[project]
name = "cuga"
version = "0.4.3"
description = "CUGA is an open-source generalist agent for the enterprise, supporting complex task execution on web and APIs, OpenAPI/MCP integrations, composable architecture, reasoning modes, and policy-aware features."
readme = "README.md"
license = "Apache-2.0"
license-files = ["LICENSE", "NOTICE"]
requires-python = ">=3.10, <3.15"
dependencies = [
    # First-party MCPAdapter replaces the archived langchain-mcp-adapters package.
    # The MCP namespace is beta: stay within the tested minor series.
    # Summary-model failures now retry then use our hard_truncation fallback,
    # instead of treating an "Error generating summary" placeholder as success.
    "langchain[mcp]>=1.4.6,<1.5",
    "langchain-ibm>=1.1.1",
    "ibm-watsonx-ai>=1.5.13; python_version>='3.14'",  # 1.5.10 StrEnum.__init__ crashes on 3.14
    "langchain-openai>=1.1.14",  # GHSA-r7w7-9xr2-qq2r: image-token-counting SSRF guard bypassed via DNS rebinding
    "langchain-core>=1.3.3",  # GHSA-pjwx-r37v-7724: unsafe deserialization via broad load() allowlists (<=1.3.2)
    # GHSA-cq5v-8q36-5273 (OOB heap read) and related aiohttp parser issues fixed in 3.14.3+.
    "aiohttp>=3.14.3",
    "httpx",
    "docker",
    "langgraph",
    "markdownify",
    "uvicorn",
    "fastapi[standard]",
    "langfuse",
    "dynaconf>=3.2.13",
    "playwright>=1.59.0",  # CVE-2025-9611 (MCP origin/DNS rebinding), CVE-2025-59288 (browser download SSL)
    "browsergym-core==0.13.0",
    "python-dotenv",
    "loguru",
    # 1.28.1+ fixes GHSA-hvrp-rf83-w775, GHSA-jpw9-pfvf-9f58, GHSA-vj7q-gjh5-988w (task/session/WebSocket auth).
    # FastMCP 4 and Evolve's CPEX integrations share the MCP 2 protocol models.
    "mcp[cli]>=2.0.0,<3",
    "psutil>=7.0.0",
    "typer>=0.15.3",
    "typer-slim>=0.15.3", # explicit: docling pulls typer meta-package which needs typer-slim
    # 4.1.2 is the stable floor shared with Evolve's server extra. FastMCP 3
    # requires MCP <2, which cannot coexist with CPEX >=0.1.4 / pii-filter >=0.3.9.
    # The fastmcp meta-package retains sole code ownership in fastmcp-slim.
    "fastmcp>=4.1.2,<5",
    # Authoritative harmony special-token vocabulary
    # (see cuga_graph.utils.harmony.harmony_special_tokens); avoids
    # hand-maintaining the list.
    "openai-harmony>=0.0.4",
    "pyyaml>=6.0.3",
    "langchain-groq>=0.3.8",
    # 0.6+/0.7.0 publish cryptography<49, which blocks GHSA-g6cj-pr64-35w5 (needs cryptography>=50).
    # Stay on 0.5.x until langchain-litellm lifts that cap; ChatLiteLLM API used here is unchanged.
    "langchain-litellm>=0.5.1,<0.6",
    # 1.89.7+ fixes GHSA-3cv6-jpf6-8222 (SSRF / provider-credential exfiltration via request-body
    # routing params, 1.89.0..1.89.6); 1.84.0+ fixed GHSA-4xpc-pv4p-pm3w (Host header auth bypass).
    "litellm>=1.89.7",
    "tavily-python",
    "fastembed>=0.4.0",
    "a2a-sdk[http-server]>=0.3.22",
    "sqlite-vec>=0.1.6", # GHSA-vrcx-gx3g-j3h8 affects <0.1.3 only; 0.1.6 is safe
    "pgvector>=0.4.2",
    "asyncpg>=0.30",
    # 2.15.0+ fixes GHSA-42vr-xj54-vc7v (RecursionError DoS in pre-verification parse, <=2.14.0) and
    # the 2.13.x batch: GHSA-ffc3-869f-jxw9, GHSA-p4g4-x82p-q773, GHSA-w2cx-738m-mc7w, GHSA-9j54-fg26-wv3r
    # (HMAC key-confusion / empty-key bypasses), GHSA-gvp8-978c-rx2q, GHSA-hxm8-2xgr-2p9m,
    # GHSA-r6x4-923q-g947, GHSA-9v7f-9g4p-ffgj, GHSA-2gx3-rcp4-g85q, GHSA-jwrc-g2q2-pq5p,
    # GHSA-w6j9-cwv2-h6wq, GHSA-8wjv-2p76-3863.
    "PyJWT[crypto]>=2.15.0",
    # 50.0.0+ fixes GHSA-g6cj-pr64-35w5 (PKCS#7 Bleichenbacher); 49+ also GHSA-jwv3-5hgf-82ww / GHSA-m2h6-j472-rp4c.
    "cryptography>=50.0.0",
    "hvac>=2.0.0",
    "boto3>=1.34.0",
    # Knowledge engine (replaces OpenRAG)
    "langchain-docling>=2.0,<3.0",
    # EasyOCR is the cross-platform safety net so OCR Just Works on a
    # fresh ``uv sync`` even when tesseract isn't on PATH. Tesseract
    # (system binary) is preferred when available — see
    # ``engine._select_ocr_engine_options``. Test
    # ``test_no_tesseract_falls_back_to_easyocr`` documents that this
    # fallback path must remain importable.
    "easyocr",
    # explicit: forces pytorch-cpu index via [tool.uv.sources], avoids 3.5GB nvidia-cuda wheels.
    # 2.13.0 fixes GHSA-rrmf-rvhw-rf47 (memory corruption in torch.jit.script, <=2.12.1).
    "torch>=2.13.0",
    "torchvision", # explicit: must match torch variant (cpu) to avoid operator registration errors
    "langchain-ollama>=0.2.0",
    "langchain-text-splitters>=1.0,<2.0",
    "beautifulsoup4>=4.12",
    "psycopg[binary]>=3.2",
    "toolguard>=0.2.17",
    "cuga-oak-health; python_version>='3.12'",
    "aiosmtpd",
]

