pyproject.toml

[build-system]
requires = ["setuptools"]
build-backend = "setuptools.build_meta"

[project]
name = "corral"
description = "Simulation agent benchmark framework."
version = "0.0.1"
readme = "README.md"
license = { text = "MIT" }
authors = [
  { name = "Martiño Ríos-García", email = "martino.rios.garcia@uni-jena.de" },
  { name = "Nawaf Alampara", email = "pvt.nawaf@gmail.com" },
  { name = "Kevin Maik Jablonka", email = "mail@kjablonka.com" },
  {name = "Chandan Gupta", email = "chandan18386@iiitd.ac.in" },
  {name = "Sajid Mannan", email = "cez218288@iitd.ac.in"}
]
keywords = ["agent", "LLM", "benchmark"]
classifiers = [
  "Development Status :: 3 - Alpha",
  "Intended Audience :: Science/Research",
  "Programming Language :: Python :: 3",
  "Programming Language :: Python :: 3.11",
  "Programming Language :: Python :: 3.12",
  "Topic :: Scientific/Engineering",
  "Operating System :: Microsoft :: Windows",
  "Operating System :: Unix",
  "Operating System :: MacOS",
]
requires-python = ">=3.10"
dependencies = [
  "numpy",
  # Ceiling raised from <1.82 to admit the OpenHands SDK's litellm>=1.84.1
  # requirement (corral[openhands]); bounded to the 1.84.x line so the bump
  # stays deliberate rather than tracking every new litellm release.
  "litellm>=1.56.4,<1.85",
  "chromadb",
  "tenacity",
  "fastapi",
  "pydantic",
  "scikit-learn",
  "requests>=2.32.3",
  "urllib3",
  "openai",
  "uvicorn>=0.34.0",
  "python-dotenv>=1.0.1",
  "anthropic>=0.42.0",
  "loguru>=0.7.3",
  "modal>=1.1.4",
  "rich>=13.9.4",
  "promptstore @ git+https://github.com/lamalab-org/promptstore.git@main",
  "wandb>=0.21.0",
  "more-itertools>=10.7.0",
  "mcp>=1.0.0",
  "fastmcp>=2.0.0",
  "pre-commit>=4.3.0",
  "pygments-styles>=0.3.0",
  "fire>=0.7.1",
  "datasets>=4.6.1",
  "snakemake[irt]>=7.32.4",
  "netcdf4>=1.7.3",
  "openai-agents==0.13.1",
  "jsonschema>=4.0.0",
]

[project.optional-dependencies]
dev = ["pytest>=7.4.0", "pytest-cov>=3.0.0", "ruff>=0.0.285", "commitizen", "pytest-mock>=3.14.0", "ipython>=8.18.1", "fire>=0.6.0", "pandas>=2.0.0", "datasets>=2.14.0", "pyarrow>=14.0.0", "huggingface-hub>=0.20.0"]
docs = [
    "mkdocs-material>=9.4.0",
    "mkdocstrings[python]>=0.22.0",
    "mkdocs-gen-files>=0.5.0",
    "mkdocs-literate-nav>=0.6.0",
    "mkdocs-terminal>=4.7.0",
    "pillow>=10.0.0",
    "cairosvg>=2.7.1"
]
db = ["rdkit", "torch", "transformers"]
# The Codex agent (corral.agents.CodexAgent) drives the OpenAI Codex harness via
# the official Python SDK, which bundles a pinned local Codex runtime. Pinned
# exactly because the SDK is beta and the bundled runtime is tied to the SDK
# release; upgrading can change harness behavior. Bump deliberately.
# 0.1.0b3 is required: it pulls openai-codex-cli-bin==0.137.0a4, the first
# runtime with glibc-Linux (manylinux) wheels. The prior 0.1.0b2 pinned
# cli-bin==0.132.0, which only shipped musllinux/macOS/Windows wheels, so
# `uv sync --all-extras` failed to resolve on glibc-Linux CI runners.
# openai-codex-cli-bin is repeated here only to name its pre-release version
# explicitly: it is a transitive dependency of openai-codex, but uv's default
# prerelease policy will not accept the 0.137.0a4 pre-release from a transitive
# `==` pin unless a direct dependency requests it explicitly.
codex = ["openai-codex==0.1.0b3", "openai-codex-cli-bin==0.137.0a4"]
# The Claude Code agent (corral.agents.ClaudeCodeAgent) drives the Claude Code
# harness via the Claude Agent SDK, which bundles a specific Claude Code CLI.
# Pinned exactly: upgrading the package can change harness behavior even with
# identical source, so bump deliberately. Requires features newer than 0.1.0:
# the claude_code system-prompt preset, the `tools` base tool set, and the
# `dontAsk` permission mode.
claude = ["claude-agent-sdk==0.2.115"]
# The OpenHands agent (corral.agents.OpenHandsAgent) drives the OpenHands
# Software Agent SDK as a black-box harness. Pinned exactly because the SDK is
# changing rapidly and its behavior is tied to the release; bump deliberately.
# 1.35.0 requires litellm>=1.84.1, which is why the base litellm ceiling above
# was raised to <1.85. It also requires Python>=3.12, so the dependency is
# gated behind a marker: on corral's lower supported Pythons (3.10/3.11) the
# extra is empty and `OpenHandsAgent` raises a clear install hint at import.
# `openhands-tools` is intentionally *not* required: the harness receives no
# terminal/file-editor/browser tools, only the corral MCP endpoint.
openhands = ["openhands-sdk==1.35.0; python_version >= '3.12'"]
irt = [
    "pymc>=5.0.0",
    "arviz>=0.18.0",
    "matplotlib>=3.7.0",
    "seaborn>=0.13.0",
    "lama-aesthetics @ git+https://github.com/lamalab-org/lama-aesthetics.git@main"
]

