Five Ways to Give Claude Code the Right Context (Not Just @)

TL;DR Most people know @filename. But there are five distinct ways to inject context into Claude Code — files, directories, images, piped data, and CLAUDE.md imports. The last one is the one almost nobody uses, and it changes how you architect your entire config.

The @ reference is one of those features that's so basic you stop thinking about it. You type @, pick a file, done. But precisely because it feels simple, most developers never explore the full syntax — and there's more here than pointing at a file.

Claude Code has five distinct input methods for providing context. Each one serves a different purpose. Knowing when to use which is the difference between Claude guessing and Claude understanding.

Result:

> @src/auth/session.ts explain the token refresh logic around line 47

Claude reads session.ts from disk, focuses on line 47,
and explains the exact expiry flow — no guessing, no drift.

The five methods

1. @ file reference

The basic move. Type @ followed by a path and Claude reads the file from disk before responding.

> @src/api/payments.ts what does this module expose?

Tab completion activates after @ — the index pre-warms on startup with session-based caching, so suggestions appear fast. You don't need to type full paths.

Two details people miss:

• Case-sensitive on Linux and macOS. @Auth.ts and @auth.ts are different files.
• Fuzzy matching: typing @Button may surface both Button.tsx and Button.test.tsx, letting you pick from a filtered list.

Combine it with natural language line references for surgical precision:

# Focus on a specific area
> @src/auth/session.ts explain the logic around line 47

# Compare across files
> @src/models/user.ts line 23 vs @src/models/admin.ts line 18 — what's different?

2. @ directory reference

Point to a directory instead of a file:

> @src/api/ what endpoints do we have?

Claude gets a directory listing — file names, not full contents. Useful for orientation before diving into specific files. Saves context compared to reading every file individually.

3. Image paste

Press Ctrl+V (or Cmd+V in iTerm2) with an image in your clipboard:

[Image #1] fix the spacing issue visible in this screenshot

Claude receives the image as visual input. The [Image #N] chip inserts at your cursor position, so you can reference it by number in your prompt. You can also drag and drop image files directly into the terminal window.

Works with screenshots of UI bugs, design mockups, error dialogs, or terminal output that's hard to copy as text.

4. Pipe data from stdin

cat error.log | claude "what's causing this?"
npm test 2>&1 | claude "which tests are failing and why?"
git diff HEAD~3 | claude "summarize what changed"

Piped data enters context the same way a file reference does. The difference: it works with any command output, not just files on disk. Useful for logs, build output, API responses, or diffs you don't want to save first.

5. CLAUDE.md @imports (the advanced pattern)

This is the one most people don't know about. Inside any CLAUDE.md file, @path/to/file imports that file's content into context at session start:

# CLAUDE.md
See @README.md for project overview and @package.json for scripts.

# Workflow
- Git conventions: @docs/git-instructions.md
- Personal overrides: @~/.claude/my-project-instructions.md

This turns @ from a prompt feature into a config architecture tool:

• Recursive imports: imported files can import other files, up to 5 levels deep
• Relative paths: resolve from the importing file, not the working directory
• Approval dialog: the first time Claude encounters external imports in a project, it asks for confirmation
• Safe in code blocks: @some-package inside a fenced code block or inline code span won't trigger an import — you can document package names without side effects

The practical pattern: keep your main CLAUDE.md under 200 lines (the official recommendation), and @import everything else. Project architecture in one file, testing conventions in another, API standards in a third. Each loads at startup, but they're modular enough to maintain separately.

Reference

Official docs: Best Practices — Provide rich content | CLAUDE.md imports