GL Reconciliation Assistant
Reconcile the general ledger against subledger data and classify breaks.
The skill treats GL and subledger/custodian extracts as untrusted data and describes analysis/reporting rather than external writes, broad permissions, or destructive defaults, supporting a relatively strong baseline; however, it does not specify sensitive financial-data minimization, retention, isolation, access control, user confirmation, rollback, or complete data-flow handling, so 7 points are deducted.
The steps, key fields, comparison fields, full-outer-join buckets, tolerances, and deliverables are broadly consistent, making the happy path plausible; however, empty files, duplicate keys, missing fields, malformed values, conflicting classifications, and diagnostic failure feedback are unspecified, and the duplicate/missing-post cause lacks a precise decision rule. The static score remains below the 10-point execution ceiling.
The target scenario, input scope, matching grain, break buckets, and downstream handoffs are reasonably clear; however, non-fit cases, trigger exclusions, file/interface requirements, Chinese-language behavior, and mainland-China environment fit are not declared, so 7 points are deducted.
The documentation is concise and stepwise, and repository context supplies a license plus maintenance and validation guidance; however, the skill itself lacks examples, a parameter contract, FAQs, known limitations, versioning, changelog, and explicit maintenance ownership, so 8 points are deducted.
The skill specifies a directly usable break report and bucket/cause summary, including sorting and matched-percentage requirements; however, there are no representative outputs, test results, or comparative-benefit evidence, and duplicate-key and complex-netting cases may require substantial manual review. The static ceiling applies and 9 points are deducted.
The source rules, bucket conditions, cause labels, and output requirements are auditable, while repository-level CI provides plugin-validation and secret-scanning evidence; however, there are no committed key-path tests, sample fixtures, third-party execution results, or independent corroboration, so only limited static verifiability is supported and 6 points are deducted.
- Input formats, missing or empty fields, duplicate keys, invalid numerics, and many-to-one matching are unspecified; implementation should add validation, diagnostic errors, and human-review rules.
- The default amount tolerance of 0.01 and zero quantity tolerance may not fit every asset class, currency precision, or accounting policy; the applicable policy should be confirmed and recorded.
- The skill may process sensitive GL and subledger data but does not define redaction, access control, retention, or transfer boundaries.
- No Chinese-output convention or mainland-China network dependency is stated; the skill itself does not show a mandatory overseas-service dependency.
What does this skill do, and when should you use it?
This skill reconciles general-ledger extracts with subledger extracts for a trade date or reporting period. It normalizes shared keys and comparison fields, performs a full outer join, and assigns each row to matched, amount break, quantity break, timing break, GL only, or subledger only. Breaks receive a likely-cause hypothesis for follow-up. The outputs are a detailed break report and summaries by bucket and likely cause.
Reads GL and subledger extracts for the same entity, asset class, and date scope; aligns keys and comparison columns such as quantity, local amount, base amount, FX rate, and posting date; normalizes dates, amounts, and identifiers; performs a full outer join; applies a default amount tolerance of 0.01 and quantity tolerance of 0; identifies matches and six break buckets; classifies breaks as likely timing, FX, mapping, duplicate or missing post, fee or accrual, or data-quality issues; and produces a break report plus summary statistics.
- A fund-operations team runs a daily position- or transaction-level reconciliation after trade activity.
- A month-end close team identifies amount, quantity, and posting-date differences before sign-off.
- An accounting or operations analyst investigates whether local amounts agree while base amounts differ because of FX sourcing or dating.
- A reconciliation owner routes material breaks to break-trace and sends the summary to the resolver for sign-off packaging.
- A multi-asset operations team classifies breaks using shared keys such as security, account, and trade date.
What are this skill's strengths and limitations?
- Supports both position-level and transaction-level reconciliation.
- Defines six explicit result buckets and normalization rules.
- Provides default tolerances and a consistent break-report sorting rule.
- Separates likely-cause hypotheses from final root-cause resolution and supports downstream handoffs.
- Licensed under Apache-2.0 and installable through the repository's plugin distribution.
- Requires GL and subledger extracts with a matching scope; the source does not specify a concrete input-file format.
- Cause labels are hypotheses for the resolver, not accounting conclusions.
- The source provides no test suite, performance benchmarks, or platform validation results.
- The source does not document a specific duplicate-detection algorithm or an executable standalone command.
How do you install this skill?
The skill is located at plugins/agent-plugins/gl-reconciler/skills/gl-recon/SKILL.md and is bundled with the gl-reconciler agent plugin. In Cowork, open Settings → Plugins → Add plugin, add https://github.com/anthropics/financial-services, and select the relevant agent or plugin; alternatively upload a zip containing plugins/agent-plugins/gl-reconciler/. In Claude Code, run: claude plugin marketplace add anthropics/financial-services; claude plugin install gl-reconciler@claude-for-financial-services. The README does not document a separate installation command for the gl-recon skill.
How do you use this skill?
Provide GL and subledger extracts covering the same entity, asset class, and date or period. Example prompt: Reconcile the GL and subledger extracts for the specified trade date using security_id, account, and trade_date; produce the break report and summary, sorted by absolute base-amount delta descending. Use the firm's policy when supplied; otherwise use the default amount tolerance of 0.01 and quantity tolerance of 0.