[project.optional-dependencies]
# 0.12.1 provides stable protocol v1 plus the Agent/Client, stdio runtime, generated
# content/capability models, and permission interfaces used by the ACP adapters.
acp = ["agent-client-protocol>=0.12.1,<0.13"]
evaluation = ["rapidfuzz"]
e2b = ["e2b-code-interpreter>=2.4.1"]
knowledge = []  # knowledge is now built-in, no extra dependencies needed
health = ["cuga-oak-health; python_version>='3.12'"]  # kept for backward compat, now also in core deps
opensandbox = ["opensandbox", "opensandbox-code-interpreter"]
# Core keeps Phoenix optional; explicitly include the server, storage and PII integrations CUGA uses.
evolve = ["altk-evolve-core[server,cli,fastembed,pgvector,pii-regex]>=1.8.0,<2; python_version>='3.12'", "einops==0.8.1; python_version>='3.12'"]
# wxO SDK requires Python >=3.11; keep it an extra so 3.10 installs are unaffected.
wxo = ["ibm-watsonx-orchestrate-sdk>=2.15.0; python_version>='3.11'"]
# OpenLit (PyPI) pins openai<2; litellm 1.83.x needs openai 2.x — cannot both be core deps for pip/uvx.
# Note: When using observability extra, ensure python-dotenv>=1.1.0 is installed to avoid conflicts
observability = ["openlit>=1.40.3"]
# profiling-only — never a runtime dep. memray has no Windows or Python 3.14 wheels.
profiling = ["memray>=1.11; sys_platform != 'win32' and python_version < '3.14'"]

[project.urls]
Homepage = "https://cuga.dev"
Repository = "https://github.com/cuga-project/cuga-agent"

