GL–Subledger Reconciliation
Reconcile general ledger and subledger data, expose breaks, and suggest likely causes for investigation.
The skill explicitly treats GL and subledger/custodian extracts as untrusted data and forbids treating their contents as instructions; it also does not claim to post to a ledger or perform external actions. However, it provides no access-control, user-confirmation, sensitive-financial-data, isolation, rollback, or retention guidance, so 7 points are deducted. The repository supplies an Apache-2.0 license and verified publisher provenance, which support traceability but do not warrant a security bonus.
The workflow, key definition, match buckets, default tolerances, and sorting rule are reasonably clear. It does not address empty files, duplicate keys, missing columns, conversion failures, classification precedence, abnormal inputs, or diagnostic failure messages, and there is no key-path test evidence for this skill; the score is therefore conservatively limited below the static ceiling.
The audience, inputs, daily/month-end scenarios, position or transaction granularity, and multi-asset use are reasonably clear. Non-fit cases, accepted file schemas, semantic trigger conditions, abnormal amount/quantity boundaries, and Chinese-language output expectations are unspecified, resulting in a 6-point deduction. The core function is local/file-based and does not entirely depend on overseas services, so no mainland-China reachability deduction is applied.
The document uses a clear stepwise structure and defines terms, matching conditions, cause categories, and outputs. The repository provides an Apache-2.0 license, a pinned revision, and plugin-validation CI. The skill itself lacks installation/dependency notes, examples, FAQs, known limitations, version history, changelog, and explicit maintenance ownership or update path, so 8 points are deducted.
The skill specifies a directly usable break report and summary, with handoffs to downstream tracing and sign-off workflows, so the core task is well framed. There are no representative outputs, acceptance criteria, or static evidence validating correctness, and edge cases may require substantial human cleanup; the score is capped at the static maximum of 7.
The source provides auditable rules, tolerances, classifications, and output requirements, while repository CI covers plugin validation and secret scanning. Those workflows do not test the reconciliation logic, and there are no committed fixtures, third-party execution results, or corroborating sources; 4 points are awarded conservatively.
- This is a static review only; validate the workflow with real fixtures covering duplicate keys, missing fields, empty inputs, sign flips, FX differences, and date conflicts before production use.
- Controls for access, masking, storage, and deletion of sensitive financial data are not specified.
- Likely-cause labels are hypotheses, not accounting conclusions or sign-off evidence; every break requires human review.
- Duplicate-key handling and precedence when quantity, amount, and date differences coexist are undefined.
What it does & when to use it
This skill reconciles general ledger and subledger extracts for the same entity, asset class, and date scope. It normalizes shared keys, comparison fields, and data types before performing a full outer join. Records are classified as matched, amount break, quantity break, timing break, GL only, or subledger only. The output includes a break report and summaries by break bucket and hypothesized cause.
Reads general ledger and subledger extracts for a common scope; normalizes dates, amounts, and identifiers; performs a full outer join at the lowest shared-grain key; tests equality using a default amount tolerance of 0.01 and quantity tolerance of 0; classifies matched and unmatched records; suggests likely causes such as timing, FX, mapping, duplicate or missing posts, fee or accrual, and data quality; produces a break report sorted by descending absolute base-amount delta and summary counts and totals by bucket and likely cause.
- A fund administration team reconciles the general ledger to a subledger after each trading day using security, account, and trade-date keys.
- A finance operations user investigates amount, quantity, and posting-date differences during month-end close.
- An asset-class operations team tests whether local amounts agree while base amounts differ, indicating a possible FX issue.
- A reconciliation owner prepares GL-only and subledger-only records for subsequent investigation and sign-off.
Pros & cons
- Covers normalization, matching, break classification, and summary generation in one workflow.
- Separates amount, quantity, timing, GL-only, and subledger-only breaks.
- Treats cause labels as investigation hypotheses rather than definitive conclusions.
- Provides explicit default tolerances for amounts and quantities.
- The source provides no test suite, sample inputs, or demonstrated run results.
- Users must supply extracted GL and subledger data; file formats and connectors are not specified.
- Cause classification is heuristic and requires human root-cause confirmation.
- The source does not define a specific asset-class coverage list or performance limits.
How to install
In Cowork, open Settings → Plugins → Add plugin, paste https://github.com/anthropics/financial-services, and select the relevant agent or vertical plugin. You may also upload a zip containing a directory under plugins/. The source does not document a standalone installation command for gl-recon.
How to use
Provide general ledger and subledger extracts covering the same entity, asset class, and date range, and state the reconciliation scope. Example: “Reconcile this GL extract to the subledger extract by security ID, account, and trade date using the default tolerances, then produce the break report and summary.”