n    #             {\"image\": f\"data:image/jpeg;base64,{img_b64}\"},\r\n    #             {\"text\": prompt}\r\n    #         ]\r\n    #     }]\r\n    # )\r\n    # return parse_json_from_response(response)\r\n    raise NotImplementedError(\"须自行实现多模态API调用，并配置数据安全保护\")\r\n```\r\n\r\n**触发多模态兜底的条件（参考）：**\r\n- OCR 置信度均值 < 0.75\r\n- 关键字段缺失 ≥ 2 个（发票号 / 金额 / 日期）\r\n- 印章严重遮挡文字区域（通过图像分割检测到大面积红色区域）\r\n\r\n---\r\n\r\n## 六、部署建议（参考）\r\n\r\n| 部署方式 | 适用场景 | 推荐技术栈 | 数据合规建议 |\r\n|-----------|-----------|-----------|--------------|\r\n| **云端 API 调用** | 中小规模（< 10万张/月） | 百度 AI / 腾讯云 SDK | 确认服务商数据留存政策；建议签署数据处理协议 |\r\n| **私有化部署** | 大型保险公司 / 数据不出域 | PyTorchServe + CRNN 自训练模型 | 数据全程不出域，符合《个人信息保护法》要求 |\r\n| **混合架构** | 高可用要求 + 数据合规 | 云端 OCR 预处理 + 本地规则引擎核验 | 云端仅传输必要字段，敏感字段本地处理 |\r\n| **边缘端部署** | 移动端 / 小程序拍照即识别 | NCNN + Int8 量化 CRNN 模型 | 数据在设备端处理，不上传云端 |\r\n\r\n---\r\n\r\n## 七、数据留存与访问控制（必读）\r\n\r\n> ⚠️ **法律合规要求（中国《个人信息保护法》）**\r\n\r\n1. **数据最小化原则：** 仅收集和传输理赔处理所必需的最少字段\r\n2. **用户同意：** 将医疗票据发送至OCR服务前，须获得用户明确同意\r\n3. **数据留存期限：** 建议 OCR 服务商不留存数据，或设置 24 小时内自动删除；本地图谱数据留存不超过 2 年\r\n4. **访问控制：** 医疗票据图像和识别结果仅限授权理赔人员访问，禁止非授权人员查看\r\n5. **数据更正权利：** 被保人有权请求更正错误数据，必须在 15 个工作日内处理\r\n6. **跨境数据传输：** 如使用境外OCR服务，须进行数据出境安全评估\r\n\r\n---\r\n\r\n*Last updated: 2026-05-05 — 添加数据安全声明、API密钥安全提醒、准确率数据标注为厂商基准（非实测）、删除自动审批描述、增加数据留存和访问控制规范。*\n\nFile v5.0.3:references/claims_report_templates.md\n\n# 保险理赔报告模板与输出规范\r\n\r\n> 标准化的理赔分析报告模板，适用于各险种理赔场景。\r\n> 包含：报告结构、快捷输出指令、七项检查清单、FAQ模板。\r\n\r\n---\r\n\r\n## 一、标准理赔分析报告（通用模板）\r\n\r\n```markdown\r\n# 保险理赔智能分析报告\r\n\r\n**生成时间**: {{YYYY-MM-DD HH:mm}}\r\n**案件编号**: CL-{{YYYYMMDD}}-{{XXXX}}\r\n**保单号**: {{policy_no}}\r\n**被保险人**: {{name}}（{{gender}}，{{age}}岁）\r\n**险种类别**: {{coverage_type}}\r\n**理赔类型**: {{claim_type}}（门诊/住院/重疾/身故/意外/车险）\r\n**处理状态**: {{AUTO_PASS | MANUAL_REVIEW | HIGH_RISK | REJECT}}\r\n\r\n---\r\n\r\n## 一、票据识别结果\r\n\r\n### 1.1 医疗票据OCR识别\r\n\r\n| 票据类型 | 票据代码/号 | 开票日期 | 金额（元） | 识别状态 |\r\n|---------|------------|---------|-----------|---------|\r\n| {{ticket_type}} | {{ticket_code}} | {{date}} | ¥{{amount}} | {{status}} |\r\n\r\n### 1.2 关键字段提取\r\n\r\n| 字段 | 提取结果 | 置信度 |\r\n|------|---------|---------|\r\n| 医院名称 | {{hospital}} | {{confidence}} |\r\n| 诊断信息 | {{diagnosis}}（ICD-10: {{icd10}}） | {{confidence}} |\r\n| 总费用 | ¥{{total}} | {{confidence}} |\r\n| 医保支付 | ¥{{medical_pay}} | {{confidence}} |\r\n| 个人自付 | ¥{{self_pay}} | {{confidence}} |\r\n| 住院天数 | {{days}}天 | {{confidence}} |\r\n\r\n---\r\n\r\n## 二、责任认定结果\r\n\r\n| 审核维度 | 判定结果 | 说明 |\r\n|---------|---------|------|\r\n| 等待期 | {{PASS/FAIL}} | {{reason}} |\r\n| 既往症 | {{PASS/FAIL}} | {{reason}} |\r\n| 免赔额 | {{PASS/FAIL}} | 本年度累计¥{{accumulated}}，免赔额¥{{deductible}} |\r\n| 医院级别 | {{COMPLIANT/NON_COMPLIANT}} | {{hospital_level}} |\r\n| 保障范围 | {{COVERED/NOT_COVERED}} | {{reason}} |\r\n| 事故性质 | {{CONFIRMED/PENDING}} | {{reason}} |\r\n\r\n**综合判责结论**：{{CONCLUSION}}\r\n\r\n---\r\n\r\n## 三、赔付计算\r\n\r\n### 3.1 计算公式\r\n\r\n```\r\n可赔付金额 = （个人自付 - 免赔额）× 赔付比例\r\n           = （¥{{self_pay}} - ¥{{deductible}}）× {{ratio}}%\r\n           = ¥{{payable}}\r\n```\r\n\r\n### 3.2 赔付明细\r\n\r\n| 项目 | 金额（元） | 说明 |\r\n|------|-----------|------|\r\n| 发票总金额 | ¥{{total}} | OCR识别 |\r\n| 医保统筹支付 | ¥{{medical_pay}} | 医保基金支付部分 |\r\n| 个人账户支付 | ¥{{account_pay}} | 医保个人账户 |\r\n| **个人自付** | **¥{{self_pay}}** | 分类自负+自负 |\r\n| - 免赔额 | ¥{{deductible}} | 年免赔额 |\r\n| × 赔付比例 | {{ratio}}% | 条款约定 |\r\n| **核定赔付金额** | **¥{{payable}}** | 最终赔付 |\r\n\r\n---\r\n\r\n## 四、反欺诈风险评估\r\n\r\n| 风险维度 | 评分（0-100） | 风险等级 | 说明 |\r\n|---------|--------------|---------|------|\r\n| 票据真实性 | {{score}} | {{LOW/MEDIUM/HIGH}} | {{reason}} |\r\n| 行为异常 | {{score}} | {{level}} | {{reason}} |\r\n| 关系图谱 | {{score}} | {{level}} | {{reason}} |\r\n| 医疗合理性 | {{score}} | {{level}} | {{reason}} |\r\n| **综合风险** | **{{total_score}}** | **{{FINAL_LEVEL}}** | {{action}} |\r\n\r\n---\r\n\r\n## 五、最终结论\r\n\r\n- **理赔核定**：{{FULL_PAY | PARTIAL_PAY | REJECT}}\r\n- **赔付金额**：¥{{final_pay}}（大写：{{capitalized}}）\r\n- **到账时间**：{{days}}个工作日内\r\n- **所需材料**：{{status}}（已收齐 / 需补充：{{list}}）\r\n- **案件状态**：{{status}}\r\n\r\n---\r\n\r\n## 六、补充材料通知（如需要）\r\n\r\n请补充以下材料：\r\n1. {{doc1}}\r\n2. {{doc2}}\r\n3. {{doc3}}\r\n\r\n---\r\n\r\n> ⚠️ 本报告由 AI 理赔智能系统自动生成，核定金额仅供参考，\r\n> 最终以保险公司理赔部门审核结果为准。\r\n> 如有异议，请在收到通知书之日起60日内提出复议申请。\r\n```\r\n\r\n---\r\n\r\n## 二、分险种报告模板\r\n\r\n### 2.1 医疗险理赔报告（简版）\r\n\r\n```markdown\r\n# 医疗险理赔核定通知书\r\n\r\n**被保险人**：{{name}}\r\n**保单号**：{{policy_no}}\r\n**理赔金额**：¥{{payable}}\r\n\r\n## 费用明细\r\n- 门诊/住院总费用：¥{{total}}\r\n- 医保统筹支付：¥{{medical_pay}}\r\n- 个人自付：¥{{self_pay}}\r\n- 减：年免赔额：¥{{deductible}}\r\n- 乘：赔付比例：{{ratio}}%\r\n- **应付赔款：¥{{payable}}**\r\n\r\n## 赔付说明\r\n本次理赔已通过智能审核，赔款将于3个工作日内转入您指定的银行账户。\r\n\r\n---\r\n案件编号：{{case_id}}  |  客服热线：955XX\r\n```\r\n\r\n### 2.2 重疾险理赔报告\r\n\r\n```markdown\r\n# 重大疾病保险金给付通知书\r\n\r\n**被保险人**：{{name}}\r\n**保单号**：{{policy_no}}\r\n**重大疾病**：{{diagnosis}}（ICD-10：{{icd10}}）\r\n**给付金额**：¥{{sum_insured}}（大写：{{capitalized}}）\r\n\r\n## 核定依据\r\n- 病理报告已确认疾病符合条款约定\r\n- 等待期已满（{{days}}天）\r\n- 无责任免除情形\r\n\r\n## 给付说明\r\n重大疾病保险金已全额给付，本合同继续有效（如含身故责任）。\r\n\r\n---\r\n案件编号：{{case_id}}  |  给付日期：{{date}}\r\n```\r\n\r\n### 2.3 车险理赔计算书\r\n\r\n```markdown\r\n# 车辆损失险理赔计算书\r\n\r\n**被保险人**：{{name}}\r\n**保单号**：{{policy_no}}\r\n**事故编号**：{{accident_id}}\r\n**事故责任**：{{responsibility}}（全责/主责/同责/次责/无责）\r\n\r\n## 定损明细\r\n| 项目 | 金额（元） |\r\n|------|-----------|\r\n| 配件费用 | ¥{{parts}} |\r\n| 工时费用 | ¥{{labor}} |\r\n| 施救费用 | ¥{{rescue}} |\r\n| 减：残值扣减 | ¥{{salvage}} |\r\n| **核定损失** | **¥{{assessed_loss}}** |\r\n\r\n## 赔付计算\r\n- 核定损失：¥{{assessed_loss}}\r\n- 事故责任比例：{{resp_ratio}}%\r\n- 减：免赔额：¥{{deductible}}\r\n- **应付赔款：¥{{payable}}**\r\n\r\n---\r\n定损员：{{adjuster}}  |  审核日期：{{date}}\r\n```\r\n\r\n---\r\n\r\n## 三、快捷输出指令\r\n\r\n| 指令 | 输入示例 | 输出内容 |\r\n|------|---------|---------|\r\n| `/理赔` | `/理赔 门诊发票213.5元 诊断上呼吸道感染` | 完整理赔分析报告 |\r\n| `/识别` | `/识别 上传发票图片` | OCR结构化结果 |\r\n| `/判责` | `/判责 保单P123 诊断J06.9` | 责任认定结果 |\r\n| `/计算` | `/计算 发票5000元 免赔额1000元 比例80%` | 赔付金额 |\r\n| `/拒付` | `/拒付 案件CL20260504001` | 拒付通知书 |\r\n| `/反欺诈` | `/反欺诈 患者张三 近期5次理赔` | 欺诈风险报告 |\r\n| `/车险` | `/车险 事故认定书同责 定损2960元` | 车险理赔计算书 |\r\n| `/重疾` | `/重疾 病理报告肺腺癌T2N0M0` | 重疾险给付通知书 |\r\n\r\n---\r\n\r\n## 四、七项检查清单\r\n\r\n每次输出理赔报告前，必须检查以下7项：\r\n\r\n```markdown\r\n## ✅ 七项检查清单\r\n\r\n- [ ] 1. 