[tool.uv]
package = true
constraint-dependencies = [
    "pillow>=12.3.0",  # GHSA-62p4-gmf7-7g93 and related Pillow OOB/DoS CVEs in <=12.2.x
    # GHSA-8988-9cw3-xx77 (HTTPS proxy TLS config ignored), GHSA-vxq7-64xx-v4gw (unbounded chunk-size
    # line), GHSA-gh4c-6fx4-qh6g (chunked deflate infinite loop), all <2.8.0; earlier GHSA-qccp-gfcp-xxvc,
    # GHSA-mf9v-mfxr-j63j (<=2.6.x).
    "urllib3>=2.8.0",
    "h11>=0.16.0",     # CVE-2025-43859: HTTP request smuggling via chunked encoding
    "setuptools>=83.0.0",  # GHSA-h35f-9h28-mq5c (MANIFEST.in bypass via Unicode NFC/NFD collision, <83); earlier GHSA-cx63-2mw6-8hw5, GHSA-5rjg-fvgr-3xxf
    # Security updates for transitive dependencies
    "authlib>=1.7.1",      # GHSA-r95x-qfjj-fjj2, GHSA-w8p2-r796-3vmq (OAuth/OIDC open redirect in 1.7.0); earlier GHSA-wvwj-cvrp-7pv5, GHSA-m344-f55w-2m6j, GHSA-7wc2-qxgw-g8gg, GHSA-7432-952r-cw78
    "idna>=3.15",          # GHSA-65pc-fj4g-8rjx: crafted input to idna.encode() bypasses the CVE-2024-3651 fix (<3.15)
    "orjson>=3.11.6",      # GHSA-hx9q-6w63-j58v
    "pyasn1>=0.6.4",       # GHSA-m4p7-r5rc-7g4j, GHSA-8ppf-4f7h-5ppj, GHSA-hm4w-wwcw-mr6r (DoS in <=0.6.3)
    "protobuf>=5.29.6",    # GHSA-7gcm-g887-7qv7
    "starlette>=1.3.1",       # GHSA-82w8-qh3p-5jfq (form DoS), GHSA-wqp7-x3pw-xc5r (UNC SSRF on Windows)
    "python-multipart>=0.0.30",  # GHSA-5rvq-cxj2-64vf (querystring DoS); GHSA-pp6c-gr5w-3c5g fixed in 0.0.27+
    "langsmith>=0.8.18",  # GHSA-f4xh-w4cj-qxq8: TracingMiddleware arbitrary file read (<=0.8.17)
    "langgraph-checkpoint>=4.1.1",  # GHSA-fjqc-hq36-qh5p: unsafe JSON deserialization when loading checkpoints
    "langgraph-sdk>=0.4.4",  # GHSA-fvww-7h3r-vfhp (custom auth silently ignores actions= on resource decorators, <=0.4.3); earlier GHSA-w39p-vh2g-g8g5 (unsafe URL path construction)
    "pydantic-settings>=2.14.2",  # GHSA-4xgf-cpjx-pc3j: NestedSecretsSettingsSource follows symlinks out of secrets_dir
    "langchain-classic>=1.0.7",  # GHSA-3644-q5cj-c5c7: untrusted prompt manifest deserialization (<=1.0.6)
    # GHSA-x3q2-h9hx-4r4j (arbitrary file read/write + command exec), GHSA-pc36-qwjq-x68c (SSRF guard
    # bypass via DNS rebinding), GHSA-4xhp-xg4w-8ppm (local file read via draw:image xlink:href),
    # GHSA-q43m-vhcp-mhvm, GHSA-9jxx-vjrv-h2rq, GHSA-p944-4xh2-x776, GHSA-cgc7-9qp3-86m3,
    # GHSA-3cr3-8m4c-fpxw, GHSA-p3fw-7699-7926, all <2.132.0; earlier GHSA-q29v-xc37-wh5m
    # (unsafe URI/path handling in HTML backend, <=2.93.x). docling-slim ships the same code.
    "docling>=2.132.0",
    "docling-slim>=2.132.0",
    "mako>=1.4.2",            # GHSA-5639-2j2p-m4mx (drive-letter URI path traversal on Windows, <=1.4.1); earlier GHSA-v92g-xgxw-vvmm (// traversal, <=1.3.10), GHSA-2h4p-vjrc-8xpq (Windows \\ traversal, <=1.3.11)
    "soupsieve>=2.8.4",  # GHSA-2wc2-fm75-p42x, GHSA-836r-79rf-4m37 (memory / ReDoS in <=2.8.3)
