Checklist Before Enabling an AI Assistant Inside a WordPress Plugin
Use this checklist before enabling an AI assistant inside a WordPress plugin. Activation is acceptable only when the product boundary, identity, credential lifecycle, allowed actions, forbidden actions, evidence owner and revocation path are all explicit.
Go or no-go criterion
Do not enable the assistant because the interface is present or because a connection test succeeds. The exit condition is a complete control record with no unknown owner, undocumented credential or untested refusal.
Before activation
- Identify the host plugin, admin screen, installed version and official documentation for that version.
- Separate the plugin’s assistant, any WAP Client integration, any MCP server and any remote backend.
- Document the data sent, endpoint, consent step, retention statement and available actions. Mark absent evidence as unverified.
- Assign a dedicated WordPress user; do not reuse a human administrator merely for convenience.
- Name every credential by integration and purpose, and record who creates, rotates and revokes it.
- Map each approved action to the required route, method, ability or tool and WordPress capability.
- Define a forbidden action that must remain refused.
- Assign the activation decision, evidence review and rollback to named owners.
Acceptance tests
- Confirm that the authenticated request resolves to the intended WordPress user.
- Execute one narrow approved read and retain a sanitized response.
- Attempt the predefined forbidden action and confirm that it is refused.
- Revoke the test credential and confirm that the same credential can no longer authenticate.
- Re-enable only with the final named credential and record its creation date and purpose.
Stop conditions
Stop if the integration cannot be tied to official versioned documentation, the credential owner is unknown, the required action needs broader access than approved, the refusal test succeeds unexpectedly, revocation recreates a credential without an understood trigger, or evidence would expose a secret or personal data.
Activation record
| Field | Required evidence |
|---|---|
| Product boundary | Plugin, WAP library, backend, connector and versions |
| Identity | Dedicated WordPress username, role and relevant capabilities |
| Actions | Approved and forbidden route, method, ability or tool |
| Credential | Label, owner, purpose, creation date and revocation trigger; never the secret |
| Decision | Approver, date, acceptance results and rollback owner |
Related guides
- How to Audit Every AI Credential on a WordPress Site
- How to Identify Which WordPress User an AI Assistant Uses
- What Is WAP AI Assistant in WordPress?
- Which WordPress Access Level Should You Give an AI?
- How to Revoke an AI Assistant’s Access to WordPress
- Plugin AI Assistant vs Claude Code, Codex and MCP for WordPress
Sources and verification
This page was checked against the following primary sources. Last source review: .
- WAP Client for WordPress Plugins · group.one / One.com
- Application Passwords · WordPress Developer Resources
- Roles and Capabilities · WordPress Developer Resources
- Model Context Protocol: Tools · Model Context Protocol