Financial Variance Commentary
Identifies material financial variances and explains the underlying business drivers.
The skill is limited to drafting commentary tables and narrative, does not instruct ledger posting or external submission, and tells the agent to flag unclear drivers rather than invent them. However, the internal-gl MCP dependency lacks least-privilege details, data-flow disclosure, sensitive-financial-data handling, user confirmation, recovery, and rollback guidance, so points are deducted.
Inputs, thresholds, output columns, and the unclear-driver fallback are specified, making the happy path plausible. Reliability is reduced because MCP availability, field contracts, missing data, zero values, denominator-zero cases, and key-path tests are unspecified; static review also caps this score.
The audience and month-end management-reporting scenario are clear, and the threshold plus always-comment rules are usable. Boundaries, non-fit cases, input formats, exclusion triggers, Chinese-language support, and mainland-China reachability are not documented; the core driver analysis also depends on internal-gl MCP, so points are deducted.
The skill is concise and layered around thresholds, fields, driver sourcing, and output; repository-level Apache-2.0 licensing and official provenance are available. The skill lacks its own version, changelog, maintenance owner, installation/dependency notes, examples, and FAQ, resulting in a mid-level score.
It specifies a directly usable variance-commentary table and a 3–5 sentence summary, with drivers tied to underlying activity rather than merely restating numbers. No representative output, calculation example, or execution evidence is supplied, and unavailable MCP data could leave the result incomplete, so points are deducted.
The rules and output structure are auditable from the text, and the repository includes general plugin-validation and secret-scan workflows. There are no skill-specific key-path tests, execution results, independent reproductions, or corroborating evidence, so verifiability remains limited.
- Before use, verify internal-gl MCP permissions, data residency and sensitive-financial-data handling, including reachability in the target network environment.
- Add input schemas, materiality-threshold precedence, zero or missing-value handling, rounding rules, and a clear fallback when MCP is unavailable.
- Require controller or other qualified finance review; the skill should not automatically publish reports or make ledger changes.
What it does & when to use it
This skill produces variance commentary for each P&L and balance-sheet line using current-period actuals, prior-period actuals, and budget for the same scope. It flags lines that meet the materiality threshold or appear on the always-comment list: revenue, headcount cost, and cash. The output contains the three values, variance amounts and percentages against prior period and budget, a one-sentence activity-based driver, and a short narrative of the biggest movers. It is intended for month-end close packages and management reporting and relies on an internal GL MCP for detailed activity evidence.
Reads current-period actuals, prior-period actuals, and budget for the same scope; filters lines using the materiality rule or always-comment list; calculates amount and percentage changes versus prior period and budget; examines journal-source breakdowns, vendor mix, headcount changes, or volume-by-rate activity through the internal GL MCP; produces a commentary table and a 3–5 sentence summary; records “driver unclear — flag for controller” when the cause cannot be established from the data.
- A fund-administration or finance-operations professional preparing variance explanations for the month-end close package.
- A management-reporting preparer comparing actual results with the prior period and budget.
- A controller reviewing required commentary for revenue, headcount cost, and cash.
- A finance team with internal GL MCP access that needs to trace variances to journal sources, vendors, staffing, or operating volume.
Pros & cons
- Compares current, prior, and budget values in one structured output.
- Requires explanations tied to underlying activity rather than repeating the variance.
- Automatically includes revenue, headcount cost, and cash on the always-comment list.
- Escalates unclear causes instead of inventing explanations.
- Reliable driver analysis depends on access to an internal GL MCP and sufficiently detailed source data.
- The source does not document MCP configuration, permissions, or interface details.
- No test suite, example inputs and outputs, or standalone deployment procedure is provided.
- The default materiality rule may need to be adjusted to firm policy.
How to install
The skill is located at plugins/vertical-plugins/fund-admin/skills/variance-commentary/SKILL.md. The README documents installing the repository as a Cowork plugin or adding its Claude Code marketplace with: claude plugin marketplace add anthropics/financial-services. It does not document a standalone installation command for this individual skill.
How to use
Provide current-period actuals, prior-period actuals, and budget for the same scope, together with the firm's materiality threshold. Ask for variance commentary for a month-end close package or management report. Unless another threshold is supplied, the default is 5% of the line or a fixed floor, whichever is greater; the skill may use the internal GL MCP to identify drivers.