---
title: What do the agent behaviour sliders do?
question: agent behaviour sliders
area: settings
order: 3
surface: desktop
audience: beginner
keywords: [agent behaviour, agent behavior, testing rigor, security rigor, response style, brain chat, sliders, how thorough, testing, security, verbosity, terse, exhaustive, extra instructions, defaults]
related: [agent-persona-and-ui-style, per-project-vs-global-settings, where-are-the-settings]
updated: 2026-07-21
---

# What do the agent behaviour sliders do?

Open the full settings page and click **Agent behaviour**. At the top are four sliders that set how your agents work: **Testing rigor**, **Security rigor**, **Response style**, and **Brain chat**. Each slider shows a plain-language word for where it sits (like "Standard"), not only a number. These are the defaults for every agent codus runs, and changes apply to the next thing an agent does, with no restart.

## Before you start

These four sliders live in the full settings page under **Agent behaviour**, not in the quick **⚙** dropdown. The lead line at the top of the section reads:

> Defaults for every agent codus runs. Each project can override these on its quadrant's ⚙ panel. Changes apply to the next thing the agent does, no restart needed.

## The four sliders

Each slider has a heading, a one-line description, and a readout showing the number and the word for the level you picked.

### Testing rigor

"How hard the agent tests the work it produces." The scale runs from 0 to 10, shown as **Off**, **Minimal**, **Light**, **Standard**, **Thorough**, then **Exhaustive** at the top. New installs sit at **Standard** (5). Higher means the agent writes and runs more tests before it calls the work done; lower means it does less checking and moves faster.

### Security rigor

"How carefully the agent reviews for security issues." The scale runs from 0 to 10, shown as **Off**, **Basic**, **Light**, **Standard**, **Thorough**, then **Adversarial** at the top. New installs sit at **Standard** (5). Higher means the agent looks harder for security problems in its own work.

### Response style

"How much the agent explains in its replies." The scale runs from 0 to 5: **Terse**, **Brief**, **Concise**, **Balanced**, **Detailed**, **Exhaustive**. New installs sit at **Balanced** (3). Lower gives you short answers with little explanation; higher gives you fuller write-ups. This changes how much the agent writes in chat, not how carefully it does the work.

### Brain chat

"How proactively the agent talks to the Codus Brain. 0 = only reports back on Brain-delegated work." The scale runs from 0 to 5: **Delegated only**, **Blockers**, **Cross-project**, **Coordinator**, **Collaborator**, **Continuous**. New installs sit at **Delegated only** (0), which means the agent stays quiet unless the Brain gave it a job, and only reports back on that. Higher makes the agent talk to the Brain more often on its own.

## Extra instructions

Under each slider is a small toggle (**▶ Extra instructions**) that expands a list where you can type your own rules for that area. Click **+ Add instruction**, type one rule per line, and remove one with the **✕** next to it. Each box shows an example of the kind of rule that fits, for instance "e.g. always add a regression test for every bug fix" under Testing rigor. These rules are added on top of wherever the slider sits.

## How do I know it worked?

There is no Save button. The moment you move a slider or add an instruction, it is stored. The new setting applies to the next thing an agent does, so a running agent picks it up on its next step without a restart.

## Common problems

- **My agent still ran lots of tests after I turned Testing rigor down.** A specific project can override the global default. Check that project's per-project setting. See [Do my settings apply to every project or only one?](per-project-vs-global-settings.md).
- **I want a different personality or a different look for what it builds.** Those are separate controls in the same section. See [How do I change my agent's tone and UI style?](agent-persona-and-ui-style.md).