菜单
Use when the user opts into a local research-interest model, wants personalized paper/preprint/trial updates, saves or rejects research suggestions, asks for a proactive briefing, or wants workflow prediction across sessions.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
用于给一个 outcome 的证据确定性做 GRADE 评级、产出 Summary of Findings (SoF) 表。触发：证据确定性、certainty of evidence、GRADE、证据质量分级、这个结论有多可靠、SoF 表、summary of findings、要不要下推荐、指南证据分级、quality of evidence、能不能信这个结论、证据强度、evidence certainty、把证据评个级。也用于"这个疗法的证据到底有多强"、"帮我给这几个结局评 GRADE"、"这条推荐的证据基础可靠吗"。禁用于：单纯核对某个 PMID 真伪（那用 evidence-audit / evidence_verify）。
Use for cross-modal target discovery from an unmet clinical need, combining literature, genetics, druggability, clinical trials, single-cell, and spatial evidence in one auditable context.
Design executable biomedical experiments from literature gaps or hypotheses, including testability, power, protocol, reagents, cost, preregistration, and failure modes.
Persist and query biomedical causal triples from BioCSSwitch conversations, literature checks, and debate outputs.
用于任何**可能含患者数据 / 临床原始记录 / 医疗档案**的场景，强制先做 PHI 检测再决定去向。触发：患者、病历、病案、临床数据、EHR、EMR、住院记录、门诊记录、handover note、progress note、discharge summary、SOAP、真实世界数据、RWD、reidentification、identifiable data、真实病人、我这里有一段病历、这是一份体检报告、帮我看一下这份 CT 报告、我上传一份门诊记录、患者档案、患儿、受试者数据、subject-level data、临床原始表、GCP 数据、eCRF、我要处理 HIPAA、PHI、脱敏、去标识化、我们医院的数据。也在提到 sensitive_mode / 敏感模式 / 隐私模式 时触发。禁用于：公开文献 / 已发表数据 —— 那些不含 PHI，扫描是浪费时间。
用于单细胞 RNA-seq 数据的标准化预处理、QC、doublet 检测、batch 整合、基因 ID 转换、细胞类型注释与内容指纹，为下游聚类、scFM embedding、DEG、轨迹或细胞通信准备可复现输入。触发：单细胞预处理、scRNA-seq QC、scanpy 预处理、AnnData、h5ad、质控、doublet、Scrublet、scDblFinder、batch correction、Harmony、scVI、BBKNN、基因 ID 转换、CellTypist、SingleR、CITE-seq、multiome、空间转录组、Visium。禁用于：bulk RNA-seq（用 geo-triage / omics_deseq2），以及 embedding 之后的 DEG/trajectory/cell communication（交给 sc-downstream-analysis）。
| name | research-partner |
| description | Use when the user opts into a local research-interest model, wants personalized paper/preprint/trial updates, saves or rejects research suggestions, asks for a proactive briefing, or wants workflow prediction across sessions. |
This skill turns repeated research interactions into a local, inspectable profile. It is not permission to retain conversation text.
research_interest_observe call, obtain explicit opt-in for local learning. Pass consent=true only within that consent scope. A user asking a biomedical question is not consent by itself.research_interest_inspect and research_interest_delete.paper_saved with its public ID and canonical topics.entity_queried.suggestion_accepted / suggestion_rejected; rejection is weak negative feedback, not proof that a topic is irrelevant.workflow_observed with a short task slug.recommendation_shown so cooldown deduplication can work.research_session_brief only if a local profile exists. Build topic_catalog from concepts already present in the local project, saved-paper index, or local KG.allow_remote_queries=false until the user or an existing policy explicitly authorizes outbound public-database queries.bio-lit / bio-trials, normalize results into the candidate schema, and call research_updates_rank locally.recommendation_shown, then record later acceptance or rejection if the user provides that feedback.This version supports session-start proactive briefings and scheduler-ready refresh actions. It does not silently run a background daemon, contact external services, or issue OS notifications. Any host scheduler must preserve the same consent, sensitive-mode, endpoint, cooldown, and deletion guarantees.