# Per-skill test environments.
#
# `python tests/run_all.py --isolated` builds one throwaway uv environment per
# skill from this file, so a skill whose upstream pins conflict with another's
# never has to share an interpreter. Nothing here is installed into the project
# environment, and `uv sync` is unaffected.
#
# Each skill's `packages` list is the set the skill's SKILL.md, references, and
# scripts actually name -- transitive dependencies are left to the resolver.
# Skills whose bundled tooling is standard-library only have `packages = []`;
# they still get a clean environment with just pytest.
#
# `python` overrides the interpreter for that skill alone. This is the reason
# per-skill environments are worth the trouble: opentrons pins numpy<2 (no
# cp313 wheels), PyTDC needs 3.11, molfeat caps at 3.10. uv downloads the
# interpreter on demand.
#
# Packages listed under [unavailable] are documented by a skill but cannot be
# installed here at all; the reason is recorded so nobody re-derives it.

[defaults]
python = "3.13"

# --- output formats and writing -------------------------------------------

[skills.docx]
packages = ["defusedxml", "lxml"]

[skills.xlsx]
packages = ["openpyxl", "defusedxml", "lxml"]

[skills.pptx]
packages = ["pillow", "defusedxml", "lxml"]

[skills.pptx-posters]
packages = ["python-pptx", "pillow", "lxml"]

[skills.pdf]
packages = [
    "pypdf",
    "pdfplumber",
    "pypdfium2",
    "reportlab",
    "pdf2image",
    "pytesseract",
    "pillow",
    "numpy",
    "pandas",
]

[skills.liteparse]
packages = ["liteparse"]

[skills.markitdown]
packages = ["markitdown", "openai"]

[skills.latex-posters]
packages = ["requests", "python-dotenv"]

[skills.infographics]
packages = ["requests", "python-dotenv"]

[skills.generate-image]
# The bundled CLI is standard-library only, and a test enforces that.
packages = []

[skills.markdown-mermaid-writing]
packages = []

[skills.venue-templates]
packages = []

[skills.scientific-writing]
packages = []

[skills.scientific-slides]
packages = [
    "pymupdf",
    "python-pptx",
    "pillow",
    "matplotlib",
    "seaborn",
    "numpy",
    "pypdf",
    "requests",
    "python-dotenv",
]

[skills.scientific-schematics]
packages = ["requests", "python-dotenv"]

[skills.scientific-visualization]
packages = ["matplotlib", "seaborn", "plotly", "pypdf", "pillow", "numpy"]

# --- literature and research workflow -------------------------------------

[skills.citation-management]
# The BibTeX parsing, rendering, and validation layers are standard library
# (scripts/_common.py). `requests` covers Crossref, PubMed, OpenAlex, and the
# DataCite fallback; `scholarly` is needed only by search_google_scholar.py.
packages = [
    "requests",
    "scholarly",
]

[skills.pyzotero]
packages = ["pyzotero", "bibtexparser", "python-dotenv"]

[skills.literature-review]
packages = ["requests", "python-dotenv"]

[skills.research-lookup]
packages = ["requests"]

[skills.exa-search]
packages = ["exa-py", "python-dotenv"]

[skills.parallel-web]
packages = []

[skills.paper-lookup]
packages = []

[skills.paperzilla]
packages = []

[skills.bgpt-paper-search]
packages = []

[skills.ontology-term-resolution]
packages = []

[skills.ncats-arax]
packages = []

[skills.peer-review]
packages = []

[skills.scholar-evaluation]
packages = []

[skills.research-grants]
packages = []

[skills.open-notebook]
packages = ["requests"]

[skills.usfiscaldata]
packages = ["pandas", "requests"]

[skills.market-research-reports]
packages = []

# --- reasoning and planning skills (standard library only) ----------------

[skills.hypothesis-generation]
packages = []

[skills.scientific-brainstorming]
packages = []

[skills.scientific-critical-thinking]
packages = []

[skills.consciousness-council]
packages = []

[skills.what-if-oracle]
packages = []

[skills.experimental-design]
packages = ["numpy", "pandas", "pyDOE3"]

[skills.iso-standards-readiness]
packages = []

[skills.analytical-method-validation]
packages = []

[skills.clinical-decision-support]
packages = []

[skills.clinical-reports]
packages = []

[skills.treatment-plans]
packages = []

[skills.get-available-resources]
packages = ["psutil"]

