Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →This skill should be used when the user asks tax-related questions, wants advice on deductions or tax savings, or needs expert guidance equivalent to a tax accountant (税理士) or life planner. Trigger phrases include: "税金について教えて", "控除は使える？", "確定申告の相談", "節税", "ふるさと納税の上限", "iDeCoの効果", "住宅ローン控除", "青色申告のメリット", "消費税はかかる？", "扶養に入れる？", "配偶者控除", "医療費控除", "法人成り", "経費になる？", "税率を教えて", "所得税の計算", "住民税", "社会保険料", "103万の壁", "130万の壁", "インボイス", "簡易課税", "税制改正", "開業届", "副業バレ", "白色申告", "税務調査", "特定支出控除", "予定納税", "中間納付"
.claude/skills/kazukinagata-tax-advisor/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 58% | 0% |
| case-07 | ✓→✗ | ▼ Worse | 12% | 0% |
| case-08 | ✓→✗ | ▼ Worse | 24% | 0% |
| case-09 | ✓→✗ | ▼ Worse | 4% | 0% |
| case-10 | ✓→✗ | ▼ Worse | 17% | 0% |
税理士・ライフプランナー相当の専門知識で、ユーザーの税務相談に回答するスキル。 令和7年分（2025年課税年度）の税制に基づく。
shinkoku.config.yaml を Read ツールで読み込む/setup スキルの実行を案内して終了するdb_path: MCP ツールの db_path 引数に使用output_dir: 進捗ファイル等の出力先ベースディレクトリ以下の構造で回答を組み立てる:
ユーザーは「収入」で質問するが、税制上の判定は「所得」で行う。以下のルールを適用する:
令和7年分（2025年）の税制改正を正確に反映する:
質問のカテゴリに応じて、以下のファイルを参照する:
| 質問カテゴリ | 参照ファイル | |-------------|-------------| | 所得税の仕組み・税率 | reference/income-tax.md | | 住民税の仕組み・税率 | reference/resident-tax.md | | 消費税・インボイス | reference/consumption-tax.md | | 源泉徴収・年末調整 | reference/withholding-tax.md |
| 質問カテゴリ | 参照ファイル | |-------------|-------------| | 所得控除の一覧・概要 | reference/income-deductions.md | | 税額控除の一覧・概要 | reference/tax-credits.md | | 医療費控除 | reference/medical-expenses.md | | 配偶者控除・特別控除 | reference/spouse.md | | 扶養控除 | reference/dependents.md | | 住宅ローン控除 | /tax-housing-loan-context を実行する | | 控除の最適化・組み合わせ | reference/deduction-optimizer.md | | 控除シミュレーションの手順 | reference/deduction-simulation-guide.md |
| 質問カテゴリ | 参照ファイル | |-------------|-------------| | 青色申告 | reference/blue-return.md | | 事業経費・家事按分 | reference/business-expenses.md | | 副業の事業所得vs雑所得判定 | reference/side-business-classification.md | | 電子帳簿保存法 | /tax-ebookkeeping-context を実行する | | 経費算入の可否判定（品目別） | reference/expense-deductibility-guide.md | | 業種別の経費ガイド | reference/industry-expense-guide.md |
| 質問カテゴリ | 参照ファイル | |-------------|-------------| | 確定申告の手続き | reference/filing-procedure.md | | 社会保険（扶養判定含む） | reference/social-insurance.md | | ライフプラン（iDeCo/NISA等） | reference/life-planning.md | | 開業届・副業の始め方・青色vs白色 | reference/startup-guide.md | | 予定納税・中間納付の管理 | reference/prepayment-management.md |
| 質問カテゴリ | 参照ファイル | |-------------|-------------| | 用語の定義・収入→所得変換 | reference/glossary.md | | よくある間違い | reference/common-mistakes.md | | 令和7年分の改正内容 | reference/tax-reform/2025.md | | 令和8年度税制改正大綱 | reference/tax-reform/2026.md | | 経過措置 | reference/tax-reform/transition.md | | 翌年以降の改正予定 | reference/tax-reform/upcoming.md | | 暗号資産の課税 | reference/crypto-tax.md | | 新NISAと確定申告 | reference/nisa-and-filing.md | | 免責事項 | /tax-legal-context を実行する |
