Base setup
The starting point: an agent with no additional configuration, used as the reference for everything else.
Before adding instructions, Skills, or tools, it is worth seeing how an agent works out of the box: with your project, your request, and nothing else. That is the base setup.
Why start here
Each topic in the path adds a piece to the agent's environment. To know whether a piece helps, you need something to compare it with. The base setup is that reference: if you add anAGENTS.md and the result improves, you know it because you saw what it looked like without one.
What it includes
- The project, with its code and its README.
- A request written in plain language.
- The tools the agent comes with by default, such as reading, editing, or running commands.
And what it leaves out: no AGENTS.md, no Skills, no MCP servers, and no deliberately configured permissions.
How to use it for comparison
- Choose a specific task and project.
- Run the task with the base setup and save the result exactly as it comes out.
- Add a single piece, for example an
AGENTS.md, and repeat with the same request and the same model. - Compare the two results and note what changed.
Its limits
A comparison like this shows what happened in one specific case; it doesn't prove that one setup is better in general. And if you change the model, the task, or the project between two runs, you can no longer tell what caused the difference.
To continue the path, readTask framing for agents.