[skills.autoskill]
packages = ["httpx", "pyyaml", "sentence-transformers"]

[skills.pi-agent]
packages = []

# --- core data, statistics, and plotting ----------------------------------

[skills.uncertainty-and-units]
packages = ["pint", "uncertainties", "numpy", "scipy"]

[skills.matplotlib]
packages = ["matplotlib", "ipympl", "numpy", "pandas", "scipy"]

[skills.seaborn]
packages = ["seaborn", "matplotlib", "numpy", "pandas", "scipy", "ipywidgets"]

[skills.polars]
packages = ["polars", "pyarrow", "pandas", "numpy", "sqlalchemy"]

[skills.dask]
packages = [
    "dask[distributed]",
    "dask-ml",
    "dask-jobqueue",
    "dask-kubernetes",
    "pandas",
    "pyarrow",
    "h5py",
    "xarray",
    "zarr",
    "s3fs",
    "gcsfs",
]

[skills.zarr-python]
packages = ["zarr", "fsspec", "s3fs", "gcsfs", "h5py", "numpy", "xarray", "dask"]

[skills.vaex]
python = "3.12"
packages = ["vaex"]

[skills.exploratory-data-analysis]
packages = ["biopython", "h5py", "numpy", "pandas", "pillow", "polars", "tifffile"]

[skills.networkx]
packages = [
    "networkx",
    "geopandas",
    "momepy",
    "matplotlib",
    "pandas",
    "plotly",
    "pyvis",
    "scipy",
]

[skills.scikit-learn]
packages = [
    "scikit-learn",
    "category-encoders",
    "imbalanced-learn",
    "umap-learn",
    "matplotlib",
    "seaborn",
    "numpy",
    "pandas",
]

[skills.statsmodels]
packages = ["statsmodels", "scikit-learn", "matplotlib", "numpy", "pandas", "scipy"]

[skills.statistical-analysis]
packages = [
    "statsmodels",
    "pingouin",
    "pymc",
    "arviz",
    "scipy",
    "seaborn",
    "matplotlib",
    "numpy",
    "pandas",
    "scikit-learn",
]

[skills.statistical-power]
packages = ["statsmodels", "lifelines", "pingouin", "matplotlib", "numpy", "pandas", "scipy"]

[skills.pkpd-modeling]
packages = ["numpy", "scipy"]

[skills.pymc]
packages = ["pymc", "arviz", "matplotlib", "numpy", "pandas", "scikit-learn"]

[skills.shap]
packages = ["shap", "matplotlib", "numpy", "pandas", "scikit-learn"]

[skills.umap-learn]
packages = ["umap-learn", "hdbscan", "matplotlib", "numba", "numpy", "scikit-learn"]

[skills.scikit-survival]
packages = [
    "scikit-survival",
    "scikit-learn",
    "ecos",
    "osqp",
    "joblib",
    "narwhals",
    "numexpr",
    "numpy",
    "pandas",
    "scipy",
]

[skills.sympy]
packages = ["sympy", "mpmath", "numpy", "scipy", "matplotlib", "ipywidgets", "ipython"]

[skills.pymoo]
packages = ["pymoo", "joblib", "optuna", "matplotlib", "numpy"]

[skills.simpy]
packages = ["simpy"]

[skills.aeon]
packages = ["aeon", "matplotlib", "numpy", "scikit-learn"]

[skills.relsa-severity-assessment]
packages = ["numpy", "pandas", "scipy", "statsmodels", "matplotlib"]

[skills.timesfm-forecasting]
packages = [
    "timesfm",
    "jax",
    "flax",
    "torch",
    "scikit-learn",
    "matplotlib",
    "numpy",
    "pandas",
    "statsmodels",
    "pillow",
]

# --- machine learning -----------------------------------------------------

[skills.transformers]
packages = [
    "transformers",
    "datasets",
    "evaluate",
    "accelerate",
    "huggingface-hub",
    "timm",
    "librosa",
    "soundfile",
    "peft",
    "pillow",
    "torch",
    "wandb",
]

[skills.hugging-science]
packages = [
    "transformers",
    "datasets",
    "huggingface-hub",
    "accelerate",
    "gradio-client",
    "torch",
    "python-dotenv",
]

[skills.pytorch-lightning]
packages = ["lightning", "torch", "torchvision", "tensorboard", "wandb", "mlflow", "matplotlib"]

