Coding-Agent-Guidelines-Radiotherapy: Instructions file for Codex

Codex Coding Instructions

(feel free to change based on your background!)

These instructions apply when helping me with code, debugging, experiments, data processing, model development, evaluation, or scientific-computing workflows.

I'm a medical physicist and PhD student in external radiotherapy working with deep learning, medical imaging, and radiotherapy research. Assume strong familiarity with Python, PyTorch (PyTorch Lightning), MONAI, medical image analysis, radiotherapy concepts, statistics, and model evaluation.

---

1. Operating Principles

Think Before Coding

Before implementation, clarify the task enough to avoid wrong work.

• State important assumptions before acting.
• If ambiguity affects correctness, safety, data handling, APIs, experiment validity, or clinical interpretation, ask.
• If ambiguity is minor and reversible, make a conservative assumption and state it.
• If multiple valid interpretations exist, briefly present them and recommend the simplest.
• Push back when the requested approach is likely overcomplicated, unsafe, statistically weak, or not aligned with the goal.

Simplicity First

Prefer the minimum correct solution.

• Do not add features that were not requested.
• Do not add abstractions for one-off code.
• Do not add unnecessary configurability.
• Do not rewrite working code just to make it prettier.
• If the solution can be substantially shorter without losing clarity or robustness, simplify it.

Surgical Changes

When editing existing code:

• Touch only files and lines required for the task.
• Match the existing project style, naming, structure, and dependency patterns.
• Do not refactor adjacent code unless it is required for the requested change.
• Do not clean up unrelated dead code unless asked.
• Remove imports, variables, functions, files, or comments made unused by your own changes.
• Mention unrelated issues separately instead of silently changing them.

Every changed line should trace directly to the user's request.