    "joserfc>=1.6.8",  # GHSA-gg9x-qcx2-xmrh: empty/nil HMAC key accepted for HS* verify (<=1.6.7)
    "json-repair>=0.60.1",  # GHSA-xf7x-x43h-rpqh: circular JSON Schema $ref DoS (<=0.60.0)
    "datamodel-code-generator>=0.81.0",  # GHSA-77xj-x4rm-935c (Protobuf weak-import path traversal, <=0.80.0); earlier GHSA-5578-w22f-pfx9 and related code-injection / SSRF / path-read CVEs
    "transformers>=5.10.0",  # GHSA-xrqw-3rrv-vx5w: save_pretrained path traversal via chat template names (<5.10.0)
    # >=2.12.0 fixes GHSA-h4x7-gw46-3wm6, GHSA-pf96-p4fj-6566 (multipart header injection /
    # conflicting Content-Length+Transfer-Encoding, both <2.11.0) and GHSA-8xx6-hgc6-gc2m
    # (unbounded streaming-decompression memory amplification, <2.12.0).
    "httpx2>=2.12.0",
    "accelerate>=1.15.0",  # GHSA-4j2p-28q2-5m79: sharded-checkpoint weight_map path traversal / DoS (<=1.14.0)
    # GHSA-82r6-8w77-94w6 (TLSStream IDNA 2003 host encoding enables cert spoofing), GHSA-5p39-cfhj-2xmp
    # (process-pool workers block on undrained stderr), both <4.14.2.
    "anyio>=4.14.2",
    # GHSA-jhr6-gm9c-rqjv: local model loading ignores trust_remote_code=False and runs custom Python (<5.6.0).
    # Pulled in by altk-evolve (evolve extra).
    "sentence-transformers>=5.6.0",
    # GHSA-p58f-9548-mpm2 (activate scripts run commands embedded in paths, <=21.7.12), GHSA-x78j-v8h9-3j2q
    # (--prompt injection in activate.bat), GHSA-94p9-xgh2-xp45 (seed wheels not integrity-checked),
    # GHSA-9h9j-4vrj-gf7g (pyvenv.cfg line injection). Dev-only, via pre-commit.
    "virtualenv>=21.7.13",
    "fsspec>=2026.6.0",  # GHSA-27vj-qcqg-25rc: server-side template injection in ReferenceFileSystem (<2026.6.0)
    "multidict>=6.9.1",  # GHSA-54p9-h82j-f925: reference leak in CIMultiDict/MultiDict items-view union (6.7.0..6.9.0)
    # Not security: hold click where it was before this re-lock. Nothing requires >=8.2, but
    # re-resolving floated py>=3.12 up to click 8.5.0, which drops CliRunner(mix_stderr=) and breaks
    # the knowledge CLI tests. 8.2 also changed CLI behaviour we haven't vetted; keep the churn out.
    "click<8.2",
]
# litellm 1.83.4 pins python-dotenv==1.0.1; fastmcp needs >=1.1.0 — allow newer for sync/lock only.
# READI pins application-wide packages; preserve CUGA's patched versions and
# the pandas range supported by watsonx. Exercise READI in the offline image check.
override-dependencies = [
    "python-dotenv>=1.1.0",
    "nltk>=3.10.3",
    "pytz>=2025.1",
    "pandas>=2.3.3,<2.4",
    "safetensors>=0.8.0",
    "urllib3>=2.8.0",  # READI 0.1.7 pins urllib3==2.7.0; keep the 2.8.0 security floor
    # READI 0.1.7 pins datasets==5.0.0; override to keep the GHSA-379c-qx7v-6h59 floor (path
    # traversal in folder-based builders, <5.0.1). A constraint can't win against READI's exact pin.
    "datasets>=5.0.1",
]