[skills.torch-geometric]
packages = [
    "torch-geometric",
    "torch",
    "captum",
    "lightning",
    "networkx",
    "pandas",
    "sentence-transformers",
    "scikit-learn",
]

[skills.stable-baselines3]
packages = ["stable-baselines3", "sb3-contrib", "gymnasium", "numpy", "torch", "tensorboard"]

[skills.pufferlib]
packages = ["gymnasium", "numpy"]

[skills.modal]
packages = ["modal"]

[skills.optimize-for-gpu]
packages = []

[skills.hypogenic]
packages = ["pyyaml"]

[skills.pyhealth]
packages = ["pyhealth", "torch"]

# --- single cell and transcriptomics --------------------------------------

[skills.scanpy]
packages = [
    "scanpy",
    "python-igraph",
    "leidenalg",
    "dask",
    "matplotlib",
    "numpy",
    "pandas",
    "scipy",
]

[skills.anndata]
packages = ["anndata", "h5py", "muon", "fsspec", "numpy", "pandas", "scipy", "scikit-learn"]

[skills.scvi-tools]
packages = [
    "scvi-tools",
    "scanpy",
    "squidpy",
    "mudata",
    "shap",
    "statsmodels",
    "seaborn",
    "matplotlib",
    "numpy",
    "optuna",
    "psutil",
    "scikit-learn",
    "torch",
]

[skills.scvelo]
packages = ["scvelo", "cellrank", "scanpy", "matplotlib", "numpy"]

[skills.cellxgene-census]
packages = ["cellxgene-census", "tiledbsoma", "tiledbsoma-ml", "scanpy", "spatialdata"]

[skills.pydeseq2]
packages = ["pydeseq2", "anndata", "matplotlib", "numpy", "pandas"]

[skills.bulk-rnaseq]
packages = ["pydeseq2", "pytximport", "gseapy", "gprofiler-official", "pandas"]

[skills.pathway-enrichment]
packages = ["gseapy", "gprofiler-official", "mygene", "numpy", "pandas"]

[skills.arboreto]
packages = ["arboreto", "dask[distributed]", "numpy", "pandas", "scipy"]

[skills.depmap]
packages = ["numpy", "pandas", "requests", "scipy"]

[skills.primekg]
packages = ["pandas"]

# --- genomics and sequence analysis ---------------------------------------

[skills.genomic-coordinates]
packages = []

[skills.pathogen-variant-surveillance]
packages = []

[skills.biopython]
packages = ["biopython", "matplotlib", "numpy", "reportlab"]

[skills.pysam]
packages = ["pysam"]

[skills.bioservices]
packages = ["bioservices", "biopython", "networkx", "pandas"]

[skills.gget]
# scanpy is pinned forward deliberately. Unpinned, the resolver walks it back to
# 1.9.8, whose numba 0.53.1 needs llvmlite 0.36.0, which cannot build on 3.13 --
# so the whole environment fails before a single test runs. >=1.10 is also what
# gget's own cellxgene extra requires.
packages = ["gget", "matplotlib", "numpy", "pandas", "scanpy>=1.10"]

[skills.scikit-bio]
packages = ["scikit-bio", "biom-format", "tables", "matplotlib", "numpy", "pandas", "seaborn"]

[skills.polars-bio]
packages = ["polars-bio", "polars", "pandas", "bioframe"]

# waypoint-bio's own package pulls torch + transformers + datasets, which the
# bundled scripts deliberately do not require: the converter is pandas-only and
# vocab_coverage.py lazy-imports the tokenizer so --help works without it.
[skills.waypoint-bio]
packages = ["pandas", "pyarrow"]

[skills.gtars]
packages = ["gtars"]

[skills.geniml]
packages = ["geniml", "gtars", "pyarrow", "scanpy"]

[skills.deeptools]
packages = ["deeptools"]

[skills.flowio]
packages = ["flowio", "numpy", "pandas"]

[skills.etetoolkit]
packages = ["ete4"]

[skills.phylogenetics]
python = "3.12"  # ete3 3.1.3 imports cgi, removed in 3.13
packages = ["ete3", "matplotlib"]

[skills.onekgpd]
packages = ["dnaerys"]

[skills.genomic-intelligence]
packages = ["requests"]

[skills.alphagenome]
packages = ["alphagenome"]

[skills.dhdna-profiler]
packages = []

[skills.glycoengineering]
packages = ["glycoshield", "pandas", "requests"]

[skills.esm]
packages = ["esm", "torch", "matplotlib", "numpy", "scikit-learn", "requests", "tqdm"]

