科研 Skill
合成路线审查
review-routes
Reviews existing synthesis routes deterministically: topology, reaction precedent, and evidence gaps. Use to compare routes from AiZynthFinder, PaRoutes, or multi-step plans.
SKILL.md
合成路线证据评审
能力
对已有合成路线执行确定性、非生成式证据评审:
- 读取 normalized route、AiZynthFinder JSON 或 PaRoutes v2 JSON;
- 校验单根、连通、无环的
mol → reaction → mol交替路线树; - 规范化 target、intermediate、terminal precursor 和每个单步反应;
- 用稳定 route/step hash 绑定
curate-reactions与search-reactionsartifact; - 传播 rejected、review、zero hit、timeout、source error 和 license 缺口;
- 检查库存快照、显式项目约束和重复路线;
- 分维度展示路线,不生成隐藏综合分数;
- 输出 weakest steps 和专家 review queue。
适用于:
- “检查这几条逆合成路线各有什么证据缺口”;
- “比较 AiZynthFinder 输出的几条路线”;
- “逐步检查路线中的反应先例”;
- “哪些路线步骤只有相似反应,没有精确先例”;
- “检查路线前体库存声明和来源许可”。
执行流程
- 确认输入是已有路线 JSON,不接收 pickle,不调用路线生成模型。
- 显式确认
input_profile、来源 SHA-256 和routes_fingerprint。 - 验证路线树并生成稳定
route_signature/step_reaction_hash。 - 如提供第五、第六 Skill artifact,按
route_id + step_id + step hash验证后绑定,禁止按数组位置猜测。 - 运行:
python scripts/review_routes.py \
--input route-review-request.json \
--output route-review.json
- 校验:
python scripts/validate_output.py route-review.json
- 报告路线 disposition、逐步 evidence level、库存/许可缺口、显式约束和 review queue。
支持的输入 Profile
normalized_route_v1
aizynthfinder_json
paroutes_v2_json
首版最多 20 条路线、每条最多 50 步、总节点最多 5000。
强制边界
- 固定
rdkit==2025.9.2; - 核心不依赖 AiZynthFinder、Syntheseus、OpenAI4S 或模型权重;
- 只读取 JSON,禁止 pickle 或可执行反序列化;
- 不生成、补全或自动修复路线;
- 不运行 forward/round-trip、可行性、条件、yield 或安全模型;
- backend rank/score 原样保留,不跨 backend 归一化;
- exact、transformation、similar、component、zero hit 和 provider failure 必须分开;
- 条件和产率仅为来源报告证据,不自动迁移到目标步骤;
- 库存必须关联快照;route export 的
in_stock只标记为来源报告; - 缺失许可保持
null,不得猜测; - 不删除重复路线,只分组;
- 默认只输出
dimensions_only,不计算综合总分; ready_for_expert_review仅表示证据包可交给专家,不表示路线可行、安全、 最优、可放大或可直接实验;- 不访问网络,不调用商业库存、路线生成器或远程反应数据库。
与其他 Skill 的关系
curate-reactions
单步反应结构、角色、产率和质量状态
↓
search-reactions
单步先例、相似度、条件、产率、来源和许可
↓
review-routes
多步拓扑、逐步证据覆盖、最弱步骤和专家复核队列
完整输入输出、状态、规则和科学边界见
references/输入输出与科学边界.md。
示例提示词
帮我评审AiZynthFinder生成的这条合成路线的反应先例和证据缺口。
所属插件
安装该插件会一并获得同领域的 32 个 Skill / MCP。
同插件的其他科研 Skill
- ASE原子模拟环境Builds and analyzes atomic structures with ASE: geometry optimization, NEB transition states, and MD trajectories. Use for setting up or analyzing atomistic simulations.
- 电池材料计算工作流Guides DFT/pymatgen workflows for battery materials: voltage profiles, phase stability, and ion migration barriers. Use for cathode, anode, or electrolyte computation.
- 催化剂设计指导Advises on catalyst composition, structure, and synthesis strategy using a traceable methodology base. Use when designing or optimizing a catalyst for a given reaction.
- 催化文献检索整理Searches arXiv, OpenAlex, and journal sites for catalysis literature and outputs a structured matrix. Use for catalysis literature reviews on a reaction or material system.