[tool.uv.sources]
torch                 = { index = "pytorch-cpu" }
torchvision           = { index = "pytorch-cpu" }
torchaudio            = { index = "pytorch-cpu" }

[[tool.uv.index]]
name = "pytorch-cpu"
url = "https://download.pytorch.org/whl/cpu"
explicit = true

# PyPI openlit declares openai<2; litellm 1.83.x needs openai 2.x. Override for uv lock / Docker / dev only.
[[tool.uv.dependency-metadata]]
name = "openlit"
version = "1.40.3"
requires-dist = [
    "anthropic>=0.42.0,<1.0.0",
    "boto3>=1.34.0,<2.0.0",
    "botocore>=1.34.0,<2.0.0",
    "openai>=1.1.1,<3.0.0",
    "opentelemetry-api>=1.38.0,<2.0.0",
    "opentelemetry-exporter-otlp>=1.38.0,<2.0.0",
    "opentelemetry-instrumentation>=0.52b0,<1.0.0",
    "opentelemetry-instrumentation-aiohttp-client>=0.52b0,<1.0.0",
    "opentelemetry-instrumentation-asgi>=0.52b0,<1.0.0",
    "opentelemetry-instrumentation-django>=0.52b0,<1.0.0",
    "opentelemetry-instrumentation-falcon>=0.52b0,<1.0.0",
    "opentelemetry-instrumentation-fastapi>=0.52b0,<1.0.0",
    "opentelemetry-instrumentation-flask>=0.52b0,<1.0.0",
    "opentelemetry-instrumentation-httpx>=0.52b0,<1.0.0",
    "opentelemetry-instrumentation-pyramid>=0.52b0,<1.0.0",
    "opentelemetry-instrumentation-requests>=0.52b0,<1.0.0",
    "opentelemetry-instrumentation-starlette>=0.52b0,<1.0.0",
    "opentelemetry-instrumentation-tornado>=0.52b0,<1.0.0",
    "opentelemetry-instrumentation-urllib>=0.52b0,<1.0.0",
    "opentelemetry-instrumentation-urllib3>=0.52b0,<1.0.0",
    "opentelemetry-sdk>=1.38.0,<2.0.0",
    "opentelemetry-semantic-conventions>=0.59b0,<1.0.0",
    "pydantic>=2.0.0,<3.0.0",
    "requests>=2.26.0,<3.0.0",
    "schedule>=1.2.2,<2.0.0",
    "xmltodict>=0.13.0,<1.0.0",
]

[tool.setuptools]
package-dir = {"" = "src"}

[tool.setuptools.packages.find]
where = ["src"]
exclude = [
    "frontend_workspaces*"
]

[tool.setuptools.package-data]
cuga = [
    "configurations/**/*.toml",
    "configurations/**/*.md",
    "settings.toml",
    "backend/**/*.jinja2",
    "backend/cuga_graph/nodes/cuga_lite/prompts/*.json",
    "backend/**/*.js",
    "backend/server/demo_setup_utils/*.json",
    "backend/observability/assets/*.json",
    "frontend/dist/**",
    "demo_tools/**",
]
scripts = ["*.py"]
system_tests = ["**/*.py", "**/*.yaml", "**/*.sh"]



[project.scripts]
load_specs = "scripts.api_utils.load_openapi_specs:main"
demo = "scripts.commands:run_demo"
registry = "scripts.commands:run_api_registry"
registry_appworld = "scripts.commands:run_api_registry_appworld"

graph = "scripts.draw_graph:main"
eval_api = "scripts.commands:run_eval_api"
appworld_eval = "evaluation.appworld_eval:run_main"
appworld_setup = "scripts.commands:setup_appworld_environment"
appworld_setup_docker = "scripts.commands:setup_appworld_environment_docker"
cuga = "cuga.cli:app"
cuga-acp = "cuga.backend.server.acp.stdio:main"
code = "evaluation.code_generator:main"
exp = "scripts.eval_gui:main"
digital_sales_mcp = "scripts.commands:run_digital_sales_mcp"
digital_sales_openapi = "scripts.commands:run_digital_sales_openapi"
test_llm = "scripts.commands:test_llm"

