Skip to main content
A coding agent takes a repository and a plain-language task, edits the code on a branch, and leaves the diff for you to review. This page is an example of how those docs can read. The steps are illustrative.

What you do

1

Open a repository

Point the agent at the repo you want changed. It reads the project before it edits anything.
2

Describe the task

Say what should be true when the work is done. Name the file or the behavior when you know it.
3

Let it work on a branch

The agent edits code, runs the check it can run, and keeps the result on a branch. It does not merge.
4

Review the diff

Read the diff. Merge it when it matches the task. Discard it when it does not.

Example prompt

What you should see

This is a sample of the docs shape, not the finished product reference.

Coding agents

Point an existing harness at the inference API.
Last modified on November 22, 2026