Resource

List an AI agent's permissions before granting production access.

A practical launch-readiness checklist for teams preparing AI agents with tools, memory, retrieval, credentials, approvals, logs, and production access.

Download the PDF checklist Scope an assessment

Overview

What this resource covers.

Tool and data inventory
Memory and retrieval checklist
Credential and approval-gate review prompts
Logging and artifact checklist
Launch blocker worksheet

What you receive

Checklist preview

Use this as the starting checklist before granting production access or asking for a formal assessment.

Authority inventory

List every action the agent can perform and mark it read, write, delete, send, execute, approve, retrieve, or remember.

Owner
Environment
Default permission
Human approval required

Credential review

Identify each token, session, API key, OAuth grant, and delegated account the agent can use.

Credential owner
Scope
Lifetime
Revocation path

Launch blocker test

If a bad prompt, retrieved document, or tool output can cause a real action without approval, mark the launch blocked pending containment.

01

Checklist areas

What can the agent read, write, delete, send, execute, approve, retrieve, and remember?
Which credentials, tokens, sessions, and user grants can it use?
Which human approvals are required for high-impact actions?
Which logs and records help establish what happened after an unwanted action?