[Preview] New Feature: Automatic Shell Output Truncation

[SandyTao520]

New Feature: Automatic Shell Output Truncation

We've released a new experimental feature that automatically truncates large shell command outputs to prevent errors from exceeding the model's context window. This tool automatically truncates lengthy shell outputs to fit within the model's context window. Our truncation logic is designed to capture the most valuable information; it retains the first 20% and the last 80% of the lines specified in your settings. This emphasizes the final section of the output, where critical results and error messages are most often found.

This is especially useful if you run commands that often produce verbose output. If you've experienced 400 Bad Request errors in the past, we encourage you to turn this on and let us know how it improves your experience!

How to Enable:
This is currently only in preview, to install

npm i -g @google/gemini-cli@preview

Navigate to the /settings dialog.
Turn on "Enable Tool Output Truncation".
Optionally, you can configure the "Tool Output Truncation Threshold" (character limit to trigger truncation) and "Tool Output Truncation Lines" (number of lines to keep).

Please let us know what you think, thanks!