# NoHalu Agent Access Checklist

Use this before connecting an AI agent to a new tool, MCP server, plugin, or account.

## Purpose

- What exact task should the agent complete?
- What result would make this connection unnecessary?

## Access

- What data can the connection read?
- Can it create, edit, delete, send, buy, or publish anything?
- Does it expose credentials, private repositories, customer data, or a wallet?
- Can the first test use read-only access, a test account, or non-production data?

## Source and behavior

- Where is the source code and documentation?
- Which version or commit are you installing?
- Which tools are actually exposed?
- What happens if a tool fails or returns unexpected data?

## Control

- Which actions require human approval?
- Where are logs or observable results?
- How do you revoke credentials and disable the connection?

## Decision

- [ ] The task is specific.
- [ ] The access is no broader than needed.
- [ ] The first test is reversible.
- [ ] The source and version are known.
- [ ] The rollback path is known.