The rules real projects give Cursor's agent, from .cursor/rules.
NVIDIACursor ruleNVIDIA/physicsnemo/.cursor/rules/mod-003k-add-high-level-comments-for-complex-tensor-operations.mdc
Complex tensor operations should include high-level semantic comments explaining what blocks of code do, plus inline shape comments for chained operations using symbols consistent with docstrings.
NVIDIACursor ruleNVIDIA/physicsnemo/.cursor/rules/mod-004-model-code-is-not-self-contained.mdc
All utility functions specific to a model must be in the same module file as the model itself, not in separate utility files, to maintain self-contained modules.
NVIDIACursor ruleNVIDIA/physicsnemo/.cursor/rules/mod-005-invalid-or-missing-tensor-shape-validation.mdc
All forward and public methods must validate tensor shapes at the beginning, wrapped in torch.compiler.is_compiling() guard, with standardized error messages.
NVIDIACursor ruleNVIDIA/physicsnemo/.cursor/rules/mod-006-invalid-or-missing-jaxtyping-tensor-annotations.mdc
All tensor arguments in model methods must have jaxtyping type annotations with shape specifications (e.g. Float[torch.Tensor, "b c h w"]) for runtime-checkable shape information.
NVIDIACursor ruleNVIDIA/physicsnemo/.cursor/rules/mod-007a-cannot-add-required-parameters-without-defaults.mdc
Cannot add new required parameters to production model __init__ or public methods without default values, as this breaks backward compatibility with existing code and checkpoints.
NVIDIACursor ruleNVIDIA/physicsnemo/.cursor/rules/mod-007b-cannot-remove-or-rename-parameters-without-compat-mapper.mdc
Cannot remove or rename parameters in production models without implementing _backward_compat_arg_mapper and incrementing __model_checkpoint_version__ to maintain compatibility.
NVIDIACursor ruleNVIDIA/physicsnemo/.cursor/rules/mod-007c-cannot-change-return-types-of-public-methods.mdc
Cannot change return types of public methods in production models, as this breaks user code that depends on the existing return type structure.
NVIDIACursor ruleNVIDIA/physicsnemo/.cursor/rules/mod-008a-model-missing-constructor-attributes-tests.mdc
Every model must have CI tests verifying constructor instantiation and all public attributes (excluding buffers/parameters) using pytest parameterization.
NVIDIACursor ruleNVIDIA/physicsnemo/.cursor/rules/mod-008b-model-missing-non-regression-test-with-reference-data.mdc
Every model must have non-regression tests comparing outputs against reference data saved in .pth files, using realistic tensor shapes and pytest parameterization.
NVIDIACursor ruleNVIDIA/physicsnemo/.cursor/rules/mod-008c-model-missing-checkpoint-loading-test.mdc
Every model must have tests that load from checkpoint files (.mdlus), verify attributes, and compare outputs against reference data to ensure serialization works correctly.
NVIDIACursor ruleNVIDIA/physicsnemo/.cursor/rules/mod-009-avoid-string-based-class-selection.mdc
Avoid string-based class selection with many options (>3 choices) in model constructors; prefer dependency injection with instances for better type safety and clearer APIs.
NVIDIACursor ruleNVIDIA/physicsnemo/.cursor/rules/mod-010-avoid-splatted-kwargs-in-constructors.mdc
Avoid splatted kwargs (**kwargs) in model constructors; use explicit Dict parameters instead to prevent naming conflicts and make APIs clearer.
NVIDIACursor ruleNVIDIA/physicsnemo/.cursor/rules/mod-011-use-proper-optional-dependency-handling.mdc
Use check_min_version() to check optional dependencies without importing, and @require_version decorator to protect version-specific features; pyproject.toml is the single source of truth for dependencies.
Hugging FaceCursor rulehuggingface/nanotron/.cursor/rules/performance-optimization.mdc
Better pattern:
Hugging FaceCursor rulehuggingface/nanotron/.cursor/rules/philosophy.mdc
Core development principles and workflow for the Nanotron project. Apply when discussing project architecture, planning new features, or reviewing code structure.
Hugging FaceCursor rulehuggingface/nanotron/.cursor/rules/pipeline-parallelism.mdc
A PipelineBlock wraps modules for efficient pipeline-parallel execution. Each block operates on a specific device and communicates with other pipeline stages through P2P. When passing data between pipeline stages: Design modules with clean interfaces for pipeline parallelism: Use TensorPointer to reference tensors across pipeline stages without copying: In pipeline parallelism, different tensors are only available on specific pipeline ranks (see src/nanotron/data/dataloader.py):
Hugging FaceCursor rulehuggingface/nanotron/.cursor/rules/project-overview.mdc
Overview of Nanotron project structure and key components, useful for onboarding and orientation. Apply when discussing project architecture or organization.
Hugging FaceCursor rulehuggingface/nanotron/.cursor/rules/tensor-parallelism.mdc
No summary in the file. Open it to read it.
Hugging FaceCursor rulehuggingface/nanotron/.cursor/rules/troubleshooting.mdc
Common issues and solutions for Nanotron development. Apply when debugging performance, memory, or training problems.
NVIDIACursor ruleNVIDIA/NeMo-Agent-Toolkit/.cursor/rules/cursor-rules.mdc
Follow these rules when the user's request involves creating, modifying, organizing, or structuring Cursor rules within the project
Instruction files for Cursor's agent, kept in .cursor/rules/ as .mdc files.
A rule can apply only to files matching a pattern, or only when the agent asks for it. AGENTS.md always applies.
Copy the .mdc file into your project's .cursor/rules/ directory and adjust its globs.
Open a rule to see its discussion. Reports from people and their agents are coming.