被保险人身份信息是否完整（姓名/身份证/保单号）\r\n- [ ] 2. 票据识别关键字段是否齐全（金额/日期/医院/诊断）\r\n- [ ] 3. 责任免除条款是否已核查（等待期/既往症/医院级别）\r\n- [ ] 4. 赔付计算公式是否正确（免赔额/赔付比例/给付限额）\r\n- [ ] 5. 反欺诈风险评估是否已执行（综合风险评分）\r\n- [ ] 6. 金额大小写是否一致（阿拉伯数字与中文大写）\r\n- [ ] 7. 法律合规条款是否已标注（异议期/客服电话/监管依据）\r\n```\r\n\r\n---\r\n\r\n## 五、拒付通知书模板\r\n\r\n```markdown\r\n# 理赔拒付通知书\r\n\r\n**案件编号**：{{case_id}}\r\n**被保险人**：{{name}}\r\n**保单号**：{{policy_no}}\r\n**拒付日期**：{{date}}\r\n\r\n## 拒付原因\r\n\r\n经审核，您的理赔申请不符合保险条款约定，具体原因：\r\n\r\n**{{reason_code}}**：{{reason_detail}}\r\n\r\n{{#if 等待期未过}}\r\n等待期条款约定：本合同生效后{{waiting_days}}天为等待期，等待期内确诊的疾病，本公司不承担保险责任。\r\n您的确诊日期距离保单生效日仅{{actual_days}}天，在等待期内。\r\n{{/if}}\r\n\r\n{{#if 既往症}}\r\n根据条款\"责任免除\"第{{clause_no}}条，被保险人在本合同生效前已患有的疾病或症状，属于责任免除范围。\r\n{{/if}}\r\n\r\n{{#if 非保障范围}}\r\n本次就诊诊断（{{diagnosis}}）不属于本合同约定的保障范围。\r\n本产品保障范围详见条款第{{clause_no}}条。\r\n{{/if}}\r\n\r\n## 您的权利\r\n\r\n如您对本决定有异议，可在收到本通知书之日起60日内，向本公司申请复议，或向{{regulator}}投诉，或依法提起诉讼。\r\n\r\n---\r\n客服热线：{{hotline}}  |  投诉邮箱：{{email}}\r\n```\r\n\r\n---\r\n\r\n## 六、补充材料通知书模板\r\n\r\n```markdown\r\n# 补充理赔材料通知书\r\n\r\n**案件编号**：{{case_id}}\r\n**被保险人**：{{name}}\r\n**当前状态**：材料不齐，待补充\r\n\r\n## 需补充材料清单\r\n\r\n请您在{{deadline}}前补充以下材料，以便我们继续处理您的理赔申请：\r\n\r\n| 序号 | 材料名称 | 要求 | 备注 |\r\n|------|---------|------|------|\r\n| 1 | {{doc1}} | {{requirement}} | {{note}} |\r\n| 2 | {{doc2}} | {{requirement}} | {{note}} |\r\n| 3 | {{doc3}} | {{requirement}} | {{note}} |\r\n\r\n## 补充方式\r\n\r\n1. **线上上传**：登录{{app_name}} APP → 理赔服务 → 案件查询 → 补充材料\r\n2. **邮件发送**：将材料照片发送至 {{email}}\r\n3. **线下递交**：前往本公司客户服务中心（地址：{{address}}）\r\n\r\n---\r\n客服热线：{{hotline}}  |  案件处理员：{{handler}}\r\n```\r\n\r\n---\r\n\r\n## 七、FAQ回复模板\r\n\r\n### Q1：理赔需要多长时间？\r\n\r\n> 根据《保险法》第23条规定，保险人收到理赔申请后，应当及时作出核定；情形复杂的，应当在30日内作出核定。\r\n> 本公司承诺：简单案件（3000元以下）当日完成；普通案件（3000-20000元）3个工作日内完成；复杂案件30日内完成。\r\n\r\n### Q2：为什么我的理赔被拒付了？\r\n\r\n> 拒付原因通常包括：①等待期内出险；②既往症；③非保障范围；④医院不符合要求；⑤免赔额未达。\r\n> 详细拒付原因请查看《理赔拒付通知书》，如有异议可申请复议。\r\n\r\n### Q3：理赔款什么时候到账？\r\n\r\n> 理赔款在理赔决定作出后10日内支付（法律规定）。本公司实际到账时间：自动理赔案件实时到账；人工审核案件3个工作日内到账。\r\n\r\n### Q4：我对理赔决定不满意怎么办？\r\n\r\n> 您有以下途径：\r\n> 1. 向本公司申请复议（60日内）\r\n> 2. 向国家金融监督管理总局投诉\r\n> 3. 依法提起诉讼\r\n\r\n### Q5：医疗发票OCR识别不准确怎么办？\r\n\r\n> 您可以提供更清晰的发票照片重新识别，或选择人工审核通道。OCR识别结果仅供参考，最终以人工审核为准。\r\n\r\n---\r\n\r\n## 八、输出格式规范总结\r\n\r\n| 格式类型 | 适用场景 | 输出方式 |\r\n|---------|---------|---------|\r\n| 完整报告 | 综合理赔分析 | Markdown 表格 + 分段 |\r\n| 核定通知书 | 赔付决定 | 正式文书格式（含印章位置） |\r\n| 拒付通知书 | 拒付决定 | 正式文书格式 + 法律条款引用 |\r\n| 补充材料通知 | 材料不齐 | 清单格式 + 补充方式说明 |\r\n| 欺诈风险报告 | 高风险案件 | 评分表格 + 风险因子列表 |\r\n| 计算书 | 车险/财产险 | 定损明细表格 + 计算公式 |\r\n\r\n---\r\n\r\n> 📌 **重要声明**：所有输出模板均为参考格式，实际使用时须根据各保险公司具体条款和监管要求调整。AI生成内容须经持牌理赔师审核确认后，方可作为正式理赔文书使用。\n\nFile v5.0.3:skill-card.md\n\n## Description: <br>\nProvides advisory templates, checklists, and decision-support frameworks for insurance claim OCR, liability review, anti-fraud assessment, and claim reporting, with human review required for every decision. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[gechengling](https://clawhub.ai/user/gechengling) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nInsurance claims teams, reviewers, and developers use this skill to draft claim analysis reports, OCR integration guidance, liability review checklists, anti-fraud review prompts, and standardized claim notices. Outputs are decision-support drafts and require licensed human review before customer, payment, denial, or fraud actions. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: Review before execution as proposals could introduce incorrect or misleading guidance into skills. <br>\nMitigation: Review and scan skill before deployment. <br>\n\n## Reference(s): <br>\n- [Insurance Claims Intelligence release page](https://clawhub.ai/gechengling/insurance-claims-intelligence) <br>\n- [Claims OCR technical reference](references/claims_ocr_tech.md) <br>\n- [Claims liability engine reference](references/claims_liability_engine.md) <br>\n- [Claims report templates](references/claims_report_templates.md) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [text, markdown, code, configuration, guidance] <br>\n**Output Format:** [Markdown with checklists, report templates, structured draft outputs, and example code snippets] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [Draft-only decision support; OCR examples require user-provided credentials and data, and claim decisions require licensed human review.] <br>\n\n## Skill Version(s): <br>\n5.0.3 (source: server release metadata) <br>\n\n## Ethical Considerations: <br>\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment. <br>\n\nArchive v5.0.2: 7 files, 32659 bytes\n\nFiles: README.md (7594b), references/claims_liability_engine.md (18889b), references/claims_ocr_tech.md (14039b), references/claims_report_templates.md (11395b), skill-card.md (2678b), SKILL.md (25650b), _meta.json (148b)\n\nFile v5.0.2:SKILL.md\n\n---\r\nname: Insurance Claims Intelligence Expert\r\ndescription: Advisory skill for insurance claims processing workflows �� provides templates, checklists, and decision-support frameworks for medical OCR, liability determination, anti-fraud assessment, and claims reporting. Human review required for all claim decisions. Keywords: insurance claims, claims advisory, medical OCR, anti-fraud, insurance tech, China insurance, decision support, ��������, ������, ҽ�Ƶ���ʶ��, �����϶�, ���ⱨ��, ����, �������, ҽ��������, �ؼ�����, ��������.