- ChEMBL生物活性数据查询Queries the ChEMBL API for bioactivity data: target lookup, IC50/Ki retrieval, and SAR tables. Use to build bioactivity datasets for a protein target.
- 量子化学计算工作流Guides quantum chemistry workflows with ORCA and xTB: geometry optimization, thermochemistry, and transition-state search. Use for first-principles energy calculations.
- 分子描述符与指纹计算Computes versioned 2D descriptors, Morgan/MACCS fingerprints, and dataset quality profiles with fixed RDKit. Use to generate reproducible features or audit data quality.
- 腐蚀电化学分析Guides corrosion analysis: electrode potentials, polarization curves, impedance, and Pourbaix diagrams. Use for electrochemical stability or corrosion-kinetics reasoning.
- 化学反应数据整理Curates structured single-step reactions against an ORD schema and RDKit, preserving originals. Use to clean reaction SMILES or tables and flag duplicates or mass-balance issues.
- RDKit简化分子处理Wraps RDKit with sensible defaults for SMILES parsing, standardization, descriptors, and conformers. Use for standard drug-discovery tasks; use rdkit for advanced control.
- 红外拉曼光谱分析Processes IR and Raman spectra with baseline correction, peak picking, and functional-group assignment. Use to interpret vibrational spectra against reference data.
- 质谱相似性比对Computes spectral similarity scores and identifies unknown compounds from mass-spectral libraries. Use for metabolite identification; use pyopenms for full proteomics pipelines.
- 分子动力学轨迹分析Analyzes MD trajectories from GROMACS, AMBER, LAMMPS, or NAMD with MDAnalysis: RMSD, RMSF, and contact maps. Use when analyzing existing trajectory files.
- 药物化学分子过滤Applies drug-likeness rules and structural alert catalogs (PAINS, NIBR) to compound libraries. Use for triaging compound libraries early in drug discovery.
- 分子数据集构建与清洗Guides molecular ML dataset building: source selection, deduplication, unit harmonization, and leakage-aware splits. Use before training molecular property models.
- 分子特征化方法Guides molecular featurization strategy for ML: parsing, fingerprints, and descriptor selection. Use when designing a featurization pipeline before model training.
- 分子机器学习特征化Converts molecules into ML-ready features using 100+ pretrained and hand-crafted featurizers. Use for QSAR modeling or virtual-screening feature generation at scale.
- 核磁共振谱图分析Processes raw NMR FID data through Fourier transform and phase correction, then picks peaks with nmrglue. Use for processing and interpreting 1D/2D NMR spectra.
- 化学文件格式转换Converts chemical file formats and generates 3D coordinates and conformers with OpenBabel and RDKit. Use when interconverting structure formats or generating 3D structures.
- 粉末XRD物相分析Simulates powder XRD patterns from candidate structures and matches them to an experimental pattern. Use for phase ID; use dedicated software for full Rietveld refinement.
- pymatgen晶体材料分析Manipulates crystal structures and queries the Materials Project API for band gaps and formation energies. Use for computational materials science on crystal structures.
- 蛋白质组质谱分析Processes proteomics and metabolomics MS data via OpenMS bindings: feature detection and peptide ID. Use for LC-MS/MS pipelines; use matchms for simpler spectral comparison.
- PySCF量子化学计算Runs ab initio and DFT calculations with PySCF: HF/DFT energies, optimization, and MP2/CCSD(T). Use for first-principles molecular energy and property calculations.
- RDKit化学信息学Provides fine-grained RDKit control: I/O, descriptors, fingerprints, substructure search, and reactions. Use for advanced workflows; use datamol for a simpler interface.
常见问题
合成路线审查能做什么?
Reviews existing synthesis routes deterministically: topology, reaction precedent, and evidence gaps. Use to compare routes from AiZynthFinder, PaRoutes, or multi-step plans。
如何安装合成路线审查?
在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「合成路线审查」并安装;在 Claude Code 或 Codex 中可以发送本页的安装提示词,或下载 Skill 包后解压到 ~/.claude/skills 或 ~/.codex/skills。
合成路线审查需要哪些权限?
该科研 Skill没有声明额外权限;执行具体操作时仍由 Agent 按自身规则请求确认。
合成路线审查经过官方审核吗?
是。SpeedSkills 收录的每个版本都会记录内容摘要与权限声明,当前版本已通过安全审核。