[skills.diffdock]
packages = ["torch", "rdkit", "openmm", "esm", "pandas"]

# --- chemistry and materials ----------------------------------------------

[skills.rdkit]
packages = ["rdkit"]

[skills.datamol]
packages = ["datamol", "rdkit", "s3fs", "gcsfs"]

[skills.medchem]
packages = ["medchem", "datamol", "rdkit", "pandas", "tqdm"]

[skills.molfeat]
python = "3.10"
packages = ["molfeat", "datamol", "torch"]

[skills.matchms]
packages = ["matchms", "numpy"]

[skills.pyopenms]
packages = ["pyopenms", "matplotlib", "numpy", "pandas"]

[skills.molecular-dynamics]
packages = ["mdanalysis", "openmm", "matplotlib", "numpy", "pandas", "scipy"]

[skills.pymatgen]
# Pinned to the skill's own "Verified snapshot (2026-07-23)", which its
# `_common.py` enforces before any network execution. Left unpinned, a new
# pymatgen-core release makes the suite fail on the snapshot check -- and
# loosening that check would throw away the guarantee it exists to make. When
# the skill re-verifies against newer releases, update both together.
packages = ["pymatgen==2026.5.4", "pymatgen-core==2026.7.16", "mp-api==0.46.4"]

[skills.rowan]
packages = ["rowan-python", "rdkit", "pandas"]

[skills.deepchem]
python = "3.11"
# torch is what the skill's own `compatibility` line and the `deepchem[torch]`
# install command require: without a backend, dc.models.GCNModel and friends do
# not exist, so no model path is reachable at all.
packages = ["deepchem", "torch", "numpy", "scikit-learn"]

[skills.pytdc]
python = "3.11"
packages = ["pytdc", "setuptools"]

[skills.cobrapy]
packages = ["cobra", "matplotlib", "pandas", "seaborn"]

# --- medical imaging and clinical data ------------------------------------

[skills.pydicom]
packages = [
    "pydicom",
    "numpy",
    "pillow",
    "pylibjpeg",
    "pylibjpeg-libjpeg",
    "pylibjpeg-openjpeg",
    "pylibjpeg-rle",
    "pyjpegls",
    "python-gdcm",
]

[skills.bids]
packages = ["pybids", "nibabel", "pydicom", "heudiconv", "dcm2bids"]

[skills.imaging-data-commons]
packages = [
    "idc-index",
    "duckdb",
    "google-cloud-bigquery",
    "pydicom",
    "simpleitk",
    "numpy",
    "pandas",
    "s3fs",
    "gcsfs",
]

[skills.pacsomatic]
packages = []

[skills.histolab]
python = "3.11"
packages = ["histolab", "pooch"]

[skills.omero-integration]
packages = ["numpy", "pillow"]

# --- neuroscience ---------------------------------------------------------

[skills.neurokit2]
packages = ["neurokit2", "numpy", "pandas"]

[skills.neuropixels-analysis]
packages = ["spikeinterface", "probeinterface", "neo", "matplotlib", "numpy", "pandas", "scipy"]

[skills.arbor]
packages = ["arbor"]

# --- quantum computing ----------------------------------------------------

[skills.qiskit]
packages = [
    "qiskit[visualization]",
    "qiskit-aer",
    "qiskit-ibm-runtime",
    "qiskit-algorithms",
    "qiskit-nature",
    "qiskit-machine-learning",
    "qiskit-optimization",
    "qiskit-addon-cutting",
    "qiskit-addon-obp",
    "qiskit-addon-sqd",
    "matplotlib",
    "numpy",
    "scipy",
]

[skills.cirq]
packages = [
    "cirq",
    "cirq-aqt",
    "cirq-google",
    "cirq-ionq",
    "cirq-pasqal",
    "azure-quantum",
    "matplotlib",
    "sympy",
]

[skills.pennylane]
packages = [
    "pennylane",
    "pennylane-lightning",
    "pennylane-cirq",
    "pennylane-ionq",
    "pennylane-qiskit",
    "jax",
    "matplotlib",
    "networkx",
    "torch",
]

[skills.qutip]
packages = ["qutip", "qutip-qip", "qutip-qtrl", "qutip-jax", "matplotlib", "numpy"]

# --- physics, astronomy, and simulation -----------------------------------

[skills.astropy]
packages = ["astropy", "matplotlib", "numpy", "pandas", "dask"]