103万・106万・130万・150万・201万等の「壁」は、それぞれ異なる制度に基づく。 以下のファイルを横断的に参照して回答する:
以下のような複合的な質問には、複数の reference ファイルを横断して回答する:
すべての回答の末尾に以下の免責事項を付記する:
---
⚠ この回答は一般的な税務情報の提供を目的としたものであり、個別の税務アドバイスではありません。
具体的な申告にあたっては、税理士等の専門家にご相談ください。
情報は令和7年分（2025年課税年度）の税制に基づいています。免責事項の詳細は /tax-legal-context を実行する。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 16,256 | 17,332 | +7% | 1 | 1 | 0% | 2,653 | 4,651 | +75% | 0 | 0 | — |
case-07 | pass→fail | 15,880 | 6,059 | -62% | 1 | 1 | 0% | 2,600 | 2,907 | +12% | 0 | 0 | — |
case-08 | pass→fail | 14,210 | 7,133 | -50% | 1 | 1 | 0% | 2,270 | 2,825 | +24% | 0 | 0 | — |
case-16 | fail→fail | 9,319 | 5,715 | -39% | 1 | 1 | 0% | 1,493 | 2,685 | +80% | 0 | 0 | — |
case-02 | pass→pass | 24,956 | 29,091 | +17% | 1 | 1 | 0% | 4,822 | 6,890 | +43% | 0 | 0 | — |
case-03 | fail→pass | 28,750 | 20,484 | -29% | 1 | 1 | 0% | 4,022 | 6,347 | +58% | 0 | 0 | — |
case-04 | fail→fail | 16,348 | 7,378 | -55% | 1 | 1 | 0% | 2,780 | 2,843 | +2% | 0 | 0 | — |
case-05 | pass→pass | 15,927 | 12,659 | -21% | 1 | 1 | 0% | 2,601 | 4,800 | +85% | 0 | 0 | — |
case-06 | pass→pass | 17,799 | 13,422 | -25% | 1 | 1 | 0% | 2,715 | 4,548 | +68% | 0 | 0 | — |
case-09 | pass→fail | 15,290 | 6,712 | -56% | 1 | 1 | 0% | 2,650 | 2,755 | +4% | 0 | 0 | — |
case-10 | pass→fail | 14,344 | 5,828 | -59% | 1 | 1 | 0% | 2,374 | 2,782 | +17% | 0 | 0 | — |
case-11 | pass→fail | 13,961 | 7,777 | -44% | 1 | 1 | 0% | 2,194 | 2,667 | +22% | 0 | 0 | — |
case-12 | pass→pass | 11,949 | 15,607 | +31% | 1 | 1 | 0% | 2,135 | 5,045 | +136% | 0 | 0 | — |
case-13 | pass→fail | 14,104 | 6,866 | -51% | 1 | 1 | 0% | 2,390 | 2,690 | +13% | 0 | 0 | — |
case-14 | pass→fail | 11,468 | 4,162 | -64% | 1 | 1 | 0% | 1,892 | 2,677 | +41% | 0 | 0 | — |
case-15 | pass→fail | 17,663 | 6,227 | -65% | 1 | 1 | 0% | 2,977 | 2,814 | -5% | 0 | 0 | — |
case-17 | pass→fail | 17,025 | 6,925 | -59% | 1 | 1 | 0% | 2,779 | 2,643 | -5% | 0 | 0 | — |
case-18 | pass→pass | 14,417 | 10,317 | -28% | 1 | 1 | 0% | 2,349 | 3,731 | +59% | 0 | 0 | — |
case-19 | pass→fail | 17,291 | 6,606 | -62% | 1 | 1 | 0% | 2,419 | 2,810 | +16% | 0 | 0 | — |
case-20 | fail→fail | 17,067 | 6,014 | -65% | 1 | 1 | 0% | 2,867 | 2,816 | -2% | 0 | 0 | — |
case-21 | fail→fail | 21,112 | 9,731 | -54% | 1 | 1 | 0% | 3,536 | 2,806 | -21% | 0 | 0 | — |
case-22 | fail→fail | 24,160 | 8,503 | -65% | 1 | 1 | 0% | 3,542 | 2,924 | -17% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 6 counted toward the lift figure. The other 16 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of -41 percentage points is the difference between those two pass rates over the 6 comparable cases. 16 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.