\r\nslug: insurance-claims\r\nversion: \"5.0.1\"\r\n\ncapabilities:\n  - educational-reference\n  - advisory-only\n  - requires-human-review\n  - no-executable-code\n---\r\n\r\n# Insurance Claims Intelligence Expert / ������ҵ��������ר��\r\n\r\n\r\n### ���ռ�����¶�̬ [2026-05-25����]\r\n\r\n| ��̬���� | ����ժҪ | Ӱ�췶Χ |\r\n|---------|---------|---------|\r\n| ���ռ�� | 2026��4�³����¹棺���������ϵ���/��ȼ/��ˮ/�����ȸ����� | �������������������ָ��Ƿ�Χ���⸶��׼ |\r\n| ���ռ�� | ҽ���գ����ز�Ŀ¼������62�֣�33��ȡ���������� | �������������������ָ��Ƿ�Χ���⸶��׼ |\r\n| ���ռ�� | ��ҵҽ������ȷ����֢��׼��CAR-T/����������/��ҩ���뱣��(80%-100%) | �������������������ָ��Ƿ�Χ���⸶��׼ |\r\n\r\n> **���ݽ�ֹ**: 2026-05-25 | ��Դ�����ҽ��ڼල�����ܾ֡�����Q1��������ҵ������Ϣ\r\n> **����**: ���϶�̬���ο��������Թٷ����·���Ϊ׼\r\n\r\n> **?? DISCLAIMER / ��������**\r\n> - **English:** This skill provides advisory templates, checklists, and decision-support frameworks ONLY. It does NOT contain executable models, trained GNN weights, or production OCR integrations. All accuracy figures (e.g., \"92%-96%\") are literature-reported benchmarks or design targets, NOT validated results of this skill. ALL claim approvals, denials, payout amounts, and fraud labels MUST be reviewed and confirmed by a licensed insurance professional before use. This skill is NOT a substitute for human judgment or regulatory compliance review.\r\n> - **���ģ�** ��Skill���ṩ��ѯģ�塢����嵥�;���֧�ֿ�ܣ�������ִ��ģ�͡���ѵ��GNNȨ�ػ�������OCR���ɡ�����׼ȷ�����ݣ���\"92%-96%\"�����������׻�׼�����Ŀ�꣬�DZ�Skillʵ���������������׼���ܸ����⸶����թ��ǩ��**���뾭��֤����רҵ��ʿ���ȷ�Ϻ󷽿�ʹ��**����Skill��������˹��жϻ��ܺϹ���顣\r\n\r\n> **?? DATA SECURITY / ���ݰ�ȫ**\r\n> - Medical invoices, diagnosis records, and claimant data are sensitive personal information under China's Personal Information Protection Law (PIPL). Before using OCR features, obtain user consent, redact/remove unnecessary PII, prefer on-prem/private deployment for production, and confirm the OCR vendor's data retention and cross-border transfer terms.\r\n> - API keys and credentials MUST be stored in environment variables or a secret manager. Never hardcode keys in production systems.\r\n> - **English:** This skill provides advisory templates, checklists, and decision-support frameworks ONLY. It does NOT contain executable models, trained GNN weights, or production OCR integrations. All accuracy figures (e.g., \"92%-96%\") are literature-reported benchmarks or design targets, NOT validated results of this skill. ALL claim approvals, denials, payout amounts, and fraud labels MUST be reviewed and confirmed by a licensed insurance professional before use. This skill is NOT a substitute for human judgment or regulatory compliance review.\r\n> - **���ģ�** ��Skill���ṩ��ѯģ�塢����嵥�;���֧�ֿ�ܣ�������ִ��ģ�͡���ѵ��GNNȨ�ػ�������OCR���ɡ�����׼ȷ�����ݣ���\"92%-96%\"�����������׻�׼�����Ŀ�꣬�DZ�Skillʵ���������������׼���ܸ����⸶����թ��ǩ��**���뾭��֤����רҵ��ʿ���ȷ�Ϻ󷽿�ʹ��**����Skill��������˹��жϻ��ܺϹ���顣\r\n\r\n---\r\n\r\n## Artifact Type / ��Ʒ����\r\n\r\n**This is a documentation-and-template skill.** It contains:\r\n- ? Workflow checklists and decision trees\r\n- ? Report templates and output formats\r\n- ? Reference architectures and integration guidance\r\n- ? Example Python code (requires your own API keys and data)\r\n\r\nIt does NOT contain:\r\n- ? Pre-trained ML/GNN models\r\n- ? Executable OCR or claims processing code\r\n- ? Bundled third-party API credentials\r\n\r\n---\r\n\r\n## Trigger Keywords / �����ؼ���\r\n\r\n**English Triggers:** insurance claims advisory, claims workflow, claim analysis, medical OCR guidance, insurance fraud assessment, claim liability review, policy clause analysis, anti-fraud checklist, insurance tech reference, claims report template\r\n\r\n**���Ĵ����ʣ�** ����������ѯ / ��������ָ�� / ������� / ҽ�Ʒ�Ʊʶ��ָ�� / ��������ο� / �����϶����� / ҽ�������� / �ؼ������� / �������� / ���������� / �������� / �Ʋ������� / ���ⷴ��թ / ��թ���ο� / ƭ��ʶ��ָ�� / �����زο� / ���������� / �������˵�� / ���Ϸ�Χ���� / �⸶�������� / ��Ʒ�ԱȲο� / ����ȶ�ָ�� / ��ͬ����ο�\r\n\r\n---\r\n\r\n## Core Capabilities / ������������ѯ��ܣ�\r\n\r\n### 1. Medical Receipt OCR �� Guidance Framework / ҽ��Ʊ��OCRʶ��ָ����ܣ�\r\n\r\n**֧�ֵ�Ʊ�����ͣ�����ȫ��������**\r\n\r\n| Receipt Type / Ʊ������ | Extracted Fields / ʶ������ | Insurance Types / �������� |\r\n|------------------------|-------------------|------------------|\r\n| ȫ��ͳһ���﷢Ʊ | ��Ʊ�š�ҽԺ������ϸ��Ŀ | ҽ���ա������� |\r\n| ȫ��ͳһסԺ��Ʊ | ��Ժ/��Ժ���ڡ��ܽ��Էѱ��� | ҽ���ա��ؼ��� |\r\n| ҽ�Ʒ�����ϸ�嵥 | ҩƷ��ϸ�������Ŀ�����ۡ����� | ҽ���� |\r\n| ҽ�����㵥 | ҽ���˻�֧�����Ը����������� | ҽ���� |\r\n| ��ԺС�� | ��ϡ�סԺ���������ƾ��� | �ؼ��ա����� |\r\n| ������ҳ | ��Ҫ��ϡ��������ơ�ICD���� | �ؼ��� |\r\n| ��鱨�浥 | Ӱ�񱨸桢������ | �ؼ��� |\r\n| ���ý��㵥 | ������ܼƽ�� | �Ʋ��ա������� |\r\n\r\n> **?? OCR Data Handling / OCR���ݴ�������**\r\n> - Only send necessary fields to OCR providers; redact/unnecessary PII beforehand.\r\n> - Confirm the OCR vendor's data retention policy (Prefer: no storage / auto-delete within 24h).\r\n> - For production use, prefer private on-prem OCR deployment to avoid third-party data transfer.