Input Directives

Directives are special syntax in user input that trigger preprocessing before the message is sent to the LLM.

@Image

Embed images directly into user messages for vision-capable models. Images are resized and JPEG-compressed before attachment. If the current model doesn’t support vision, a warning is shown (message still sent).

@Image[screenshot.png] What's wrong with this UI?

Configure in .aura/config/features/vision.yaml:

Setting Default Description
dimension 1024 Max pixel dimension (width or height)
quality 75 JPEG compression quality (1-100)

@Bash

Run a shell command and inject its output as context before the user message. Output (including errors) is placed in a preamble block.

@Bash[git log --oneline -5] Summarize recent commits

@File

Inject file contents (or a directory listing) as context before the user message. Files over 500 lines are truncated; directories produce a shallow listing.

@File[src/main.go] What does this file do?

@Path

Replace the directive with a bare path, expanding environment variables inline — no file reading.

@Path[$HOME/.config/aura] Check this path

Autocomplete

In the TUI, typing @ triggers directive completion:

  • @Image[ shows a file picker filtered to image files (PNG, JPG, GIF, WebP, BMP)
  • Press Tab or Right Arrow to accept the completion

Input Size Guard

Messages using @Bash or @File directives that would push context usage above user_input_max_percentage (default: 80%) are rejected before entering the conversation. The error is shown in the TUI and the message is not sent.

Configure the threshold in .aura/config/features/tools.yaml:

tools:
  user_input_max_percentage: 85

Argument Splitting

Slash command arguments use quote-aware splitting (via shlex). This means you can use quoted strings with spaces as single arguments:

/assert "context_above:70 and auto" "/compact"
/assert bash:"go build ./..." "Fix the build errors."

If quote parsing fails (e.g. unmatched quotes in natural text), arguments fall back to whitespace splitting.

Supported Formats

Images and PDFs: PNG, JPG, JPEG, GIF, WebP, BMP, PDF

The @Image directive handles PDFs by extracting one image per page. For single-image analysis and text extraction, the Vision tool is also available.


Back to top

Copyright © 2026 idelchi. Distributed under the MIT License.