Claude Code's official guidance recommends using @ file references to pull the exact content of relevant files directly into a prompt so Claude has concrete code to follow, rather than describing where code lives or what it does. This is explicitly called out in the 'Provide rich content' section of the Claude Code best practices: 'Reference files with @ instead of describing where code lives. Claude reads the file before responding.' Because the task is a one-off integration and the developer already knows exactly which three modules exemplify the required patterns, there is no need to add this information to CLAUDE.md — CLAUDE.md is intended for persistent, project-wide conventions that should apply in every session (per Anthropic's guidance, a CLAUDE.md edit is warranted for a recurring convention or repeated mistake, not a one-off correction). Asking Claude to explore the codebase (option C) would waste context and time rediscovering files the developer has already identified. Describing the patterns in natural language (option B) is less precise and reliable than supplying the actual code, since Claude performs better when given concrete examples rather than paraphrased descriptions of conventions. Directly including the three files via @ references gives Claude the exact transaction handling, error format, and audit logging conventions to mirror when generating the new payment processing module.
Community Discussion