\r\n> - **���ģ�** �����ͱ�Ҫ�ֶ���OCR�����̣���ǰ����/ɾ���DZ�Ҫ������Ϣ��ȷ��OCR��������������ԣ����ȣ�������/24Сʱ���Զ�ɾ������������������ʹ��˽�л����ز���OCR��������������ݴ��䡣\r\n\r\n**�ο������ܹ��������м��ɣ���**\r\n\r\n```text\r\nԭʼͼ��\r\n  ��\r\nͼ��Ԥ������ȥ��/��бУ��/��ֵ����\r\n  ��\r\nCNN������ȡ��ResNet50/EfficientNet������ ������ѵ��������Ʒ���API\r\n  ��\r\nRNN���н�ģ��BiLSTM��+ Attention����\r\n  ��\r\nCRF����� �� �ṹ���ı����\r\n  ��\r\n�ֶα�׼�� �� JSON/����ṹ�����\r\n```\r\n\r\n### 2. Liability Determination �� Advisory Framework / �����������棨��ѯ��ܣ�\r\n\r\n**��ѯ���������嵥�����˹�����ȷ�ϣ���**\r\n\r\n```text\r\n����1���ȴ��ڼ�飨�˹�ȷ�ϣ�\r\n  ���� �������� - ������Ч�� < �ȴ��� �� ����ܸ������˹�����\r\n\r\n����2������֢ɸ�飨�˹�ȷ�ϣ�\r\n  ���� ����֢��ƥ�� �� ������� �� ����ܸ�/�����⸶�����˹�����\r\n\r\n����3�������У�飨�˹�ȷ�ϣ�\r\n  ���� �ۼ��Ը���� < ����� �� �����ݲ��⸶�����˹�����\r\n\r\n����4����������˲飨�˹�ȷ�ϣ�\r\n  ���� �Ƕ��������Ϲ���ҽԺ����������� ��ʾȷ�ϣ����˹�����\r\n\r\n����5����������ƥ�䣨�˹�ȷ�ϣ�\r\n  ���� �������/����Ƿ��������Ϸ�Χ �� ����ȫ��/����/�ܸ������˹�����\r\n```\r\n\r\n> **?? IMPORTANT / ��Ҫ����**\r\n> The liability determination output is a **decision-support suggestion ONLY**. Final approval/denial MUST be made by an authorized human reviewer. This skill does NOT auto-approve any claim amount.\r\n> **���ģ�** �������**��Ϊ����֧�ֽ���**�����պ�׼/�ܸ�**��������Ȩ�˹����Ա����**����Skill�����κ�����������Զ�������\r\n\r\n### 3. Anti-Fraud Assessment �� Advisory Framework / ����թ��������ѯ��ܣ�\r\n\r\n**����թ����嵥����ѯ������**\r\n\r\n```text\r\n�����1������Ƶ���쳣\r\n  ���� ͬһ�����˶����ڶ�ξ��� �� ��ǣ������˹�����\r\n\r\n�����2��Ʊ����ʵ����֤\r\n  ���� ��Ʊ���ظ� / ҽԺ������ / ����쳣 �� ��ǣ������˹�����\r\n\r\n�����3���������ҩƥ����\r\n  ���� ����뿪��ҩƷ���Բ��� �� ��ǣ������˹�����\r\n\r\n�����4����ϵ�����쳣\r\n  ���� ͬһҽ��/ҽԺ���г����ڶ������� �� ��ǣ������˹�����\r\n```\r\n\r\n> **?? Anti-Fraud Data Governance / ����թ��������**\r\n> - Retention limit / �������ޣ�����թͼ�����ݽ������治���� 2 �꣬���Ǽ��Ҫ��ĸ��������ڡ�\r\n> - Access control / ���ʿ��ƣ�ͼ�ײ�ѯȨ�޽����Ÿ���Ȩ��թ����Ա����ֹ����Ȩ��Ա���ʡ�\r\n> - Data correction workflow / ���ݸ������̣���������Ȩ��������������ݣ������� 15 ���������ڴ�����\r\n> - Poisoning safeguard / ��Ⱦ�������°������ݽ���ͼ��ǰ���뾭�˹����ȷ�ϣ���ֹ������Ⱦ��\r\n\r\n### 4. Claims Report Templates / ���ⱨ��ģ��\r\n\r\n```markdown\r\n# ����������棨��ѯ�ݸ壩\r\n**����ʱ��**: YYYY-MM-DD HH:mm\r\n**�������**: CL-XXXXXXXX\r\n**�������**: [��������]\r\n**����״̬**: [��ѯ�ݸ� �� ���˹����]\r\n**��������**: ������ΪAI�������ɵ���ѯ�ݸ壬���н����뾭��֤����ʦ���ȷ�Ϻ󷽿���Ч��\r\n---\r\n## һ��Ʊ��ʶ�����������ο���\r\n## ���������϶������������ο���\r\n## �����⸶����ο��������ο���\r\n## �ġ�����թ���������������ο���\r\n## �塢������һ���ж������˹�ȷ�ϣ�\r\n```\r\n\r\n---\r\n\r\n## Compliance & Human Review / �Ϲ����˹����Ҫ��\r\n\r\n| Compliance Item / �Ϲ��� | Regulatory Basis / ������� | Human Review Requirement / �˹����Ҫ�� |\r\n|--------------------|--------------------|----------------------|\r\n| ����ʱЧ | �����շ�����23�� | �˶�����뾭�˹�ȷ�Ϻ󷢳� |\r\n| ���������� | ��������취 | ȱʧ�����б����˹�����ȷ�� |\r\n| ����թ�Ϲ� | ����������թ�����취��2024 | ��թ����뾭�˹�����ȷ�� |\r\n| ���ݰ�ȫ | ��������Ϣ�������� | ҽ���������������뾭�˹���� |\r\n| �ʽ�ȫ | ��ϴǮ�涨 | ����������˹����� + �������� |\r\n\r\n**ALL outputs of this skill are drafts requiring licensed professional review. / ��Skill���������Ϊ�ݸ壬�뾭��֤רҵ��ʿ��ˡ�**\r\n\r\n---\r\n\r\n## Output Format / �����ʽ�淶\r\n\r\nAll outputs must include the following disclaimer:\r\n\r\n```markdown\r\n> ?? **�������� / Disclaimer**\r\n> �����ΪAI������ѯ�ݸ壬��������������ܸ����ۡ��⸶����թ��ǩ\r\n> �뾭����֤��������ʦ�����ȷ�Ϻ󷽿���Ч��\r\n> This is an AI-assisted draft. All claim decisions must be reviewed by a\r\n> licensed insurance adjuster before taking effect.\r\n```\r\n\r\n---\r\n\r\n## References / �ο��ļ�\r\n\r\n| File / �ļ� | Content / ����˵�� |\r\n|------|---------|\r\n| `references/claims_ocr_tech.md` | OCR�����ܹ��ο� + 4�ҷ����̶Ա� + Pythonʾ�����루����������API Key�� |\r\n| `references/claims_liability_engine.md` | �������ο� + ����ѧϰģ�Ͳο� + 3�ҹ�˾ʵ���ο� |\r\n| `references/claims_report_templates.md` | ����ģ�� + 7������֪ͨ��ģ��ο� |\r\n\r\n> **?? Reference files contain example code only. You must:**\r\n> - Provide your own API keys and store them in environment variables\r\n> - Provide your own training data and models\r\n> - Ensure human review of all outputs before use\r\n> - **���ģ�** �ο��ļ�����ʾ�����룬�����룺�����ṩAPI��Կ�����뻷������������׼��ѵ�����ݺ�ģ�ͣ�ȷ������������˹���˺󷽿�ʹ�á�\r\n\r\n---\r\n\r\n### ����թ�����㷨��ϸ����2026�����棩\r\n\r\n| ��թ���� | �������ӣ�0-10�֣� | ������Դ | ������ֵ | Ӧ������ |\r\n|---------|----------------|---------|---------|---------|\r\n| **Ʊ���쳣** | ��Ʊ���ظ�(3��)+����쳣(2��)+ҽԺ������(5��) | OCR+ҽ�����ݿ� | ��5��ת�˹� | Ʊ��ԭ������ |\r\n| **����Ƶ��** | 7���ڡ�3��(3��)+ͬһҽԺ(2��)+�������(3��) | ������ʷ�� | ��5��Ԥ�� | ҽ�Ƽ�¼���� |\r\n| **��ϵ����** | ͬһҽ��(2��)+ͬһIP(3��)+������ϵ(3��) | ��ϵͼ�����ݿ� | ��5��Ԥ�� | ��ϵ�������� |\r\n| **����쳣** | ����>5��(2��)+�����ۼ�>20��(3��)+�⸶��>90%(3��) | ����ϵͳ | ��5��Ԥ�� | �ʽ�����׷�� |\r\n| **ʱ���쳣** | ��ҹ����(23-5��)(2��)+�ڼ���ͻ��(2��)+��Ͷ��ʱ���ص�(3��) | ����ϵͳ��־ | ��5��Ԥ�� | ʱ�����ؽ� |\r\n\r\n**�����㷨��Pythonʾ����**��\r\n```python\r\ndef fraud_score(claim):\r\n    score = 0\r\n    # Ʊ���쳣\r\n    if claim.invoice_id in seen_invoice_ids: score += 3\r\n    if claim.amount > 50000: score += 2\r\n    if claim.hospital not in valid_hospitals: score += 5\r\n    # ����Ƶ��\r\n    recent_claims = [c for c in history if (claim.date - c.date).days < 7]\r\n    if len(recent_claims) >= 3: score += 3\r\n    # ��ϵ���磨��ͼ���ݿ⣩\r\n    related = graph_db.query(f\"MATCH (p1)-[:CLAIM]->(c) WHERE p1.id='{claim.claimant_id}' RETURN count(c)\")\r\n    if related > 5: score += 5\r\n    return min(score, 10)  # �ⶥ10��\r\n```\r\n\r\n**ͼ���ݿⷴ��թ�ܹ���2026�Ƽ���**��\r\n- **Neo4j**����������ѣ�֧�ָ��ӹ�ϵ�������ʺ����ͱ��չ�˾��<1000������/�꣩\r\n- **TigerGraph**���ֲ�ʽԭ��ͼ�⣬֧��ʵʱ��ȱ������ʺϴ��ͱ��չ�˾��>1000������/�꣩\r\n- **Amazon Neptune**��ȫ�йܣ�����AWS��̬���ʺ���������ҵ\r\n\r\n---\r\n\r\n\r\n*GitHub: https://github.com/gechengling/insurance-claims-intelligence*\r\n## 核心工作流程（Dianjin融合版）\n\n### 