Skip to content

AGENTS.md

Latest commit

 

History

History
72 lines (61 loc) · 3.45 KB

AGENTS.md

File metadata and controls

72 lines (61 loc) · 3.45 KB

DeepLidar Pretrain Project - Claude Instructions

Project Overview

You are working on a Python project for analyzing image annotations and bounding box data.

Critical Requirements

  • Short and concise code is preferred
  • The project uses "uv" for running scripts
  • NO EMOJIS - emojis are unprofessional and should be avoided completely
  • Provide a concise action plan before implementing any ideas - outline the approach first
  • Be direct, professional and if necessary, skeptical.
  • DO NOT INVENT HARDCODED INFORMATION - Never make up specific paths, filenames, directory structures, or other concrete details unless they are directly found in the codebase
  • Code comments should be brief and used sparingly

Environment

  • Environment variables are stored in .env (not accessible to you)
  • Environment is set in scripts using dotenv, but is not set in bash typically
  • Dataset is stored on an external volume with root directory specified by TIF_ROOT_DIR environment variable, assume this is fine.
  • pretraining_annotations.csv is a symlinked 2.3GB file with image paths and bounding boxes
  • SLURM scripts for GPU run on the hpg-turin partition with access to L4 GPUs. Normal CPU-only scripts do not need a partition specified.
  • Use test_minimal.csv and test_minimal_out for real data testing.

Code Standards

  • Follow PEP 8 Python style guidelines
  • When writing code, consider the runtime and cyclomatic complexity of the implmentation
  • Run pre-commit when you're finished.
  • DO NOT include complexity information in docstrings - keep docstrings focused on functionality

Testing

  • Use uv run pytest to run tests with appropriate flags.
  • Instructions in the tests folder.

Dependencies

  • ALWAYS use uv for ALL dependency management - never edit pyproject.toml directly
  • Use uv add <package> to add dependencies
  • Use uv run <command> to execute scripts
  • Use uv sync to install dependencies

Git Workflow

Follow Conventional Commits specification for all commit messages.

Types

  • feat: new feature
  • fix: bug fix
  • docs: documentation changes
  • style: formatting, missing semicolons, etc (no code change)
  • refactor: code change that neither fixes a bug nor adds a feature
  • test: adding missing tests or correcting existing tests
  • chore: changes to build process or auxiliary tools

Commit Rules

  • NEVER use git add . or git add -A or other commands that blindly include files.
  • Always add files individually using specific file paths: git add file1.py file2.py
  • Show the user what will be staged before adding files
  • Ask for confirmation on which files should be included in the commit
  • Use lowercase for type and description
  • Keep description under 50 characters
  • Use imperative mood (e.g., "add" not "added" or "adds")
  • No period at the end of the description
  • Stay objective and factual - avoid subjective words like "improve", "enhance", "better", "clean up", "simplify"
  • Be specific about what changed - describe the actual modification, not its perceived benefit
  • Use concrete technical terms rather than qualitative assessments
  • Avoid unnecessary justifications - state what changed, not why it's good

Examples

✅ Good: feat: add new validation for image paths ✅ Good: fix: handle null pointer in image validation ✅ Good: refactor: extract tile size calculation to separate function

❌ Bad: fix: improve error handling ❌ Bad: refactor: clean up tile algorithm ❌ Bad: feat: better image loading performance