[project.urls]
repository = "https://github.com/lamalab-org/corral"
documentation = "https://lamalab-org.github.io/corral/"
changelog = "https://github.com/lamalab-org/corral/blob/main/CHANGELOG.md"

[tool.setuptools.package-data]
corral = ["py.typed"]

[tool.pyright]
include = ["corral"]
exclude = ["**/__pycache__"]

[tool.pytest.ini_options]
minversion = "6.0"
addopts = ["-p no:warnings", "--import-mode=importlib"]
xfail_strict = true
log_cli_level = "warn"
pythonpath = "src"
testpaths = ["tests"]

[tool.black]
exclude = '''
/(
    \.git
  | \.tox
)/
'''
skip-magic-trailing-comma = true

[tool.isort]
profile = 'black'
skip_gitignore = true

[tool.coverage.run]
source = ["src"]

[tool.coverage.report]
exclude_also = [
  "if TYPE_CHECKING:",
  "if __name__ == .__main__.:",
  "except ImportError",
]

[tool.ruff]
target-version = "py310"
lint.select = [
  "A",     # flake8-builtins
  "ARG",   # flake8-unused-arguments
  "ASYNC", # flake8-async
  "B",     # flake8-bugbear
  "C4",    # flake8-comprehensions
  "DTZ",   # flake8-datetimez
  "E",     # pycodestyle error
  "EXE",   # flake8-executable
  "F",     # pyflakes
  "FA",    # flake8-future-annotations
  "FLY",   # flynt
  "I",     # isort
  "ICN",   # flake8-import-conventions
  "INT",   # flake8-gettext
  "ISC",   # flake8-implicit-str-concat
  "LOG",   # flake8-logging
  "NPY",   # numpy-specific rules
  "PD",    # pandas-vet
  "PERF",  # perflint
  "PIE",   # flake8-pie
  "PL",    # pylint
  "PT",    # flake8-pytest-style
  "PTH",   # flake8-use-pathlib
  "PYI",   # flake8-pyi
  "Q",     # flake8-quotes
  "RET",   # flake8-return
  "RSE",   # flake8-raise
  "RUF",   # Ruff-specific rules
  "SIM",   # flake8-simplify
  "SLOT",  # flake8-slots
  "T20",   # flake8-print
  "TCH",   # flake8-type-checking
  "TID",   # flake8-tidy-imports
  "TRIO",  # flake8-trio
  "UP",    # pyupgrade
  "W",     # pycodestyle warning
  "YTT",   # flake8-2020
]
lint.ignore = [
  "E501",    # Line too long
  "ISC001",  # single-line-implicit-string-concatenation
  "PERF203", # try-except-in-loop
  "PLR",     # Design related pylint codes
  "PT004",   # Fixture does not return anything
  "PT011",   # pytest.raises
  "PT012",   # pytest.raises
  "RET505",  # Unnecessary `elif` after `return`
]
lint.unfixable = [
  "T20",  # Removes print statements
  "F841", # Removes unused variables
]
lint.pydocstyle.convention = "numpy"
lint.isort.known-first-party = ["corral"]
src = ["src"]
extend-include = ["*.ipynb"]

[tool.ruff.lint.per-file-ignores]
"__init__.py" = ["F401"]
"tests/**" = ["ANN", "ARG", "B008", "D", "E402", "PTH", "S101"]

[tool.docformatter]
pre-summary-newline = true
black = true

[tool.mypy]
ignore_missing_imports = true
namespace_packages = true
explicit_package_bases = true
no_implicit_optional = false
disable_error_code = "annotation-unchecked"


[dependency-groups]
dev = [
    "isort>=6.0.1",
    "pytest>=8.4.1",
    "pytest-mock>=3.14.0",
    "ruff>=0.12.4",
    "streamlit>=1.46.1",
]
docs = [
    "cairosvg>=2.7.1",
    "mkdocs-terminal>=4.7.0",
    "mkdocs-gen-files>=0.5.0",
    "mkdocs-literate-nav>=0.6.0",
    "mkdocs-material>=9.4.0",
    "mkdocstrings[python]>=0.29.1",
    "pillow>=10.0.0",
]