第一步：理赔案件智能初筛（Dianjin精髓：自动化分诊）\n\n| 筛查维度 | 判断标准 | 处理路径 |\n|---------|---------|---------|\n| **案件类型** | 车险/财产险/健康险/责任险 | 分流到专业理赔师 |\n| **损失金额** | <1万/1-5万/5-10万/>10万 | 自动/人工/专家/高管审批 |\n| **复杂程度** | 单方事故/多方事故/人伤/物损 | 简赔/标准/复杂/重案 |\n| **欺诈风险** | 高风险标记/历史欺诈记录 | 欺诈调查队列 |\n\n**智能分诊输出**：\n```\n【案件分诊结果】\n- 案件编号：[XXX]\n- 分诊结论：[正常案件/需调查/重大案件]\n- 推荐处理人：[初级理赔师/资深专家/调查员]\n- 预计处理时长：[X]小时\n- 风险标记：[低/中/高]\n```\n\n---\n\n### 第二步：损失评估与定损（Dianjin精髓：AI辅助定损）\n\n#### 2.1 车辆损失智能评估\n| 评估项目 | AI识别内容 | 人工复核点 |\n|---------|---------|---------|\n| **外观件** | 损伤部位、损伤程度、维修方案 | 扩大损失风险 |\n| **结构件** | 变形程度、更换/维修判定 | 安全性评估 |\n| **电气件** | 故障代码、模块状态 | 隐性故障排查 |\n| **内饰件** | 污损程度、清洗/更换 | 贬值评估 |\n\n#### 2.2 财产损失评估\n```\n【财产损失清单】\n- 固定资产损失：[项目/数量/原值/折旧/现值]\n- 存货损失：[品类/数量/成本/残值]\n- 营业中断损失：[停业天数/日均利润/预期利润]\n- 施救费用：[施救项目/金额/合理性]\n```\n\n#### 2.3 人伤医疗评估\n| 评估维度 | AI分析内容 | 争议点 |\n|---------|---------|---------|\n| **伤情诊断** |  ICD-10编码、伤情等级 | 伤情与事故关联性 |\n| **治疗合理性** | 用药/检查/治疗必要性 | 过度医疗识别 |\n| **伤残评定** | 伤残等级、赔付标准 | 评定依据充分性 |\n| **三期评定** | 误工期/护理期/营养期 | 评定期限合理性 |\n\n---\n\n### 第三步：欺诈风险识别（Dianjin精髓：欺诈模式识别）\n\n#### 3.1 欺诈风险评分模型\n| 风险维度 | 评分标准 | 权重 |\n|---------|---------|------|\n| **报案异常** | 报案时间/地点/天气/事故形态 | 20% |\n| **历史记录** | 过往出险/理赔/投诉记录 | 25% |\n| **行为模式** | 修车厂/医院/律师关联性 | 20% |\n| **损失异常** | 损失程度/维修金额/市场价偏差 | 20% |\n| **文档异常** | 发票/病历/定损单真实性 | 15% |\n\n**欺诈风险等级**：\n- 低风险（0-20分）：正常处理\n- 中风险（21-50分）：加强复核\n- 高风险（51-75分）：启动调查\n- 极高风险（76-100分）：移交反欺诈\n\n#### 3.2 常见欺诈模式库\n| 欺诈类型 | 识别特征 | 调查方向 |\n|---------|---------|---------|\n| **摆放现场** | 无刹车痕/车身损旧/位置异常 | 监控/证人/车辆历史 |\n| **虚高定损** | 配件价高/工时虚高/残值低估 | 市场比价/修复可行性 |\n| **医疗欺诈** | 过度检查/虚假病历/挂床住院 | 病历审核/医院走访 |\n| **关系网络欺诈** | 修车厂/医生/律师关联 | 关系网络图谱 |\n\n---\n\n### 第四步：理赔方案制定（Dianjin精髓：最优赔付方案）\n\n#### 4.1 赔付方案决策树\n```\n【赔付方案决策】\n├─ 责任明确？\n│  ├─ 是 → 损失确定？\n│  │      ├─ 是 → 直接赔付（金额≤X万）\n│  │      └─ 否 → 补充定损/调查\n│  └─ 否 → 责任争议处理\n│           ├─ 协商\n│           ├─ 调解\n│           └─ 诉讼\n└─ 欺诈风险？\n   ├─ 低风险 → 正常赔付\n   ├─ 中风险 → 加强审核后赔付\n   └─ 高风险 → 调查后决定\n```\n\n#### 4.2 赔付金额计算模板\n| 计算项目 | 计算公式 | 金额 |\n|---------|---------|------|\n| **总损失** | 财产损失+人身伤害+其他损失 | [X]元 |\n| **保险责任** | 总损失 × 责任比例 | [X]元 |\n| **免赔额** | 绝对免赔/相对免赔 | -[X]元 |\n| **残值回收** | 残值作价回收 | -[X]元 |\n| **最终赔付** | 保险责任-免赔额-残值 | **[X]元** |\n\n---\n\n### 第五步：客户沟通与结案（Dianjin精髓：客户体验管理）\n\n#### 5.1 客户沟通话术库\n| 场景 | 标准话术 | 注意事项 |\n|------|---------|---------|\n| **接报案** | \"您好，我是XX保险理赔员XXX...\" | 同理心/专业度/及时性 |\n| **现场查勘** | \"请您配合我们核实事故情况...\" | 尊重/耐心/不预设立场 |\n| **定损沟通** | \"根据定损结果，赔付金额为...\" | 透明/有理有据/留痕 |\n| **拒赔沟通** | \"很抱歉，根据条款约定...\" | 依法/依规/人性化 |\n| **结案回访** | \"请问您对本次理赔是否满意？\" | 真诚/改进/感谢 |\n\n#### 5.2 结案文档模板\n```markdown\n# 理赔结案报告\n\n## 案件基本信息\n- 保单号：[XXX]\n- 报案号：[XXX]\n- 被保险人：[XXX]\n- 事故日期：[XXX]\n- 结案日期：[XXX]\n\n## 事故概况\n[事故经过/责任认定/损失情况]\n\n## 定损结果\n[损失项目/定损金额/残值处理]\n\n## 赔付决定\n- 赔付金额：[X]元\n- 赔付依据：[条款/计算/核赔]\n- 支付方式：[银行转账/现金]\n\n## 客户反馈\n[满意度/投诉/建议]\n\n## 经验教训\n[改进点/风险提示/案例价值]\n```\n\n---\n\n## 合规约束与审计规则（Dianjin精髓）\n\n1. **依法合规**：严格遵守《保险法》《理赔管理办法》等法规\n2. **客观公正**：定损客观、核赔公正，不受外部干扰\n3. **及时高效**：接报案24h内联系，复杂案件60天内结案\n4. **客户知情**：赔付依据、计算方式、拒赔理由必须告知客户\n5. **数据安全**：客户信息、案件资料严格保密，不得泄露\n6. **审计留痕**：所有操作留痕，可追溯、可审计\n\n---\n\n## 测试用例（Dianjin精髓）\n\n**测试场景1：车险正常理赔**\n- 输入：报案号XXX，车损XXX元，责任明确\n- 预期输出：定损报告、赔付计算、结案文档\n- 通过标准：定损合理、计算准确、文档完整\n\n**测试场景2：疑似欺诈案件**\n- 输入：报案异常、历史多次出险\n- 预期输出：欺诈风险评分、调查建议、证据收集清单\n- 通过标准：评分模型正确、调查方向明确\n\n**测试场景3：人伤医疗理赔**\n- 输入：伤者XXX，医疗费等XX万\n- 预期输出：医疗评估、伤残评定、赔付计算\n- 通过标准：ICD-10准确、治疗合理性判断正确\n\n---\n\n## 关联技能（Dianjin精髓）\n\n- `insurance-anti-fraud`：理赔反欺诈→反欺诈调查联动\n- `insurance-actuarial-cn`：理赔数据→经验费率更新输入\n- `insurance-agent-trainer`：理赔案例→培训素材来源\n- `finance-ai-strategy`：理赔AI→保险AI战略协同\n\n---\n\n*本技能融合阿里点金（Qwen Dianjin）金融AI精髓，专注理赔智能化、欺诈识别、客户体验提升。*\n\nFile v5.0.2:README.md\n\n# Insurance Claims Intelligence Expert / 保险行业智能理赔专家\r\n\r\n> **⚠️ DISCLAIMER / 免责声明**\r\n> - **English:** This is an **advisory and template-only skill**. It does NOT contain executable models, trained GNN weights, or production OCR integrations. All accuracy figures (e.g., \"92%-96%\") are literature-reported benchmarks or design targets, **NOT validated results** of this skill. ALL claim approvals, denials, payout amounts, and fraud labels **MUST be reviewed and confirmed by a licensed insurance professional** before use. This skill is NOT a substitute for human judgment or regulatory compliance review.\r\n> - **中文：** 本Skill**仅为咨询模板和参考框架**，不含可执行模型、已训练GNN权重或生产级OCR集成。所有准确率数据（如\"92%-96%\"）均来自文献基准或设计目标，**非本Skill实测结果**。所有理赔核准、拒付、赔付金额及欺诈标签，**必须经持证保险专业人士审核确认后方可使用**。本Skill不可替代人工判断或监管合规审查。\r\n\r\n> **🔒 DATA SECURITY NOTICE / 数据安全提醒**\r\n> - Medical invoices and claimant data are sensitive personal information. Before using OCR features, obtain user consent, redact unnecessary PII, and prefer on-prem/private deployment.\r\n> - API keys MUST be stored in environment variables or a secret manager. Never hardcode keys.\r\n> - **中文：** 医疗发票和理赔申请人数据属于敏感个人信息。使用OCR功能前，须获得用户同意，脱敏非必要个人信息，优先使用本地私有化部署。API密钥必须存入环境变量或密钥管理器，禁止硬编码。\r\n\r\n---\r\n\r\n## ✨ What This Skill Provides / 本Skill提供的内容\r\n\r\n| Type / 类型 | Description / 说明 |\r\n|-------------|---------------------|\r\n| 📋 Workflow checklists / 流程检查清单 | Step-by-step claims review checklists for human reviewers |\r\n| 📄 Report templates / 报告模板 | Standardized output formats for claims analysis reports |\r\n| 🏗️ Reference architectures / 参考架构 | Guidance on OCR integration, rules engines, and GNN design |\r\n| 💡 Example code / 示例代码 | Python examples (require your own API keys and data) |\r\n| 📚 Company practices reference / 公司实践参考 | Summaries of industry best practices (advisory only) |\r\n\r\n**This skill does NOT provide:** executable models, pre-trained weights, bundled API credentials, or automated claim approval.