Skip to content
All plugins
reviewanomalyauditaccountantsprincipalsRuns on the XBert MCPClaude Code or Claude Desktop

XBert Anomaly Review

Every anomaly XBert that fired for the period — ranked, with the underlying transaction and a per-firing recommended action.

Updated 23 Jul 20261 skill1 commandby XBert Intelligence

Anomalies in the ledger are detected deterministically by Custom XBerts on the data XBert MCP exposes today — same vendor + amount + date duplicates on Bill or BankTransaction, period-over-period balance jumps, same vendor coded to a different account than its recurring pattern, round-number-recurring postings, volume spikes per vendor or account. The XBerts do the detection; this plugin does not.

Before reviewing, the plugin verifies the Anomaly Review agent is set up in your Connect portal with linked Custom XBerts — if it isn't, the review stops and walks you through the setup instead of presenting generic worklist data as an anomaly review.

XBert Anomaly Review surfaces what the linked anomaly XBerts caught for the period as a flat ranked list, verifying each firing against the underlying records before recommending an action (ABN/address comparison on duplicate-contact firings, current outstanding amount on overdue-invoice firings). Per firing: posting date, doc number, vendor, amount, account, the XBert that fired and its detection reason, the verification result, the count of prior firings on the same vendor or same account in the period (a deterministic fact, not a pattern label), and the recommended next action. XBerts whose firings contradict their intent are flagged with a suggested prompt refinement to apply in the XBert portal. Optional sub-grouping by deterministic facets (vendor, account, period bucket) keeps a long list scannable — never under invented pattern labels.

What the plugin explicitly does NOT do (no backing data in MCP today): identify the posting user, detect after-hours patterns or time-of-day attribution, infer intent (round-tripping), name pattern clusters across firings, track cluster-level resolution status, or claim Journal data. Duplicate-journals, reversal-pair and round-tripping detection all need Journal data which is not exposed today. If the user wants those, the answer is the same in every case: it needs a deterministic Custom XBert sitting on a data surface MCP exposes, and that surface doesn't exist yet.

Output lands as a working paper in your project folder — review it before anything goes to a client.

What you can do with it

4 use cases

Monthly anomaly review during close

Before signing off the month, walk every anomaly XBert that fired with the underlying transaction and the recommended action — per firing.

Pre-audit anomaly sweep

Ahead of an external audit, surface and resolve per-firing anomalies so the auditor sees a clean ledger with documented evidence.

New-client first-month sanity check

First month on a newly onboarded client surfaces the firings the linked anomaly XBerts caught on legacy data.

Recurring-issue investigation

When the same vendor or account keeps appearing in firings, the per-vendor / per-account recurrence count surfaces the pattern without inventing a label.

How it changes your work

What XBert users actually get back when they bring this into their practice.

XBerts detect, plugin surfaces

Detection is owned by the Custom XBerts deterministically; the plugin reasons only over what fired. No invented pattern labels, no user attribution, no Journal data claims.

Deterministic facets, not invented clusters

If sub-grouping helps, the label IS the data facet — '5 firings on Vendor X', '3 firings on Account Y', '8 firings in the last week of the period'. Never a name the XBerts cannot independently confirm.

Audit-ready evidence pack

Word working paper carrying every firing with its source XBert reference and the underlying Bill / BankTransaction / Invoice — filable straight into the client folder.

First install this plugin in Claude Code or Claude Desktop — see Install below. The slash command & its skills load automatically once installed. Then:

How it works

The flow from start to finish — under your control at every step.

  1. Type /anomaly-review in Claude Code or Claude Desktop

  2. Plugin checks the Anomaly Review agent is set up in your Connect portal with linked Custom XBerts — if not, it stops and walks you through the setup instead of reviewing generic worklist data

  3. Confirm the client and the review period

  4. Plugin pulls the period's firings from the linked anomaly XBerts and verifies each one against the underlying records (e.g. ABN/address on duplicate contacts, current outstanding amount on overdue invoices)

  5. Every firing is presented as a row in a ranked list with the underlying transaction, the XBert reason, the verification result, and a recommended next action

  6. XBerts whose firings contradict their intent are flagged with a suggested prompt refinement to apply in the XBert portal

  7. Optional sub-grouping by vendor / account / period bucket if the list is long (deterministic facets, never invented pattern labels)

  8. Build the working-paper payload (stamped with the plugin version), render via xbert-working-paper:render-docx, and report the saved path

Prerequisites

  • XBert account
  • Client connected to a ledger
  • Anomaly Review agent added via Automation → Agents → Browse Agents in the Connect portal
  • Anomaly XBerts linked to the Anomaly Review agent (Agent → Linked XBerts)
  • XBert Working Paper, the companion render plugin that writes the Word document — install it from this same catalogue (it needs Python 3.10+)

Install this plugin

Run in Claude Code or Claude Desktop
 1/plugin marketplace add ThirstyStudios/XBert-Plugins 2/plugin install xbert-anomaly-review@xbert

Add the marketplace once, then install any plugin from it.

Plugins add slash commands and skills on top of the XBert MCP — same data, same permissions, same audit trail as asking directly.

Other plugins