We should add more instructions that LLMs can use to review PRs. This will help show our intent better, and can even be helpful for human reviewers. One example of these instructions is libcudacxx-style/SKILL.MD.
We can try getting these instructions by asking an LLM to go through our code and extracting similar characteristics for other subprojects.
Another avenue worth exploring is asking an LLM to look through our past PRs and seeing what insights it can extract from the reviews left by the maintainers
We should add more instructions that LLMs can use to review PRs. This will help show our intent better, and can even be helpful for human reviewers. One example of these instructions is libcudacxx-style/SKILL.MD.
We can try getting these instructions by asking an LLM to go through our code and extracting similar characteristics for other subprojects.
Another avenue worth exploring is asking an LLM to look through our past PRs and seeing what insights it can extract from the reviews left by the maintainers