菜单
Decode BO suggestions back to nearest real reactions via DRFP similarity.
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
Orchestrate an end-to-end chemistry or materials optimization study from a plain-English research question to BO execution and a paper draft.
BO execution layer — initializes a run from a resolved experiment spec, records observations, and continues through suggest/observe/report.
Initialize a BO run from a dataset or explicit search-space JSON.
Generate a final BO report and summarize optimization status.
Run an external evaluator loop for a BO run using a pre-provisioned backend id.
Design and stabilize an expensive or fragile chemistry evaluator before BO setup.
| name | bo-decode-drfp |
| description | Decode BO suggestions back to nearest real reactions via DRFP similarity. |
Use this skill when the BO engine has suggested fingerprint vectors and the user wants to find the closest real reactions from the original catalog.
uv run python -m bo_workflow.converters.reaction_drfp decode \
--catalog <CATALOG_CSV> --query '<JSON>'
Optional flags: --k <N> (number of nearest neighbors, default 3).
rank, similarity, and all non-fingerprint columns from the catalog (including the original reaction SMILES).--query accepts a JSON dict of fingerprint bit values (inline or path to a .json file), matching the format produced by suggest.--catalog can be the catalog.csv from encode, or any external CSV with matching fp_* columns (e.g. a large database of commercially available reactions). Useful for human-in-the-loop workflows where BO suggests fingerprints outside the training set.