Skill directory

The application provides 23 bundled Skills. This directory groups them by the work they support. Entries describe the shipped method; they do not claim that every external model, dependency or service is installed.

For methods distributed through the app, use the marketplace installation guide. The directory below helps you choose a research method; it is not a live inventory of marketplace versions.

Check readiness before selecting a method

1. Open the Skill in Settings and read its complete requirements and third-party notices.
2. Compare the input type with your actual data. A bulk RNA-seq table is not a single-cell AnnData object; a molecular drawing is not a docking result.
3. Inspect the selected runtime and packages. For remote work, select a usable Compute Host and inspect its environment before submitting.
4. Request one bounded run, inspect the actual output, and retain the input/version references before scaling up.

Two additional manifest entries, self-awareness and skill-creator, are internal framework resources. They are not user-facing directory entries. Personal or imported Skills, including rnaseq-count-qc, are separate from the count of 23.

The three Environment application Skills and Customize stay enabled; see activation rules. Use the bundled manifest to identify shipped methods and Remote compute for host setup and result delivery.

Browse by research task

Protein structure

Skill	Input	Dependencies and execution	Output to inspect
AlphaFold2 · alphafold2	Protein FASTA; monomer or complex	ColabFold, model weights, GPU; optional public MSA service	Predicted structures and confidence scores
Boltz · boltz	Protein/DNA/RNA/ligand complex specification	Boltz package, weights, GPU; MSA access where requested	Complex structure and confidence; optional affinity output
Chai-1 · chai1	Multi-entity FASTA	chai-lab, weights, GPU	All-atom complex and confidence
ESMFold2 · esmfold2	Sequences or complex inputs	Biohub esm package, weights, CUDA; distinct from fair-esm	Structure predictions; ESMC representations where requested
OpenFold3 · openfold3	Protein/nucleic-acid/ligand specification	OpenFold3, weights/access, CUDA and configured kernels	Complex structures and scores

Protein design

Skill	Input	Dependencies and execution	Output to inspect
DiffDock · diffdock	Target PDB plus ligand SMILES/SDF	DiffDock repository, weights and GPU	Ranked ligand poses; pose confidence is not affinity
ProteinMPNN · proteinmpnn	Backbone PDB, designed/fixed chains and residues	Repository, checkpoints, torch/numpy; small jobs support CPU	Designed sequences and scores
LigandMPNN · ligandmpnn	Backbone plus ligand/metal/nucleic-acid context	Repository and Python dependencies; small jobs support CPU	Sequences and threaded structures
SolubleMPNN · solublempnn	Protein backbone	ProteinMPNN repository and soluble checkpoints; CPU possible	Sequences under the soluble-model prior

Sequence and cells

Skill	Input	Dependencies and execution	Output to inspect
ESM-2 · fair-esm2	Protein sequences	fair-esm and weights; bundled procedure uses GPU	Embeddings, logits or contact predictions
Borzoi · borzoi	DNA windows with stated genome/coordinates	borzoi-pytorch, weights and CUDA	Predicted genomic tracks or reference/alternate deltas
Evo 2 · evo2	DNA sequences or prefixes	Evo 2 weights, compatible CUDA and sufficient memory	Sequence likelihoods, embeddings or generated DNA
scGPT · scgpt	Single-cell AnnData with gene vocabulary mapping	scGPT package, checkpoint and GPU	Cell embeddings or annotation outputs
scvi-tools · scvi-tools	Single-cell counts and batch/label metadata	scvi-tools/scanpy/anndata; bundled training workflow expects GPU	Latent representation, label transfer or model-based comparisons

Evidence and writing

Skill	Input	Dependencies and execution	Output to inspect
Literature Review · literature-review	Research question, identifiers or papers	Source retrieval; OpenAlex key for OpenAlex operations	Verified evidence synthesis and citations
Indication Dossier · indication-dossier	An indication framed as a patient population	Research tools and source access	Resumable research waypoints and a dossier

Environment

Skill	Input	Dependencies and execution	Output to inspect
Environment & Packages · env-management	Missing package or version question	Selected Python/R runtime and allowed package source	Package inspection, managed installation and import check
Compute Environment Setup · compute-env-setup	Named environment on an SSH/Slurm host	Configured host and user/admin-managed activation	Setup instructions and validation record
Remote Compute (SSH) · remote-compute-ssh	A workload and eligible Compute Host	SSH credentials, host, scheduler when applicable	Submitted job, harvested results and published artifacts

Authoring

Skill	Input	Dependencies and execution	Output to inspect
Customize · customize	A requested Skill or Specialist change	Working agent; native customization operations	Saved package or role with read-back verification
Figure Style · figure-style	Actual data and one final figure	Notebook function and plotting dependencies	Inspected plot with legible labels and faithful data
Figure Composer · figure-composer	One claim and immutable data-version references	Main Agent, delegation, plotting and review	Multi-panel figure and review iterations
Paper Narrative · paper-narrative	Manuscript/abstract, captions and ordered figure deck	Grounded artifact versions and review tools	Paper brief and an ordered figure argument

Implementation reference: manifest.json.