JavaLangRuntimeException

docs/spec/ の仕様書を元に、docs/work/YYYYMMDD_<feature>.md として実装計画書を作成する。Phase 分解・影響範囲・テスト戦略・リスクを構造化し、実装前にユーザー承認を取る。「実装計画立てて」「work 書いて」「どう進めるか計画して」などで起動。

差分や指定コードを構造化されたレビュー観点（正確性・設計・テスト・可読性・セキュリティ）で読み、優先度付きで指摘する。修正は提案するが勝手に書き換えない。「レビューして」「コードレビュー」「セカンドオピニオン」などで起動。

未知のコードベースを最短で把握するための構造化探索を行う。言語・ビルドツール・アーキテクチャ・テスト方針・主要なエントリポイントを短時間でレポート化する。「このプロジェクト教えて」「コードベース調べて」「初見で入ったから概要ほしい」などで起動。

未コミットの変更を依存関係と関心事に基づいて適切な粒度のコミットに分割する。自動生成物・手動実装・テスト・ドキュメントを分離し、レビューしやすい履歴を作る。「コミット分けて」「良い粒度でコミット」「この変更コミットにして」などで起動。

バグ調査を場当たり的でなく体系的に進める。再現→仮説→検証→修正→回帰テストの順序を守り、仮説と事実を分けて記録する。「バグ調査して」「なぜか動かない」「デバッグ手伝って」などで起動。

汎用的な開発オーケストレーター。docs/spec/ の仕様と docs/work/ の実装計画書を軸に、Phase 分解と PDCA サイクルで実装を進める。自分は実装せず、各 Phase を Task で専門 agent / Explore / Plan に委託する。「開発進めて」「実装オーケストレートして」「機能実装して」などで起動。

実 DB を起動して Handler → Usecase → Service → Repository を貫通させる integration test を書く。fixture 分離、TestMain/global setup、トランザクション単位のロールバックによるテスト間隔離、API エンドポイントを HTTP 経由で叩く E2E 的検証を扱う。単体テスト(mock 前提)は対象外で `backend-test-writer` に委譲する。「integration test 書いて」「E2E テスト追加して」「DB 込みのテスト書いて」などで起動。

コミット履歴と差分から質の高い Pull Request 説明文を生成する。Summary / Changes / Test plan / Risk を構造化し、レビュアーが 30 秒で理解できる PR を作る。「PR 説明書いて」「PR description 作って」「pull request の本文を整えて」などで起動。

Orchestrator skill for the Abstract section - delegates writing to the abstract subagent (this file is the rulebook the subagent loads at startup), routes open questions to paper-researcher and reviews to paper-reviewer1/2; the main conversation never writes prose itself. Use this skill whenever the user is writing or revising an abstract for an academic paper (international conference full paper, journal article, workshop paper). Triggers include mentions of "abstract", "アブスト", "要旨", "要約", requests to "write/revise/fix the abstract", or when the user shares an abstract draft for review. Also trigger when the user is finalizing a paper and needs the abstract aligned with the rest of the manuscript. This skill enforces a 5-part structure that leads with the technical contribution (NOT the user study details) and keeps contributions structurally clear (numbered list OR prose enumeration — both are acceptable).

Orchestrator skill for the Conclusion section - delegates writing to the conclusion subagent (this file is the rulebook the subagent loads at startup), routes open questions to paper-researcher and reviews to paper-reviewer1/2; the main conversation never writes prose itself. Use this skill whenever the user is writing or revising the Conclusion section of a paper. Triggers include "Conclusion", "結論", "まとめ", "締め", "最終章", or when the user says "the paper is almost done, just need to write the conclusion". This skill ensures the conclusion does three things in order: (1) briefly restate the problem and approach, (2) summarize key findings in relation to the hypotheses, (3) gesture toward broader implications. It should be concise (typically 2-4 paragraphs) and must align with the Abstract and Contributions bullet list in the Introduction.

