← Về thư mục
📄 / / profiles / r-and-d / raw_evidence / OpenAI4S / skills / README.md

Bundled Skills

中文说明

This tree exposes 604 bundled Skills: 43 curated OpenAI4S recipes plus the 561 recipes in the pinned GPTomics/bioSkills collection. A Skill is a recipe—code plus the operational knowledge needed to run it—not a provider JSON Tool. Disclosure is progressive: curated Skills receive one summary line each, while the large third-party collection receives one line total and expands through search or exact loading. Only a selected Skill's SKILL.md and optional kernel.py sidecar are loaded.

Install

These recipes are not OpenAI4S-specific, and every Skill page below repeats this section with its own name filled in. With Node 18+ and git on PATH (npm resolves a github: spec through it), and nothing cloned:

npx github:PKU-YuanGroup/OpenAI4S install --all --target claude
npx github:PKU-YuanGroup/OpenAI4S install --collection bioskills --target claude
npx github:PKU-YuanGroup/OpenAI4S install alphafold2 boltz --dir ./my-skills

--target claude writes to ~/.claude/skills, claude-project to ./.claude/skills, openai4s to <data_dir>/user-skills, and --dir <path> to anywhere you name. The resolved absolute path is printed before anything is written there, and --dry-run stops at that plan. A reinstall refuses to overwrite a copy you have edited or one it did not install, and uninstall removes only the files it wrote. To install recipes included in the npm release, use npx @pku-yuangroup/openai4s-skills …. The npm catalog can differ from this repository; use @pku-yuangroup/openai4s-skills@0.2.0 to fix the release at 603 Skills (42 curated + 561 bioSkills).

Without Node, one directory comes out of the source tarball on its own, and python3 -m zipfile -c <name>.zip <name> turns it into something an upload field will take. The tarball is the whole repository (over 100 MB), and the pipe is a POSIX shell recipe (macOS, Linux, WSL):

curl -L https://codeload.github.com/PKU-YuanGroup/OpenAI4S/tar.gz/refs/heads/main \
  | tar -xz --strip-components=2 OpenAI4S-main/skills/alphafold2

The click-through form of that same download is the repository zip, which is also the route to take on Windows: extract it and copy skills/<name>/ out. If you already run OpenAI4S there is nothing to install — the wheel ships every Skill in this tree, and a bundled Skill takes precedence over a same-named copy in <data_dir>/user-skills. Targets, provenance, and what the installer refuses to do: ../tools/skills-installer/.

