> ## Documentation Index
> Fetch the complete documentation index at: https://docs.oasm.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Instruction

> Add standing custom instructions to the agent's system prompt in every one of your conversations

The **Instruction** setting holds your own standing instructions for the agent: how you want it to work, what to prefer and how to report. They are added to the agent's system prompt at the start of every turn, in Ask mode and Agent mode, so you don't have to repeat them in each chat.

Open it from **Settings → Instruction** in the AI Studio sidebar. The header reads "Custom instructions added to the agent system prompt".

![Agent settings Instruction section with a large text box showing example lines Prefer passive recon before active scans, Report findings in a table with severity and affected asset, Keep answers short, the counter 0 / 20,000 and a Save button in the header](https://mintcdn.com/oasmplatform/JcAER2Ib37AN8_8A/images/screens/ai-instruction.webp?fit=max\&auto=format\&n=JcAER2Ib37AN8_8A\&q=85\&s=73c84151cd3516285f8e4fb560704b13)

Settings → Instruction, empty, with the example placeholder

## Scope

The instruction is personal. It applies to the conversations you start in the current workspace, not to other members' chats and not to your chats in other workspaces. Each member writes their own.

It is sent to the main agent only. [Subagents](https://docs.oasm.io/ai-studio/settings/agents.md) run on their own system prompts and the brief the main agent writes for them, so state a rule a subagent must follow in your request, where the main agent can include it in the brief (its `must_do` and `must_not` lists).

## Write or change the instruction

**1. Open Instruction**

In AI Studio, open **Settings → Instruction**.

**2. Write your instructions**

Type in the text box. The counter below it shows how many characters you used, up to **20,000**.

**3. Save**

Click **Save** in the header. The console confirms with "Instruction saved". **Discard** restores the saved text.

The change applies from the next message you send, in new and existing conversations.

To remove the instruction, clear the text box and click **Save**. Leading and trailing whitespace is trimmed when saving, so whitespace alone counts as empty.

| Rule | Value |
| - | - |
| Maximum length | 20,000 characters |
| Saved as | The trimmed text; an empty box clears the instruction |
| Permission to change | `agent.write` ([Permissions](https://docs.oasm.io/permissions.md)) |

## Examples of good instructions

Good instructions are short, concrete and stable: things that are true for most of your work, written as rules.

```text theme={null}
- Prefer passive recon before active scans.
- Report findings in a table with severity, affected asset and evidence.
- Keep answers short; lead with the conclusion.
- Never run brute-force or denial-of-service checks.
- Treat anything under *.staging.example.com as out of scope.
- When you are unsure whether a finding is real, say so and list the check that would confirm it.
```

Avoid:

* **Secrets.** Never put API keys, passwords or tokens in the instruction.
* **One-off tasks.** Put those in the chat message instead.
* **Engagement-specific rules.** Scope and rules of engagement for one client belong in that project's instructions (see below).

## How it combines with project instructions

A [project chat](https://docs.oasm.io/ai-studio/project-chats.md) can also carry the project's own instructions, set in the project's **Settings** tab. In a project chat the agent receives both: your personal instruction first, then the project's instruction, which is marked as set by the project's members. Neither replaces the other, so keep them consistent:

* **Instruction** (this setting): how *you* like to work, across every chat in the workspace.
* **Project instructions**: what applies to *one project*, for every member who chats from it.

## Related

> [card] [Project chats](https://docs.oasm.io/ai-studio/project-chats.md)
>
> Project instructions and reference files

> [card] [Memory](https://docs.oasm.io/ai-studio/settings/memory.md)
>
> Let the agent remember facts across conversations

> [card] [Skills](https://docs.oasm.io/ai-studio/settings/skills.md)
>
> Longer procedures the agent loads only when needed


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.