[skills.fluidsim]
packages = ["fluidsim", "fluidfft", "pyfftw", "h5py", "numpy"]

[skills.openpiv]
packages = ["openpiv", "numpy", "scipy", "scikit-image", "matplotlib"]

[skills.lab-hardware-cad]
# build123d caps at <3.15 and pulls the OpenCascade kernel via cadquery-ocp-novtk,
# a large wheel. The bundled suite is standard-library only by design -- it covers
# the standards database and the fit arithmetic, which are the checks that gate a
# fabrication decision -- so the kernel is listed for parity with SKILL.md rather
# than because the tests need it.
python = "3.12"
packages = ["build123d", "matplotlib"]

# --- geospatial -----------------------------------------------------------

[skills.geopandas]
packages = ["geopandas", "pyogrio", "pyproj", "shapely", "pyarrow", "matplotlib", "sqlalchemy"]

[skills.geomaster]
packages = [
    "geopandas",
    "shapely",
    "pyproj",
    "rasterio",
    "rioxarray",
    "xarray",
    "folium",
    "osmnx",
    "contextily",
    "cartopy",
    "mapclassify",
    "dask-geopandas",
    "planetary-computer",
    "pystac-client",
    "laspy",
    "torchgeo",
    "xgboost",
    "scikit-learn",
    "scikit-image",
    "earthengine-api",
    "rtree",
    "fiona",
    "geopy",
    "rasterstats",
    "esda",
    "mercantile",
    "networkx",
]

# --- lab platforms and integrations ---------------------------------------

[skills.benchling-integration]
packages = ["benchling-sdk", "biopython", "boto3", "httpx"]

[skills.dnanexus-integration]
packages = ["dxpy>=0.400"]

[skills.latchbio-integration]
packages = ["latch"]

[skills.labarchive-integration]
packages = ["labapi"]

[skills.lamindb]
packages = ["lamindb", "bionty", "lamindb-wetlab"]

[skills.opentrons-integration]
python = "3.12"
packages = ["opentrons"]

[skills.pylabrobot]
packages = ["pylabrobot"]

[skills.tiledbvcf]
packages = ["tiledb", "tiledb-cloud"]

[skills.nextflow]
packages = ["nf-core"]

[skills.tamarind]
packages = ["tamarind", "requests"]

[skills.adaptyv]
packages = []

[skills.ginkgo-cloud-lab]
packages = []

[skills.protocolsio-integration]
packages = []

[skills.database-lookup]
packages = ["zeep"]

[skills.matlab]
packages = ["h5py", "scipy"]

[skills.pathml]
python = "3.12"
packages = []

[skills.torchdrug]
python = "3.10"
packages = ["torch"]

# --- packages a skill documents that cannot be installed here -------------
#
# The runner prints these when it sets up the skill, so the gap is visible in
# the test output rather than rediscovered each time.

[unavailable]
adaptyv-sdk = "adaptyv skill: distributed from GitHub only, not on PyPI"
tiledbvcf = "tiledbvcf skill: not published to PyPI; conda-forge or a source build only"
matlabengine = "matlab skill: requires a local MATLAB installation to build against"
pufferlib = "pufferlib skill: source-only native archive, no wheel builds on 3.13"
hypogenic = "hypogenic skill: pins torch<=2.4.1, which has no cp313 wheels"
torchdrug = "torchdrug skill: torch-scatter 2.1.2 sdist needs torch present at build time"
pathml = "pathml skill: python-javabridge needs a JDK, plus OpenSlide and Bio-Formats"
omero-py = "omero-integration skill: zeroc-ice 3.6.5 has no wheels and needs a C++ build"
bids-validator-deno = "bids skill: Deno tool, installed with npm/deno rather than pip"
rsgislib = "geomaster skill: conda-forge only"
pdal = "geomaster skill: conda-forge only, needs the PDAL C++ library"
open3d = "geomaster skill: no cp313 wheels"
mpi4py = "fluidsim skill: needs a site MPI implementation and compilers"
dgl = "molfeat skill: no cp313 wheels"
"flash-attn" = "esm skill: needs a CUDA toolchain to build"
deepspeed = "pytorch-lightning skill: needs a CUDA toolchain to build"
bitsandbytes = "transformers skill: CUDA-only, Linux wheels"
"cudf-cu12" = "optimize-for-gpu skill: RAPIDS wheels are Linux + CUDA only (whole group)"