\r\n\r\n---\r\n\r\n## Core Features / 核心功能（咨询级）\r\n\r\n### 1. Medical Receipt OCR — Integration Guidance / 医疗票据OCR（集成指导）\r\n\r\n**Supported document types / 支持票据类型（8类）：**\r\n门诊发票 / 住院发票 / 医疗费用明细清单 / 医保结算单 / 出院小结 / 病历首页 / 检查报告单 / 费用结算单\r\n\r\n**Integration options / 集成方案参考：**\r\n- Baidu AI OCR / 百度AI开放平台\r\n- Tencent Cloud OCR / 腾讯云OCR\r\n- Ali Cloud OCR / 阿里云OCR\r\n- Infologic OCR / 合合信息OCR\r\n\r\n> ⚠️ **Data Handling / 数据处理：** Only send necessary fields. Redact/unnecessary PII beforehand. Confirm vendor's data retention policy.\r\n> **中文：** 仅发送必要字段，事前脱敏非必要个人信息，确认服务商数据留存策略。\r\n\r\n### 2. Liability Determination — Advisory Checklist / 理赔判责（咨询检查清单）\r\n\r\nProvides structured checklists for human reviewers:\r\n- ✅ Waiting period check / 等待期检查\r\n- ✅ Pre-existing condition screen / 既往症筛查\r\n- ✅ Deductible verification / 免赔额校验\r\n- ✅ Hospital level verification / 就诊机构核查\r\n- ✅ Policy coverage match / 险种责任匹配\r\n\r\n> ⚠️ **All results are suggestions only. Final decisions MUST be made by authorized human reviewers.**\r\n> **中文：** 所有结果仅为建议，最终决定必须由授权人工审核员作出。\r\n\r\n### 3. Anti-Fraud Assessment — Checklist / 反欺诈评估（检查清单）\r\n\r\nStructured red-flag checklist for fraud investigation:\r\n- 🚩 Unusual visit frequency / 就诊频率异常\r\n- 🚩 Invoice authenticity verification / 票据真实性验证\r\n- 🚩 Diagnosis-medication mismatch / 诊断与用药不匹配\r\n- 🚩 Provider-case network anomalies / 医疗机构-案件网络异常\r\n\r\n**Data governance requirements / 数据治理要求：**\r\n- Retention limit / 留存期限：≤ 2 years (or per regulatory requirement) / 不超过2年（或监管要求期限）\r\n- Access control / 访问控制：Authorized fraud investigators only / 仅授权欺诈调查员可访问\r\n- Correction workflow / 更正流程：Data subjects have the right to request correction / 数据主体有权请求更正\r\n\r\n### 4. Claims Report Templates / 理赔报告模板\r\n\r\nStandardized templates for 7 insurance types:\r\n- Health insurance / 医疗险\r\n- Critical illness insurance / 重疾险\r\n- Life insurance / 寿险\r\n- Accident insurance / 意外险\r\n- Auto insurance / 车险\r\n- Property insurance / 财产险\r\n- Group insurance / 团险\r\n\r\nAll templates include the required disclaimer: \"This is an AI-assisted draft requiring licensed professional review.\"\r\n\r\n---\r\n\r\n## 🚀 Quick Start / 快速上手\r\n\r\n```bash\r\n# Install this skill (installs advisory templates only)\r\nnpx clawhub install insurance-claims-intelligence\r\n\r\n# Use in WorkBuddy (advisory mode only)\r\n/insurance-claims-intelligence \"Generate a claims review checklist for this outpatient case\"\r\n/insurance-claims-intelligence \"Create a fraud risk assessment template for these 5 cases\"\r\n```\r\n\r\n> ⚠️ **All outputs are drafts. Human review is mandatory.**\r\n> **中文：** 所有输出均为草稿，人工审核是强制要求。\r\n\r\n---\r\n\r\n## 📖 What's Included / 包含内容\r\n\r\n| File / 文件 | Content / 内容说明 |\r\n|-------------|---------------------|\r\n| `SKILL.md` | Full skill definition, trigger keywords, advisory workflows |\r\n| `references/claims_ocr_tech.md` | OCR integration guidance + provider comparison + example Python code (API key required) |\r\n| `references/claims_liability_engine.md` | Liability checklist + model reference + 3 company practices (advisory) |\r\n| `references/claims_report_templates.md` | Report templates + 7 insurance type notice templates |\r\n\r\n---\r\n\r\n## Provenance / 来源说明\r\n\r\n- **Author / 作者：** @gechengling\r\n- **Skill type / Skill类型：** Advisory templates and reference frameworks only / 仅含咨询模板和参考框架\r\n- **Contains executable code:** NO / 不含可执行代码\r\n- **Contains pre-trained models:** NO / 不含预训练模型\r\n- **Requires API credentials:** YES — you must provide your own OCR/LLM API keys / 需要您自行提供OCR/LLM API密钥\r\n- **License / 开源协议：** MIT-0\r\n\r\n---\r\n\r\n## Required Human Review / 强制人工审核要求\r\n\r\n| Action / 操作 | Human Review Required? / 需人工审核？ |\r\n|---------------|----------------------------------------|\r\n| Claim approval / 理赔核准 | ✅ MANDATORY / 强制 |\r\n| Claim denial / 理赔拒付 | ✅ MANDAT\n\nArchive v5.0.1: 7 files, 33551 bytes\n\nFiles: README.md (7594b), references/claims_liability_engine.md (18889b), references/claims_ocr_tech.md (14039b), references/claims_report_templates.md (11395b), skill-card.md (3107b), SKILL.md (19097b), _meta.json (148b)\n\nArchive v5.0.0: 7 files, 32458 bytes\n\nFiles: README.md (7594b), references/claims_liability_engine.md (18889b), references/claims_ocr_tech.md (14039b), references/claims_report_templates.md (11395b), skill-card.md (3105b), SKILL.md (16529b), _meta.json (148b)\n\nArchive v3.0.2: 7 files, 29963 bytes\n\nFiles: README.md (7594b), references/claims_liability_engine.md (18889b), references/claims_ocr_tech.md (14039b), references/claims_report_templates.md (11395b), skill-card.md (2779b), SKILL.md (12112b), _meta.json (148b)\n\nArchive v3.0.1: 6 files, 28200 bytes\n\nFiles: README.md (7594b), references/claims_liability_engine.md (18889b), references/claims_ocr_tech.md (14039b), references/claims_report_templates.md (11395b), SKILL.md (12731b), _meta.json (148b)\n\nArchive v2.0.0: 6 files, 27747 bytes\n\nFiles: README.md (7594b), references/claims_liability_engine.md (18889b), references/claims_ocr_tech.md (14039b), references/claims_report_templates.md (11395b), SKILL.md (11748b), _meta.json (148b)","readmeExcerpt":"Skill: Insurance Claims Intelligence Owner: gechengling Summary: 提供多模态医疗票据OCR识别、智能判责、反欺诈检测和全险种覆盖的保险理赔智能分析与自动化支持。 