Subdirectories

Directory Responsibility
admet_genetic/ A genetic loop over seed SMILES, scored with RDKit descriptors, QED, SA-Score, and ADMET-AI. The sidecar ships no GA engine on purpose: mutation, crossover, filters, and scoring weights are yours to design for the objective at hand. Every logged candidate must carry the lineage that produced it.
alphafold2/ AF2 and AF2-Multimer through the ColabFold colabfold_batch runner, so a prediction is one FASTA and one command instead of a local database mount. The MSA comes from the public MMseqs2 server, which means the sequence is sent there. Proteins only. For ligands or nucleic acids, route to boltz, chai1, or openfold3.
audit-dataset/ The check to run before anything is trained or published: schema drift, missingness, duplicate rows and IDs, target imbalance, and entities shared across train, validation, and test. Pure stdlib. A clean structural audit still says nothing about whether the data is representative or the labels are right.
bioprobench/ Scoring a model's protocol reasoning against the BioProBench benchmark: question answering, step ordering, error correction, generation, and LLM-judged error reasoning. The contract is the trap — it scores one file that already merges ground truth into each model response, and a plain model-output file returns zeros under "status": "failed". Check Failed_Rate.
bioskills/ A pinned, read-only import of all 561 MIT-licensed GPTomics/bioSkills recipes across 63 bioinformatics categories. Its bilingual boundary docs, upstream license, and SHA-256 manifest preserve provenance; individual recipes remain searchable/loadable without placing 561 descriptions in every system prompt.
boltz/ Open-weights co-folding of protein, DNA, RNA, and ligand chains, with an optional small-molecule affinity head. Among the four co-folders it is the default for binder-validation campaigns: fully open MIT weights and the fastest sampler.
borzoi/ DNA in, predicted assay coverage out: RNA-seq, CAGE, DNase, and ChIP tracks over roughly 524 kb windows. Score a non-coding variant by running ref and alt windows and comparing the per-track delta. Reach for evo2 instead when you want sequence likelihood rather than assay tracks.
catalyst_sar_screening/ Single-atom-catalyst SAR on graphene M–N–C sites, hard-locked to FAIRChem UMA. Heuristics, lookup tables, and other MLIPs are forbidden, and so is handing the repo's committed demo outputs to a user as a result: every answer must come from a fresh pipeline run. If the weights hub is unreachable it stops and asks rather than substituting a method.
chai1/ The same co-folding surface as boltz from a different model, which is the point: run both and keep the designs that pass either. Its Python entry point makes it the easier of the two to drive from inside a loop, and Apache-2.0 explicitly permits commercial use.
diffdock/ Blind docking. No search box: the diffusion model places the ligand anywhere on the surface, and a confidence head ranks the samples. That confidence tracks pose correctness, not binding free energy, and the values are not comparable across complexes, so pair it with a scorer before triaging hits.
esmfold2/ The Biohub ESM release: all-atom co-folding that works from a single sequence, plus the ESMC language models for embeddings, mutation scoring, and contacts. Pick it over the other co-folders when you can live without an MSA.
evaluate-model/ Held-out scoring for binary classification and regression, with tie-aware ROC AUC and a deterministic bootstrap. Half of it is discipline rather than arithmetic: choose the metric before seeing the test set, compare against a baseline, check subgroups. Bootstrap intervals describe sampling variability and do not correct leakage or dataset shift.
evidence-walkthrough/ The reference end-to-end pass, and the one to run first: fixed query, local analysis, versioned artifacts that declare the inputs they were derived from, then a session package a recipient verifies with openai4s verify-package and no daemon. The accessions are fixed so two runs are comparable, which is also what makes it usable as a benchmark case. A pass means the package is intact, not authentic — the format carries no signature.
evo2/ A long-context DNA language model. Per-nucleotide likelihoods for variant-effect scoring, embeddings of genomic windows, and generation from a prefix. It gives sequence probability, where borzoi predicts the assay tracks.
example_stats/ The worked example of a user-authored Skill (origin: personal), and a usable one: mean, std, median, quantiles, z-scores, and Pearson correlation over plain Python lists, no NumPy or pandas. Read this one when you are about to write your own.
fair-esm2/ Meta's ESM-2 through the fair-esm package: per-residue and per-sequence embeddings, masked-LM mutation scoring, contact prediction. Mind the namespace collision. fair-esm and the Biohub fork behind esmfold2 both import as esm and are different libraries.
figure-composer/ The middle tier of the three figure Skills: make one multi-panel figure good. It turns a one-sentence claim into a panel outline on a 12-column grid, fans out a sub-agent per panel, tiles and letters the result, then runs an adversarial composite review for at most three rounds.
figure-style/ The inner tier: the rules for one plot. Deliberately a checklist and not a house look, covering data fidelity, label economy, chart choice by data shape, and a render-then-verify pass. The correctness sections bind everywhere; the aesthetic sections are defaults a deliberate alternative can override. Every panel sub-agent loads this.
indication-dossier/ Five resumable phases that build a dossier on one indication framed as a patient population rather than a disease: who they are, epidemiology, biology, standard of care, regulatory precedent, landmark trials. It expects the clinical-trials and pubmed MCP servers and falls back to web search against the public sources when they are not connected.
ligandmpnn/ Inverse folding when the design surface is not protein alone. Small molecules, nucleic acids, and metals are atoms the network can see, and proteinmpnn would simply ignore them. Its runner is also the one that threads designed sequences back onto the structure and writes PDBs.
literature-review/ From "what is the seminal paper for X" to a full multi-source review. The discipline is the content: retrieve before writing, resolve every DOI instead of recalling it, check CrossRef for retractions, and produce paragraphs that open on your synthetic claim rather than on an author's name.
mineral_spectra_analysis/ Unmixing the Raman spectrum of an unknown mineral mixture. Preprocess once, then loop: detect residual peaks, match the library, refit all components with NNLS, subtract. The blind loop must not read truth.json; ground-truth evaluation is a separate step that only runs once the answer is fixed.
