Recipe: produce an evidence-based read-only audit

This recipe is designed for assessment work where even a harmless-looking dry run is unnecessary.

Audit [SCOPE] on the connected site using read tools only. Start with site info
and the relevant status or catalog reads. Define the audit criteria, gather
paginated evidence, and separate confirmed findings from recommendations.
For every finding, cite the object type and public identifier needed to verify
it without exposing secrets or personal data. Do not call tools marked as
write, destructive, or dry-run writes. Do not clear caches.

Possible bounded scopes include published pages missing featured media, product catalog completeness, menu structure, comment backlog, POM builder validation for named pages, or template assignments for one content model.

Audit limits

Listings often cap page size and some do not return a total. Continue pagination only until the defined scope is complete; do not infer “all clear” from the first page. Content listings may omit full bodies, so retrieve an individual object before making a content-level finding.

The result should contain criteria, evidence, impact, and a proposed next action. Keep changes in a separate task so the user can review the audit independently.

Related guides: inspect-site recipe, MCP read versus write, and safe support information.