READ — inject file contents into the conversation context
/read FILE [FILE…]
/read loads one or more workspace files and injects their contents into the conversation as a user-role message. The model sees the file contents in the very next turn and can answer questions about them, suggest edits, or use them as reference material.
Multiple files are concatenated with a blank line between each, all in a single injected message. Each file’s content is preceded by a header showing its workspace-relative path.
FILE is a path relative to the workspace root. Absolute paths outside the workspace are rejected. Symlinks are not followed. Sensitive files (e.g. .env, id_rsa, harvey.yaml) are blocked regardless of permissions.
The agents/ directory is off-limits to /read to prevent skills and configuration from being inadvertently exposed.
Context window impact: reading large files can quickly consume the model’s context window. Check /status after reading to see the token impact.
Read a single file:
harvey > /read src/main.goRead several files at once:
harvey > /read README.md docs/ARCHITECTURE.mdRead then ask a question:
harvey > /read harvey.go
harvey > What does the ragAugment function do?/read-dir [PATH] — read all files in a directory /read-pdf FILE [PAGES] — extract and inject text from a PDF file /attach FILE — attach an image, PDF, or text file (auto-detects format) /files [PATH] — list directory contents /status — check remaining context window space /help read-dir /help read-pdf /help attach