The @ Reference System and Rich Context Injection

@ references, image paste, the ! bash prefix, and Ctrl+G editor mode are the four mechanisms for injecting rich context into Claude Code prompts. Used together, they eliminate the gap between what you know and what Claude knows.

Quick Reference

→@filename: Claude reads the file before responding — content is in context before reasoning starts
→@directory/: Claude reads the directory listing for navigation context
→Image paste: Cmd+V (Mac/Linux) or Alt+V (Windows) — screenshots, mockups, error messages
→! prefix: run shell commands inline — output injected as context (!git log -10)
→Ctrl+G: open in $EDITOR for multi-paragraph prompts
→Ctrl+S: stash the current message to finish later

@ File and Directory References

@ references are processed before the prompt is sent to the model. The file content is in Claude's context before it reasons about your request — not injected as part of a 'please read this' instruction Claude might deprioritize.

@ reference syntax

@ vs telling Claude to read

Writing '@src/auth.ts' injects the file content upfront. Writing 'read src/auth.ts' is a tool call instruction Claude processes as part of the response. @ references are more reliable for critical context you need Claude to reason from.

Image Paste

Claude Code accepts images pasted directly into the prompt. This is faster and more accurate than describing what you see in text — especially for UI bugs, error messages, and design mockups.

The ! Bash Prefix

The ! prefix runs a shell command inline and injects the output into the prompt as context. Use it to inject real runtime data Claude cannot see by reading files.

Sign in to read this article

This is a premium article. Sign in with your Google account to continue.