[dependency-groups]
# Image-only dependencies: uv applies the READI compatibility overrides above.
# Keep these out of the published evolve extra, whose pip consumers do not use uv overrides.
evolve-image = [
    "altk-evolve-core[pii-semantic]>=1.8.0,<2; python_version>='3.12'",
    "readi-privacy==0.1.7; python_version>='3.12'",
    "en-core-web-trf @ https://github.com/explosion/spacy-models/releases/download/en_core_web_trf-3.8.0/en_core_web_trf-3.8.0-py3-none-any.whl ; python_version>='3.12'",
]
dev = [
    "pytz>=2025.1",
    "ruff>=0.9.4",
    "tabulate",
    "beartype",
    # >=3.10.3 fixes a batch of GHSAs disclosed 2026-09-02: GHSA-x99w-6fgc-pmfw (pickle allowlist
    # bypass RCE), GHSA-m4rf-3fr8-xwx3 (Stanford wrapper JVM arg injection, incomplete fix of
    # CVE-2026-12841), GHSA-3gq4-3j92-5w49 (corpus sandbox bypass), GHSA-p4rw-rvv2-7xwr (symlink
    # pathsec bypass), GHSA-6ww7-3frv-cqxh (proxy SSRF bypass), GHSA-cv22-g7mw-8v73 (Crubadan
    # path traversal), GHSA-5gh2-94qg-qppq (allowlist dotted-name RCE), GHSA-f794-5jv7-7672
    # (hardlink overwrite), GHSA-6hwm-xvph-95vm (Graphviz dot search-path hijack), plus DoS fixes
    # GHSA-97qj-x29f-37w7, GHSA-ww6m-cw3f-q94g, GHSA-vp2x-qp44-57v7, GHSA-ff5c-cp5c-9wjf,
    # GHSA-cw6x-m8jw-qmrh. See https://github.com/nltk/nltk/security/advisories for details.
    # NOTE: GHSA-8mgp-746c-j5xp (model-artifact APIs bypassing pathsec: TransitionParser,
    # AveragedPerceptron, PerceptronTagger, save_maxent_params) has no patched version as of
    # 3.10.3 upstream. CUGA does not call these model train/save/load APIs (grepped clean), so
    # it isn't reachable here, but the Dependabot alert (#619) can't be closed by a version bump.
    "nltk>=3.10.3",  # GHSA-p4gq-832x-fm9v and related path traversal / ReDoS / SSRF in <=3.9.x
    "termcolor",
    "rapidfuzz",
    "pytest>=8.3.5",
    "pytest-asyncio>=1.1.0",
    "pytest-xdist>=3.6.1",
    "cugaviz",
    "pre-commit>=4.3.0",
    "questionary>=2.1.1",
    "rich>=14.2.0",
    "pip-licenses>=5.5.1",
]
google-genai = [
    "langchain-google-genai>=2.0.11",
]
sandbox = [
    "llm-sandbox>=0.1.3",
]


[tool.pytest.ini_options]
testpaths = [
    "tests",
    "src/cuga/backend",
    "src/cuga/sdk_core",
    "src/system_tests",
]
python_files = ["test_*.py", "*_test.py", "*_test_*.py", "load_test*.py"]
python_classes = ["Test*"]
addopts = ["-ra", "--strict-markers", "-m", "not manual and not pgvector", "--import-mode=importlib"]
asyncio_mode = "auto"
norecursedirs = [".git", ".tox", "dist", "build", "docs", "frontend_workspaces"]
markers = [
    "e2e: live external services / full agent stack",
    "slow: long-running integration tests",
    "unit: fast isolated tests",
    "pgvector: requires CUGA_KNOWLEDGE_PGVECTOR_CONNECTION_STRING",
    "load: concurrent load tests (honors --load-test-users)",
    "stability: LLM-backed e2e stability suite (src/system_tests/e2e)",
    "windows_smoke: Windows CI smoke subset of stability tests",
    "manual: requires manually started services; excluded from default collection",
]

[tool.ruff]
# Exclude a variety of commonly ignored directories.
exclude = [
    ".bzr",
    ".direnv",
    ".eggs",
    ".git",
    ".git-rewrite",
    ".hg",
    ".ipynb_checkpoints",
    ".mypy_cache",
    ".nox",
    ".pants.d",
    ".pyenv",
    ".pytest_cache",
    ".pytype",
    ".ruff_cache",
    ".svn",
    ".tox",
    ".venv",
    ".vscode",
    "__pypackages__",
    "_build",
    "buck-out",
    "build",
    "dist",
    "node_modules",
    "site-packages",
    "venv",
]

# Same as Black.
line-length = 110
indent-width = 4

# Assume Python 3.9
target-version = "py312"

[tool.ruff.lint]
# Enable Pyflakes (`F`) and a subset of the pycodestyle (`E`)  codes by default.
select = ["E4", "E7", "E9", "F"]
ignore = []


# Allow fix for all enabled rules (when `--fix`) is provided.
fixable = ["ALL"]
unfixable = []

# Allow unused variables when underscore-prefixed.
dummy-variable-rgx = "^(_+|(_+[a-zA-Z0-9_]*[a-zA-Z0-9]+?))$"

[tool.ruff.lint.pyflakes]
allowed-unused-imports = ["webarena","evaluation.helper_functions"]

[tool.ruff.format]
# Like Black, use double quotes for strings.
quote-style = "preserve"

# Like Black, indent with spaces, rather than tabs.
indent-style = "space"

# Like Black, respect magic trailing commas.
skip-magic-trailing-comma = false

# Like Black, automatically detect the appropriate line ending.
line-ending = "auto"