pyproject.toml

[build-system]
requires = ["setuptools>=69"]
build-backend = "setuptools.build_meta"

[project]
name = "code-task-forge"
version = "1.0.0"
description = "A SWE-bench-style harness for judging coding-agent patches: workspace isolation, patch policy, fail-to-pass / pass-to-pass with restored tests, held-out tests, and a failure taxonomy, measured on a pool of labelled candidate patches."
readme = "README.md"
requires-python = ">=3.10"
license = {text = "MIT"}
authors = [{name = "Yixuan Zhang"}]
# The harness runs candidate tests with `python -m pytest` in its own interpreter.
dependencies = ["pytest>=8"]

[project.optional-dependencies]
dev = ["ruff>=0.6"]
api = ["fastapi>=0.110", "uvicorn>=0.29", "httpx>=0.27"]

[project.scripts]
code-task-forge = "code_task_forge.cli:main"

[tool.setuptools.packages.find]
where = ["src"]

[tool.pytest.ini_options]
testpaths = ["tests"]

[tool.ruff]
line-length = 120
target-version = "py310"
extend-exclude = ["benchmark/repo", "benchmark/hidden"]

[tool.ruff.lint]
select = ["E", "F", "I", "B", "UP"]