What does this skill do, and when should you use it?
This skill uses current-period actuals, prior-period actuals, and budget for the same scope to produce a variance commentary table. It flags lines that meet the materiality threshold or appear on the always-comment list: revenue, headcount cost, and cash. To explain why a line moved, it consults underlying activity through the internal-gl MCP, including journal-source breakdowns, vendor mix, headcount changes, and volume-by-rate analysis. When the evidence is insufficient, it writes “driver unclear — flag for controller” instead of inventing an explanation.
Reads current-period actuals, prior-period actuals, and budget for a consistent scope; applies the supplied materiality threshold, defaulting to 5% of the line or a fixed floor, whichever is greater; flags relevant P&L and balance-sheet lines; calculates amount and percentage changes versus prior period and budget; uses the internal-gl MCP to inspect underlying activity; produces a commentary table and a 3–5 sentence narrative covering the period’s biggest movers.
- A financial controller preparing the month-end close package needs commentary on material P&L and balance-sheet movements.
- A management-reporting team needs current, prior-period, and budget comparisons in one monthly review output.
- A finance analyst needs to trace a variance to journal sources, vendor mix, headcount changes, or volume-by-rate activity.
- A controller needs explicit escalation when the available data does not establish a credible driver.
How do you install this skill?
- Before use, verify that the internal-gl MCP exists, is least-privilege, and document the data it reads and any cross-border transfer.
- Add rules for zero denominators, negative balances, reclassifications, missing budgets, and other abnormal inputs; controller review should cover every “driver unclear” result.
- Treat the output as a draft requiring accounting review and approval; it should not flow directly into close records or management reporting without sign-off.
- MCP Server
internal-gl MCP
In Cowork, open Settings → Plugins → Add plugin, paste https://github.com/anthropics/financial-services, and select Month-End Closer from the available plugins or agents. The README does not provide a standalone Claude Code install command for Month-End Closer; it documents adding the marketplace and then installing selected named agents.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/anthropics/financial-services.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/plugins/agent-plugins/month-end-closer/skills/variance-commentary" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Using these current, prior, and budget figures, produce variance commentary for lines at or above 5% and for revenue, headcount cost, and cash. Use the internal-gl MCP to explain the drivers and flag unsupported drivers for the controller.
Invoke it for a month-end close or management-reporting task and provide current-period actuals, prior-period actuals, budget, the materiality threshold, and any available underlying activity data. Example: “Using these current, prior, and budget figures, produce variance commentary for lines at or above 5% and for revenue, headcount cost, and cash. Use the internal-gl MCP to explain the drivers and flag unsupported drivers for the controller.”
What are this skill's strengths and limitations?
- Compares current, prior-period, and budget values with both amount and percentage changes.
- Requires drivers to explain underlying activity rather than restating the variance.
- Includes explicit always-comment treatment for revenue, headcount cost, and cash.
- Provides a defined fallback when the data does not support a driver.
- Depends on the internal-gl MCP for underlying activity, but the source does not document its setup, permissions, or coverage.
- The SKILL.md provides no test suite, sample output, or platform validation evidence.
- The fixed-floor component of the default materiality threshold is not numerically defined.
- It cannot establish a driver beyond the activity data supplied or exposed through the MCP.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Month-End Variance Commentary this page ✓ Anthropic · Official | 42 · Not recommended | ★ 40k | 1mo ago | Apache-2.0 |
| Finance & Accounting Super-Skill | 46 · Use with care | ★ 26 | 7mo ago | — |
| Financial Model Auditor ✓ Anthropic · Official | 53 · Use with care | ★ 40k | 1mo ago | Apache-2.0 |
| Model Checker for Financial Audits | 53 · Use with care | ★ 1.8k | 11d ago | Apache-2.0 |
| 3-Statement Financial Model Skill | 51 · Use with care | ★ 1.8k | 11d ago | Apache-2.0 |
How did FollowSkills review this skill?
The skill only produces a commentary table and narrative and does not instruct the agent to post entries or perform external actions; the README requires human review and sign-off, reducing direct-effect risk. However, it does not disclose financial-data flows, MCP permission scope, user confirmation, isolation, or recovery controls, so points are deducted.
Inputs, thresholds, output columns, and the “driver unclear” fallback are reasonably specified. However, the required internal-gl MCP is not verified in the supplied connector inventory, and there are no skill-specific tests, abnormal-input rules, or diagnosable error messages. The score is capped by static review and reduced for dependency uncertainty.
The target use case is clear: month-end close and management reporting. Inputs and trigger conditions are also fairly specific, including materiality and the always-comment list. Non-fit boundaries, Chinese-language behavior, mainland-China reachability, and the required internal MCP’s availability are not established, so points are deducted.
The SKILL.md has a concise structure covering thresholds, fields, driver guidance, and output format. The repository provides an Apache-2.0 license, contribution workflow, sync tooling, and CI validation, and publisher provenance is verified. The skill lacks its own version, changelog, installation/dependency notes, examples, FAQ, maintenance owner, and limitation disclosure, so points are deducted.
The skill specifies a directly usable deliverable: line-level variances with amounts, percentages, activity-based drivers, and a 3–5 sentence summary. However, driver sourcing depends on an unverified internal-gl MCP, and there is no representative output or execution evidence; human review remains necessary. The static-review cap keeps this at or below 7, with further deduction for uncertain completeness and usability.
The revision and assessed path are explicit, repository CI validates plugin structure, and the README documents human sign-off and connector architecture. There is no skill-specific test suite, reproducible example, or third-party execution evidence, so conclusions rely mainly on static documentation review and points are deducted.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →