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
Resource
Before an AI agent reaches production, inventory its real authority.
A practical launch-readiness checklist for teams preparing AI agents with tools, memory, retrieval, credentials, approvals, logs, and production access.
What you receive
Concrete outputs for a specific decision.
- Tool and data inventory
- Memory and retrieval checklist
- Credential and approval-gate review prompts
- Logging and artifact checklist
- Launch blocker worksheet
Work product
Checklist preview
Use this as the starting checklist before granting production access or asking for a formal assessment.
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?
- What logs and artifacts prove what happened after a bad action?
Related paths
Next step