Orchestrator skill for the Discussion section - delegates writing to the discussion subagent (this file is the rulebook the subagent loads at startup), routes open questions to paper-researcher and reviews to paper-reviewer1/2; the main conversation never writes prose itself. Use this skill whenever the user is writing or revising the Discussion section of a paper — the section that interprets the Results, explains underlying mechanisms, and connects findings to prior work. Triggers include "Discussion", "考察", "6章", "interpret results", "mechanism", or when the user is explaining *why* their results came out a particular way. Critically, this skill enforces precise interpretation of cited works (checking whether a prior study is truly contrasted or actually consistent with the current findings) and separates mechanistic explanation from the descriptive Results section. Also use this skill when explaining interaction effects, lack-of-interaction findings, or reconciling findings with or against prior literatur

Orchestrator skill for the Introduction section - delegates writing to the introduction subagent (this file is the rulebook the subagent loads at startup), routes open questions to paper-researcher and reviews to paper-reviewer1/2; the main conversation never writes prose itself. Use this skill whenever the user is writing or revising the Introduction section of an academic paper. Triggers include "write the intro", "イントロを書く", "introduction を直す", "第1章", "Heilmeier", requests to review intro drafts, or when a user explicitly says the abstract is done and they want to move to the introduction. This skill enforces the Heilmeier's Catechism structure (4 paragraphs corresponding to questions 1-4), and critically, requires Contributions at the end as a bulleted list (not Research Questions). Readers want to know conclusions, not questions — so the intro ends with what was found, not what was asked.

Orchestrator skill for the Limitations section - delegates writing to the limitations subagent (this file is the rulebook the subagent loads at startup), routes open questions to paper-researcher and reviews to paper-reviewer1/2; the main conversation never writes prose itself. Use this skill whenever the user is writing or revising the Limitations, Limitations and Future Work, or Future Directions section of a paper. Triggers include "Limitations", "Future Work", "limitations", "将来研究", "今後の課題", "制約", "将来の方向性". Also trigger when a reviewer asks the authors to "discuss generalizability" or "address sample limitations". This skill helps the user acknowledge limitations honestly (avoiding both over-apology and hiding weaknesses) and frame them as motivated future directions rather than pure weaknesses.

Orchestrator skill for the Method / System / Experimental Setup section - delegates writing to the method subagent (this file is the rulebook the subagent loads at startup), routes open questions to paper-researcher and reviews to paper-reviewer1/2; the main conversation never writes prose itself. Use this skill whenever the user is writing the Method, System, Experimental Setup, Apparatus, or Implementation section of a paper — the chapter that describes the technical system or platform built before running the user study. Triggers include "Method", "Experimental Setup", "System", "実装", "手法", "システム概要", "装置", "3章", or "setup". This section should justify design decisions (not merely describe them), explicitly link the system design back to the research gap from the Introduction, and set up the independent variables that the subsequent user study will manipulate.

Paper-writing pipeline manager (a.k.a. paper-write-orchestrator / paper-write-manager). Use when the user wants the MSPW agent team to write, extend, revise, or review an academic manuscript — triggers include "論文を進めて", "〜章を書いて/直して", "レビューして反映して", "specを実装して", "/paper-orchestrator". Orchestrates the section-writer subagents (abstract … conclusion), paper-researcher, paper-reviewer1/2 and the writing-principles checker via the Agent tool, relays ALL inter-agent communication, and batches judgment calls to the user via AskUserQuestion. The orchestrator itself never writes manuscript prose, never reviews content, and never answers research questions from memory.

Spec-writing pipeline manager for paper changes. Use BEFORE editing the manuscript when a change deserves a design document — restructuring sections, changing claims/contributions, adding analyses/experiments, major revisions, reviewer-response plans. Triggers include "specを作って", "設計書", "変更計画", "リバイズ方針をまとめて", "/paper-spec-orchestrator". Produces docs/spec/NNNN-<slug>.md via the paper-spec-writer subagent, resolves unknowns via paper-researcher, and gets the spec approved by the user via AskUserQuestion. Never edits the manuscript itself; implementation is handed off to /paper-orchestrator.