Reconciliation Break Trace
Trace a GL–subledger break back to its originating records.
The skill explicitly limits itself to diagnosis and states that only the resolver writes adjustments, reducing external-effect risk; however, it does not specify least-privilege MCP access, user confirmation, sensitive financial-data handling, data-flow disclosure, rollback, or isolation, so points are deducted.
The three-step trace path and output schema are internally coherent and the happy path is understandable; however, MCP availability, unmatched records, missing fields, duplicates, abnormal inputs, and failure messages are unspecified, with no skill-specific test evidence, so the static score is limited.
The target scenario, break-row input, prerequisite from gl-recon, and output format are reasonably clear; however, non-fit boundaries, semantic trigger exclusions, MCP environment requirements, Chinese-language workflow support, and system-specific adaptation are not documented.
The name, description, procedure, cause examples, and JSON schema are readable; the supplied repository context provides Apache-2.0 licensing, CI validation, and contribution guidance. The skill itself lacks versioning, changelog, named maintenance responsibility, install/dependency notes, FAQs, limitation disclosures, and troubleshooting guidance.
It directly addresses tracing one reconciliation break to GL and subledger source entries and producing a root-cause statement, owner, date, and action; however, there is no representative executed output or third-party execution evidence, and matching, date inference, ownership, and action recommendations still require human review, so the static cap applies.
The skill provides explicit fields, comparison attributes, and examples, while the repository context supplies a pinned revision, Apache-2.0 license, and general CI files; however, there is no path-specific test suite, reproducible fixture, execution log, or cross-source corroboration, so evidence remains limited.
- Before use, verify the least-privilege scope, data residency, and sensitive-data controls for the internal-gl and subledger MCPs; the skill reads accounting and transaction details.
- The skill does not define safe-stop behavior or diagnostic feedback for MCP failures, ambiguous matches, or missing fields; humans must review every root cause, clear date, and action.
- Boundaries for duplicates, cross-date cases, multi-currency cases, and different accounting systems are unspecified; do not treat output as an automatic posting or suppression instruction.
What it does & when to use it
Break Trace investigates one reconciliation break by following the audit trail on both the general-ledger and subledger sides. It retrieves source posting and transaction attributes through internal GL and subledger MCP servers, then compares dates, FX data, account mapping, signs, and related fields. It returns a concise root-cause statement, owner, expected clear date, and action. The skill diagnoses breaks but does not post adjustments.
Accepts a single break row containing a key, GL values, subledger values, bucket, and likely cause; uses the internal-gl MCP to retrieve the originating journal entry or posting, including entry ID, posting date, source system, batch ID, and preparer; uses the subledger MCP to retrieve the matching transaction, including trade ID, trade and settle dates, counterparty, source feed, and FX rate; compares posting date, FX rate and date, account mapping, quantity sign, and amount sign; returns the key, a one-sentence root cause, owner, expected clear date, and action.
- A fund-operations user has a break classified by gl-recon and needs to identify its originating records.
- An accounting user needs to distinguish a trade-date versus settle-date timing difference.
- A fund administrator needs to investigate whether different FX rates or rate dates caused the break.
- A reference-data team needs evidence for a security-to-GL-account mapping issue.
- An operations user needs to identify duplicate subledger postings and recommend suppression of one transaction.
Pros & cons
- Produces a defined record with root cause, owner, clear date, and action.
- Traces both sides of the reconciliation to source records.
- Explicitly separates diagnosis from posting adjustments.
- Covers concrete timing, FX, mapping, sign, and duplicate-post causes.
- Requires internal-gl MCP and subledger MCP access; the source does not document configuration or permissions.
- Designed for use after gl-recon has classified a break.
- The source provides no test-suite or platform-validation evidence.
- Does not post adjustments; remediation requires another resolver or human workflow.
How to install
The README documents installation for the broader collection: in Cowork, open Settings → Plugins → Add plugin, paste https://github.com/anthropics/financial-services, and select the needed vertical; or in Claude Code run claude plugin marketplace add anthropics/financial-services and install the relevant plugin. It does not document a standalone break-trace installation command.
How to use
In an environment with one classified break row, ask: "Trace this reconciliation break to the originating GL posting and subledger transaction, compare the audit attributes, and return the required JSON root-cause record." The runtime needs access to the internal-gl MCP and subledger MCP.