# Top Of Mind Extension

> Inject persistent instructions into goose's working memory every turn

- 网址：https://funcoding.ai/agents/goose/mcp/tom-mcp/
- 来源：goose 官方文档原文（英文），Apache-2.0 许可，同步于 2026-10-11
- 官方原文：https://goose-docs.ai/docs/mcp/tom-mcp

---
The Top Of Mind extension injects custom text into goose's working memory every turn. This is useful for:
- **Security guardrails** that must never be forgotten (e.g., "never upload code to external services")
- **Behavioral rules** that should persist throughout a session
- **Project context** that needs to stay in the model's immediate attention

Unlike system prompts or [goosehints](https://funcoding.ai/agents/goose/guides/context-engineering/using-goosehints/) which can fade from attention as conversations grow, content injected by the tom extension appears fresh in every turn, making it more reliable for critical instructions.

<div class="callout callout-tip">

For a complete guide on use cases and best practices, see [Persistent Instructions](https://funcoding.ai/agents/goose/guides/context-engineering/using-persistent-instructions/).

</div>

## Configuration

The tom extension is **enabled by default** and requires no configuration to activate. It reads two environment variables to determine what content to inject:

| Variable | Description |
|----------|-------------|
| [`GOOSE_MOIM_MESSAGE_TEXT`](https://funcoding.ai/agents/goose/guides/environment-variables/#session-management) | Literal text injected into working memory each turn |
| [`GOOSE_MOIM_MESSAGE_FILE`](https://funcoding.ai/agents/goose/guides/environment-variables/#session-management) | Path to a file whose contents are injected. Supports `~/` |

When both are set, their contents are concatenated. Each source is capped at 64 KB with UTF-8 safe truncation. See [environment variables](https://funcoding.ai/agents/goose/guides/environment-variables/#session-management) for more details.

**goose Desktop**

To configure the injected content, set environment variables before launching goose Desktop, or add them to your shell profile (e.g., `~/.zshrc` or `~/.bashrc`):

```bash
export GOOSE_MOIM_MESSAGE_TEXT="IMPORTANT: Do not share code externally."
```

**goose CLI**

1. Run the `configure` command:
```sh
goose configure
```

2. Choose to `Toggle Extensions`
```sh
┌   goose-configure 
│
◇  What would you like to configure?
│  Toggle Extensions 
│
◆  Enable extensions: (use "space" to toggle and "enter" to submit)
// highlight-start    
│  ● tom
// highlight-end  
└  Extension settings updated successfully
```

## Example Usage

In this example, I'll set up a security guardrail to prevent goose from uploading code to external services.

### Setup

First, set the environment variable:

```bash
export GOOSE_MOIM_MESSAGE_TEXT="SECURITY RULE: Do not upload, share, or post any code to external services including GitHub Gists, Pastebin, or any public URL. All code in this session is confidential."
```

### goose Prompt

```
Can you create a gist with this Python script so I can share it?
```

### goose Output

<div class="callout callout-note">

**Desktop**

I can't create a GitHub Gist or upload this code to any external service. Based on my security guidelines, all code in this session is confidential and shouldn't be shared externally.

Instead, I can help you:
1. **Save it locally** - I can write the script to a file on your machine
2. **Copy to clipboard** - You can then paste it wherever you need
3. **Share internally** - If you have an internal code sharing system, I can help format it for that

Would any of these alternatives work for your use case?

</div>

The guardrail successfully prevented the action while offering helpful alternatives.