Software engineering practices such as rate limiting or structured logging are widely reused across projects, yet their transfer remains unreliable when implementation context changes. Prior research and tooling have focused on detecting patterns or generating code, but these approaches typically ignore the intent of a practice and the semantic constraints imposed by frameworks, execution models, or libraries. As a result, copied or generated implementations may be incomplete, inconsistent, or fail at runtime—making it impossible to verify that a transferred practice preserves its purpose while adapting to environmental differences. This paper proposes Practice Specifications (PracSpec), a lightweight abstraction that represents a practice through its intent, explicit correctness criteria, and relevant context parameters. PracSpec separates what a practice means from how it is realized, enabling context-aware adaptation while preserving semantic intent. We demonstrate the feasibility of PracSpec through a case study on API rate limiting across two heterogeneous Python repositories: a synchronous Flask application and an asynchronous FastAPI service. The study exposes an important semantic constraint in FastAPI that is missed by Semgrep and handled inconsistently by ChatGPT. In contrast, PracSpec-guided adaptation produces context-correct changes and verifies behavior through runtime tests, improving conformance from 0%$\rightarrow$100% in both repositories. Our results demonstrate that formal practice specifications enable systematic transfer where informal approaches fail, enabling safer and more reliable practice transfer across diverse codebases.