
ICD-10-CM Coding
SkillAnthropic 官方医疗技能:像专业病案编码员一样读门诊病历,判断哪些诊断该上报,按病历记载的精确程度给出可计费的 ICD-10-CM 编码。需接入 ICD-10 编码连接器(healthcare 插件自带),每个编码都经它核对。适用美国门诊场景,结果需持证编码员复核。
ICD-10-CM Coding Skill
Extract billable ICD-10-CM diagnosis codes from clinical notes the way a professional coder builds a claim.
Part of the Claude for Healthcare plugin, whose bundled ICD-10 Codes connector provides the skill's lookup tools out of the box. Further code-system skills (procedure coding, drug/supply coding) and a coding-audit workflow slot in alongside it as they're built.
What it does
Given a clinical note (visit note, encounter summary, discharge note), the skill produces the diagnosis codes a coder would submit on the claim for that encounter:
- Claim selection — decides which conditions belong on the claim (reason for visit, conditions managed this visit) and which stay off (history mentions, uncertain diagnoses, symptoms already explained by a coded diagnosis).
- Documented specificity — codes exactly what the clinician documented: unspecified when the note doesn't subtype, specific when it does, never inferring complications from labs or imaging alone.
- Verified lookup — every code is confirmed against a current ICD-10-CM source rather than recalled from memory, so post-2022 code-set changes (e.g. F32.A) are handled correctly.
Lookup
All code lookup goes through the bundled ICD-10 Codes connector (search_codes, lookup_code, validate_code) — no shell or network access is requested by the skill itself. If the connector is unavailable, the skill stops and asks for it rather than coding from memory: unverified recall is the main source of stale-code-set errors.
Scope
Written for and validated on outpatient encounter coding (primary care, ambulatory follow-ups, minor injury visits). Inpatient-specific conventions (POA indicators, DRG considerations) and specialty regimes with heavy 7th-character logic (poisonings/adverse effects, obstetrics) are not specifically covered.
This skill assists coding workflows; it does not replace certified coder review. Final claim responsibility remains with the billing provider.
相關資源

Prior Authorization Review
SkillAnthropic 官方医疗技能,用于美国医保预授权审核:读取临床资料,核验医生资质与诊断、手术编码,对照 Medicare 覆盖政策评估医疗必要性,给出批准、拒绝或待补材料的建议并起草通知函。需接入 healthcare 插件自带的 CMS、NPI、ICD-10 连接器,结论须经临床审核人员确认。
anthropics/healthcare452
Claude for Healthcare
PluginAnthropic 官方医疗插件:从病历提取结构化信息,给出 ICD-10 诊断与 CPT/HCPCS 操作编码,审核事前授权,筛查医保欺诈,起草临床试验方案,内置 CMS 医保覆盖、NPI、PubMed 等托管服务;连接医院 FHIR 系统需接口权限。面向美国医疗体系,结果需医务人员复核。
anthropics/healthcare452
OpenClaw Medical Skills
SkillFreedomIntelligence 团队从十余个开源技能库汇编的 869 个医学技能,含 SOAP 病历与出院小结、药物相互作用、临床指南与 PubMed/FDA 查询、基因组分析和医疗器械法规。主要为 OpenClaw 设计,也可用于其他 agent;部分技能需第三方 API Key,结果须由医生复核。
FreedomIntelligence/OpenClaw-Medical-Skills3.1k
Agent Skills for Healthcare & Life Sciences
SkillAWS 官方医疗与生命科学技能合集:42 个技能覆盖医学编码与术语映射、FHIR / HL7 病历解析、医保理赔与事前授权规则、医学影像、药物流行病学与基因组分析,以美国规范为准;分析类技能需本地装相应软件。仅供参考,不可用于临床决策,结果需专业人员复核。
awslabs/hcls-agent-skills76
AWS HealthLake MCP Server
MCPAWS 官方出品,让 agent 管理 AWS HealthLake 中的 FHIR 医疗数据:增删改查、高级检索、调取患者全部记录、导入导出,可开只读模式防误改。需要 AWS 账号、HealthLake 数据存储(付费)和本机 AWS 凭证;涉及临床的结论需专业人员复核。
awslabs/mcp9.9k
Mirobody
MCP自托管的健康数据引擎,把化验单、体检报告和可穿戴设备数据统一编码成标准指标,供 agent 查趋势、画图,每个数字注明来源文件。uvx mirobody mcp 可直接使用指标编码工具;查完整健康档案需用 Docker 自部署并配大模型 API Key。结果需医生复核。
thetahealth/mirobody1.5k