Docs · How to

How to: configure security & scopes

A full walkthrough of NIBWP security — scopes, force-draft, rate limits and the audit log — and a safe default configuration.

NIBWP is built so that capable automation stays controllable — but the controls only protect you if you use them. This guide walks the Security and Settings panel: how scopes work, how force-draft keeps AI writes reviewable, how rate limits stop runaway jobs, and how to think about least privilege. Spend ten minutes here and you can hand agents real power without losing sleep.

TL;DR

Security in NIBWP rests on four pillars: scoped tokens (least privilege), force-draft (AI writes become drafts you approve), rate limits (throttle bulk jobs), and the audit log (a record of everything). Configure them once and every agent inherits the guardrails.

The four pillars

Everything in the Security panel is one of four ideas. Understand them and you understand NIBWP’s safety model entirely.

PillarWhat it doesWhere you set it
ScopesLimit what each token can doOn the token, in Connect
Force-draftMake AI writes land as draftsSettings / Security panel
Rate limitsThrottle calls per tokenSettings / Security panel
Audit logRecord every callAlways on; viewed in its panel

Pillar 1 — Least-privilege scopes

The single most important habit: give each agent the smallest scope that does the job. A research assistant gets read. A content drafter gets write, not publish. Reserve admin for almost nothing. Because each client has its own token, you can be generous with one agent and strict with another.

Never give a chat client an admin tokenIf an agent can change settings or users, a single bad prompt can do real damage. Scope down, and gate anything irreversible behind human review. Admin scope is for trusted server-side automations, not interactive chat.

Pillar 2 — Force-draft

Turn this on and every AI write becomes a draft. The agent proposes content and changes; a human approves before anything is published. This one setting converts “an AI that can change my site” into “an AI that drafts changes for me to approve” — a profound difference in risk.

Force-draft is your seatbeltLeave it on for any agent that writes. The tiny cost — you click publish — buys total protection against an agent putting something live you did not intend.

Pillar 3 — Rate limits

Cap how many operations a token can perform per minute. This protects your database from a runaway loop — if an agent gets stuck repeating a write, the rate limit stops it before it does harm. For big legitimate jobs, the agent simply works within the limit and reports progress.

wp-cli
$ wp nibwp limits\nwrite: 60/min per token - read: 300/min - destructive: 10/min

Pillar 4 — The audit log

Every call is recorded — tool, token, arguments, result, timing. The audit log is both your security review (what did agents do?) and your debugging tool (why did that fail?). It is always on. See the audit log guide for reading it.

A safe default configuration

If you do nothing else, set this and you are in good shape:

  • Force-draft on for every writing agent
  • One scoped token per client, named for its purpose
  • Destructive abilities off unless a task needs them
  • Tokens in environment variables, never committed
  • A recent backup before any large automated job

FAQ

What is the safest scope to start with?
Read. It can look but never change. Graduate to write with force-draft when you are ready.
Does force-draft slow me down?
Barely — you click publish on what you approve. The protection far outweighs the click.
Can I see what an agent did?
Yes — the audit log records every call with full detail. Nothing is invisible.

Lock it down right

Set scopes, force-draft and limits once — every agent inherits the guardrails.

Get NIBWP

Connect your first agent.

Follow the quickstart, or book a demo and we'll walk you through it.