openfold3/ An Apache-2.0 reproduction of AlphaFold3, so this is the co-folder to reach for when AF3-faithful behaviour is the point. Weights are gated on HuggingFace and need an accepted access form. The MSA server is on by default, so the sequence leaves the machine unless you opt out.
paper-narrative/ The outermost figure tier, and it starts a step earlier than expected: it reads the manuscript and the whole deck, then a handling-editor reviewer answers whether Figure 1 would get the paper sent out. Out comes the arc, panels sitting in the wrong figure, analyses that are missing, and what to kill. The brief it derives from your manuscript is model-generated, so review it before acting.
pdf-explore/ Parse a PDF once in the kernel and keep the page text, then work from an outline, relevance search, per-page extraction, and figure crops. Built for answers that draw on several places at once, or on every page. For a single lookup of one to four pages you will quote immediately, skip it and read the pages directly.
plan-ml-experiment/ What gets written down before training starts: hypothesis, baseline, metric, decision rule, and a split boundary that survives grouped or chronological structure. Reproducibility is mechanical here, resting on config fingerprints, dataset checksums, recorded seeds, and artifact manifests. Determinism does not prove validity, and repeating one biased split does not repair it.
protein-design-mcp/ Composing the bundled protein-design MCP tools rather than driving any one model: target-conditioned binder backbones, constrained sequence design, monomer and complex prediction, physical scoring and relaxation, sequence-naturalness scoring, minimization, and reproducible candidate comparison. It states its own boundary — the RFdiffusion tool behind it requires target hotspots and expresses no epitope-free, motif-scaffolding, unconditional or membrane-aware generation — and vendors no weights or GPU environment; the connector and its backends are configured separately.
protein-mutation-enhancement/ An orchestration layer, not a model. It builds deterministic mutant libraries with stable IDs such as A12V+G47D, merges sequence, structure, property, and assay scores into one ranking, and decides whether a gain-of-function round stops or expands. The heavy model calls are delegated to fair-esm2 and esmfold2.
proteinmpnn/ The default inverse-folding step when the design surface is protein only: backbone geometry in, sequence out, small enough to run on CPU for a handful of designs. It writes sequences and nothing else, so use the ligandmpnn runner if you need threaded PDBs, and switch skills as soon as a cofactor or soluble expression is in play.
reaction-atom-mapping/ RXNMapper atom correspondence and changed-bond extraction for a complete reaction. It requires both reaction sides and must not be used as a target-only retrosynthesis model or feasibility test.
reaction-condition-recommendation/ Parrot condition hypotheses for a fixed reaction, preserving checkpoint-specific label vocabularies and temperature support while separating model output from literature or ELN evidence.
reaction-forward-prediction/ ReactionT5v2 forward product prediction and round-trip recovery checks for proposed retrosynthetic steps. Product rank is model agreement, not experimental feasibility.
reaction-yield-estimation/ Bounded ReactionT5v2 yield screening for fully specified reactions, with an explicit domain gate and no route-wide success-probability claim.
rfdiffusion/ Backbone generation for de novo binders, hotspot conditioning, and motif scaffolding. The recipe fixes Hydra quoting and contig semantics, preserves .trb mappings and batch provenance, and requires downstream sequence design plus independent monomer/complex validation; generation alone is not binding evidence.
remote-compute-nvidia/ Dispatch to NVIDIA NIM in either of two forms that share one job contract. self_hosted runs an nvcr.io container on your own GPU; hosted needs no GPU but every job request leaves for NVIDIA's managed gateway. Only the declared key variables are forwarded to the confined helper, and they are scrubbed from the log tails that leave the sandbox.
remote-compute-ssh/ The orchestration half of running on the user's own SSH or SLURM host: partitions, env activation, job scripts, staging, harvest, recovery. Not the science. Each submit puts an approval modal in front of the user and spends their allocation, so the shape of a good run is to read what is already known about the host, ask once for what is not, land the first submit, and write down what you learned.
retrosynthesis_planning/ The multi-step scientific task: AiZynthFinder searches from a target to a declared stock, after which routes are normalized, de-duplicated, ranked, structurally audited, and rendered for chemist review. Single-step, forward, mapping, condition, and yield questions have their own Skills.
scgpt/ A transformer foundation model for single-cell data: cell embeddings for clustering, zero-shot or fine-tuned cell-type annotation, and gene-level representations for perturbation or GRN work. Checkpoints are raw directories rather than HuggingFace repos. The code is MIT and no source states a license for the weights.
single-cell-rna-analysis/ CPU-first Scanpy workflow for cell-called human/mouse 10x scRNA-seq and snRNA-seq counts: single-sample descriptive QC/clustering or donor-aware comparative inference, explicit-only Harmony, evidence-assisted annotation, checkpoints and a checksummed result bundle.
scvi-tools/ The probabilistic counterpart to scgpt: scVI for a batch-corrected latent space, scANVI for label transfer from a partly annotated reference, and Bayesian differential expression. It needs raw integer UMI counts. For spatial deconvolution or mapping, use cell2location, DestVI, or Tangram instead.
solublempnn/ ProteinMPNN's architecture retrained on a soluble-PDB subset, which shifts the output away from the surface hydrophobics the full-PDB model happily places. Reach for it when designs are aggregating or going to inclusion bodies. It trades a few points of native recovery for that bias, and a sequence-only prior is not a solubility measurement.
single-step-retrosynthesis/ RetroChimera one-step precursor proposals through the existing isolated, manifest-checked Syntheseus adapter. It deliberately stops before stock search or recursive route planning.
using-model-endpoint/ Documents a planned endpoint-scoped inference workflow: a Python kernel whose egress is scoped to one registered endpoint, with BASE_URL preloaded and no job lifecycle. The current Host implements endpoint registration and probes, but does not wire this provider into ComputeManager or create the scoped kernel yet.
volcengine-datapro/ A deliberately narrow MCP recipe for professional-dataset search: discover dataPro_search, make the real query call, and accept only an integer structured result code of zero as usable. Tool discovery alone is never an authentication verdict.

Where this fits