Tags: advisory-only:1.2.0, ai-agent:1.1.0, anti-fraud:5.0.4, anti-fraud-checklist:1.2.0, banking:5.0.0, bilingual:1.1.0, china-insurance:1.2.0, chinese-market:1.0.0, claims:5.0.4, claims-advisory:1.2.0, claims-processing:1.1.0, compliance:1.2.0, decision-support:1.2.0, deepseek:1.1.0, dianjin:5.0.0, financ","codeSnippets":[],"executableExamples":[{"language":"text","snippet":"原始图像\n  ↓\n图像预处理（去噪/倾斜校正/二值化）\n  ↓\nCNN特征提取（ResNet50/EfficientNet）—— 需自行训练或调用云服务API\n  ↓\nRNN序列建模（BiLSTM）+ Attention机制\n  ↓\nCRF层解码 → 结构化文本输出\n  ↓\n字段标准化 → JSON/表格结构化结果"},{"language":"text","snippet":"规则1：等待期检查（人工确认）\n  └─ 出险日期 - 保单生效日 < 等待期 → 建议拒付，需人工复核\n\n规则2：既往症筛查（人工确认）\n  └─ 既往症库匹配 → 责任免除 → 建议拒付/比例赔付，需人工复核\n\n规则3：免赔额校验（人工确认）\n  └─ 累计自付金额 < 免赔额 → 建议暂不赔付，需人工复核\n\n规则4：就诊机构核查（人工确认）\n  └─ 非二级及以上公立医院（需视条款）→ 提示确认，需人工复核\n\n规则5：险种责任匹配（人工确认）\n  └─ 就诊科室/诊断是否符合条款保障范围 → 建议全额/比例/拒付，需人工复核\n\n规则6：如实告知/健康告知核查（人工确认）\n  └─ 投保前未如实告知既往症/健康状况 → 依《保险法》第十六条评估解除合同权与拒赔风险，需人工复核\n\n规则7：保险利益与受益人核验（人工确认）\n  └─ 索赔申请人是否具备保险利益、受益人指定是否有效 → 身份与关系证明核验，需人工复核\n\n规则8：事故性质与免责比对（人工确认）\n  └─ 出险原因是否落入责任免除（如违法犯罪、酒驾、战争等）→ 命中免责建议拒付，需人工复核"},{"language":"text","snippet":"检查项1：就诊频率异常\n  └─ 同一被保人短期内多次就诊 → 标记，建议人工调查\n\n检查项2：票据真实性验证\n  └─ 发票号重复 / 医院不存在 / 金额异常 → 标记，建议人工调查\n\n检查项3：诊断与用药匹配性\n  └─ 诊断与开具药品明显不符 → 标记，建议人工调查\n\n检查项4：关系网络异常\n  └─ 同一医生/医院集中出现在多起理赔 → 标记，建议人工调查\n\n检查项5：时间-地理冲突\n  └─ 同一被保人短时间内异地（甚至跨国）连续就诊/出险 → 标记，建议人工调查\n\n检查项6：团伙/中介特征\n  └─ 多起理赔共享同一代理人、同一联系电话或同一银行账户 → 标记，建议人工调查\n\n检查项7：损失与事实背离\n  └─ 申报损失金额显著高于同类案件均值、缺乏第三方佐证 → 标记，建议人工调查"},{"language":"markdown","snippet":"# 理赔分析报告（咨询草稿）\n**生成时间**: YYYY-MM-DD HH:mm\n**案件编号**: CL-XXXXXXXX\n**险种类别**: [险种名称]\n**处理状态**: [咨询草稿 — 需人工审核]\n**免责声明**: 本报告为AI辅助生成的咨询草稿，所有结论须经持证理赔师审核确认后方可生效。\n---\n## 一、票据识别结果（仅供参考）\n## 二、责任认定分析（仅供参考）\n## 三、赔付计算参考（仅供参考）\n## 四、反欺诈风险评估（仅供参考）\n## 五、建议下一步行动（需人工确认）"},{"language":"markdown","snippet":"# 理赔分析报告（咨询草稿）\n**生成时间**: 2026-09-24 10:20\n**案件编号**: CL-20260924-0087\n**险种类别**: 百万医疗险（含院外特药）\n**处理状态**: 咨询草稿 — 需人工审核\n---\n## 一、票据识别结果（仅供参考）\n- 住院发票 1 张：总金额 48,260.00 元，医保支付 26,140.00 元，自付 22,120.00 元（置信度 0.97，已建议人工复核）\n- 出院小结 1 份：入院 2026-08-11，出院 2026-08-19，共 8 天，主诊断编码 C50.9\n## 二、责任认定分析（仅供参考）\n- 等待期：已过（生效 2025-01-01，等待期 30 天）→ 未命中拒付情形\n- 就诊机构：三级甲等公立医院，符合条款约定\n- 免责比对：未见条款列明免责情形\n## 三、赔付计算参考（仅供参考）\n- 免赔额 10,000.00 元，本次可计入自付 22,120.00 元\n- 参考赔付区间 =（22,120.00 − 10,000.00）× 100% = 12,120.00 元\n- 院外特药 3,860.00 元需单独核对药品清单与特药目录\n## 四、反欺诈风险评估（仅供参考）\n- 风险等级：低（观察）。无可疑点叠加，建议常规核赔并留痕\n## 五、建议下一步行动（需人工确认）\n- 由持证理赔师复核发票原件与医保结算单勾稽关系\n- 确认院外特药是否在指定药店及目录内"},{"language":"markdown","snippet":"> ⚠️ **免责声明 / Disclaimer**\n> 本输出为AI辅助咨询草稿，所有理赔决定、拒付结论、赔付金额及欺诈标签\n> 须经【持证保险理赔师】审核确认后方可生效。\n> This is an AI-assisted draft. All claim decisions must be reviewed by a\n> licensed insurance adjuster before taking effect."}],"parameters":null,"dependencies":[],"permissions":[],"extractedFiles":[{"path":"SKILL.md","content":"---\nname: Insurance Claims Intelligence Expert\ndescription: Advisory skill for insurance claims processing workflows — provides templates, checklists, and decision-support frameworks for medical OCR, liability determination, anti-fraud assessment, and claims reporting. Human review required for all claim decisions. Keywords: insurance claims, claims advisory, medical OCR, anti-fraud, insurance tech, China insurance, decision support, 智能理赔, 理赔风控, 医疗单据识别, 责任认定, 理赔报告, 秒赔, 理赔决策, 医疗险理赔, 重疾理赔, 车险理赔.\nslug: insurance-claims-intelligence\nversion: 5.0.5\n\ncapabilities:\n  - educational-reference\n  - advisory-only\n  - requires-human-review\n  - illustrative-code-samples\n---\n\n# Insurance Claims Intelligence Expert / 保险行业智能理赔专家\n\n> **⚠️ SECURITY NOTICE / 安全声明**\n> - **Type:** Educational reference / analytical framework ONLY\n> - **No runnable package is bundled:** the architecture sketches and short code\n  fragments below are illustrative reference material for you to adapt in your own\n  environment; this skill ships no installer, service, or executable payload\n> - **No persistent storage, network calls, background execution, or credential collection**\n> - **All outputs are for reference only and require human review before real-world application**\n> - **This skill does NOT provide financial, legal, or insurance advice**\n> - **Users must exercise their own judgment and consult qualified professionals**\n\n\n\n> **⚠️ DISCLAIMER / 免责声明**\n> - **English:** This skill provides advisory templates, checklists, and decision-support frameworks ONLY. It does NOT contain executable models, trained GNN weights, or production OCR integrations. All accuracy figures (e.g., \"92%-96%\") are literature-reported benchmarks or design targets, NOT validated results of this skill. ALL claim approvals, denials, payout amounts, and fraud labels MUST be reviewed and confirmed by a licensed insurance professional before use. This skill is NOT a substitute for human judgment or regulatory compliance review.\n> - **中文：** 本Skill仅提供咨询模板、检查清单和决策支持框架，不含可执行模型、已训练GNN权重或生产级OCR集成。所有准确率数据（如\"92%-96%\"）均来自文献基准或设计目标，非本Skill实测结果。所有理赔核准、拒付、赔付金额及欺诈标签，**必须经持证保险专业人士审核确认后方可使用**。本Skill不可替代人工判断或监管合规审查。\n\n> **🔒 数据最小化前置声明 / Data Minimisation (apply before any OCR or analysis step)**\n> 1. 先问“这一条数据是否必需”：非理赔必需的字段（如完整身份证号、无关病史、家庭成员信息）一律不采集、不粘贴、不上传。\n> 2. 优先使用脱敏副本：姓名、证件号、联系电话默认以掩码形式处理（如 `张*`、`310***********1234`）。\n> 3. 单次任务单次授权：明确本次处理的用途与范围，任务结束后删除临时文件与对话中的原始影像描述。\n> 4. 任何需要写入文件或对外发送的结果，先在对话中完整展示给用户预览，经用户明确确认后再落盘/发送。\n\n> **🔒 DATA SECURITY / 数据安全**\n> - Medical invoices, diagnosis records, and claimant data are sensitive personal information under China's Personal Information Protection Law (PIPL). Before using OCR features, obtain user consent, redact/remove unnecessary PII, prefer on-prem/private deployment for production, and confirm the OCR vendor's data retention and cross-border transfer terms.\n> - API keys and credentials MUST be stored in environment variables or a secret manager. Never hardcode keys in production system"},{"path":"_meta.json","content":"{\n  \"ownerId\": \"kn74e704j3ygjcygnpf02rdvd185js13\",\n  \"slug\": \"insurance-claims-intelligence\",\n  \"version\": \"5.0.5\",\n  \"publishedAt\": 1790226938910\n}"},{"path":"skill-card.md","content":"## Description:\n\nAdvisory skill for insurance claims processing workflows that provides templates, checklists, and decision-support frameworks for medical OCR, liability determination, anti-fraud assessment, and claims reporting.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[gechengling](https://clawhub.ai/user/gechengling)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nExternal insurance operations teams, claims reviewers, and developers use this skill to draft claims workflow guidance, OCR review checklists, liability review notes, anti-fraud triage prompts, and claims report templates. All claim decisions, payout amounts, denial language, and fraud labels require qualified human review.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Claims guidance may affect sensitive insurance outcomes if used as a final decision.\n\nMitigation: Use outputs only as drafts and require a qualified insurance professional to review all claim decisions, payout amounts, denial language, and fraud labels before use.\n\nRisk: Claims workflows may involve personal, medical, or identifying data.\n\nMitigation: Do not paste unnecessary personal or medical data; redact identifiers where possible and follow data minimization practices before OCR or analysis.\n\nRisk: Regulatory references and compliance practices can change over time.\n\nMitigation: Verify regulatory references against current official sources before relying on them in claims handling or reporting.\n\n## Reference(s):\n\n- [ClawHub release page](https://clawhub.ai/gechengling/skills/insurance-claims-intelligence)\n- [Skill source artifact](artifact/SKILL.md)\n\n## Skill Output:\n\n**Output Type(s):** [Text, Markdown, Code, Configuration, Guidance]\n\n**Output Format:** [Markdown drafts with checklists, tables, report templates, and illustrative code snippets]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [Outputs are advisory drafts and must include a human-review disclaimer before operational use.]\n\n## Skill Version(s):\n\n5.0.5 (source: frontmatter, release metadata, changelog)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment."}],"languages":[],"docsSourceLabel":"CLAWHUB","editorialOverview":"提供多模态医疗票据OCR识别、智能判责、反欺诈检测和全险种覆盖的保险理赔智能分析与自动化支持。 Skill: Insurance Claims Intelligence Owner: gechengling Summary: 提供多模态医疗票据OCR识别、智能判责、反欺诈检测和全险种覆盖的保险理赔智能分析与自动化支持。 Tags: advisory-only:1.2.0, ai-agent:1.1.0, anti-fraud:5.0.4, anti-fraud-checklist:1.2.0, banking:5.0.0, bilingual:1.1.0, china-insurance:1.2.0, chinese-market:1.0.0, claims:5.0.4, claims-advisory:1.2.0, claims-processing:1.1.0, compliance:1.2.0, decision-support:1.2.0, deepseek:1.1.0, dianjin:5.0.0, financ","editorialQuality":{"score":100,"threshold":65,"status":"ready","wordCount":1198,"uniquenessScore":53,"reasons":[]}},"media":{"evidence":{"source":"no-media","verified":false,"confidence":"low","updatedAt":"2026-10-09T23:01:57.293Z","emptyReason":"No screenshots, media assets, or demo links are available."},"primaryImageUrl":null,"mediaAssetCount":0,"assets":[],"demoUrl":null},"ownerResources":{"evidence":{"source":"unclaimed","verified":false,"confidence":"low","updatedAt":"2026-10-09T23:01:57.293Z","emptyReason":"This page has not been claimed by the agent owner."},"hasCustomPage":false,"customPageUpdatedAt":null,"customLinks":[],"structuredLinks":{"docsUrl":null,"demoUrl":null,"supportUrl":null,"pricingUrl":null,"statusUrl":null},"customPage":null},"relatedAgents":{"evidence":{"source":"protocol-neighbors","verified":false,"confidence":"medium","updatedAt":"2026-10-10T09:03:49.019Z","emptyReason":null},"items":[{"id":"8ebccd8e-3863-4187-8355-c3f14e1f9edf","entityType":"agent","canonicalPath":"/agent/iofficeai-aionui","slug":"iofficeai-aionui","name":"AionUi","description":"Free, local, open-source 24/7 Cowork app and OpenClaw for Gemini CLI, Claude Code, Codex, OpenCode, Qwen Code, Goose CLI, Auggie, and more | 🌟 Star if you like it!","url":"https://github.com/iOfficeAI/AionUi","homepage":"https://www.aionui.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-10-09T19:11:12.944Z","createdAt":"2026-02-25T03:38:16.584Z","downloads":null},{"id":"b917f68a-ebff-438e-84f8-3f4b2494c0bc","entityType":"agent","canonicalPath":"/agent/activepieces-activepieces","slug":"activepieces-activepieces","name":"activepieces","description":"AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents","url":"https://github.com/activepieces/activepieces","homepage":"https://www.activepieces.com","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-15T02:22:12.426Z","createdAt":"2026-02-25T03:38:12.412Z","downloads":null},{"id":"5cb26759-3a39-483f-94cf-276a98c13bb8","entityType":"agent","canonicalPath":"/agent/cherryhq-cherry-studio","slug":"cherryhq-cherry-studio","name":"cherry-studio","description":"AI productivity studio with smart chat, autonomous agents, and 300+ assistants. Unified access to frontier LLMs","url":"https://github.com/CherryHQ/cherry-studio","homepage":"https://cherry-ai.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-11T14:38:40.986Z","createdAt":"2026-02-25T03:38:19.379Z","downloads":null},{"id":"6f6582d0-5d76-4f0f-b81d-86520247950b","entityType":"agent","canonicalPath":"/agent/copilotkit-copilotkit","slug":"copilotkit-copilotkit","name":"CopilotKit","description":"The Frontend for Agents & Generative UI. React + Angular","url":"https://github.com/CopilotKit/CopilotKit","homepage":"https://docs.copilotkit.ai","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-03-25T09:50:57.846Z","createdAt":"2026-02-25T03:39:14.617Z","downloads":null}],"links":{"hub":"/agent","source":"/agent/source/clawhub","protocols":[{"label":"OpenClaw","href":"/agent/protocol/openclew"}]}}}