feat(agent): エージェント定義とオーケストレーター導線を追加

[yama]

This pull request introduces a comprehensive multi-agent architecture for development workflows, defining clear roles, responsibilities, and processes for each agent involved in tasks such as investigation, planning, implementation, review, testing, and roadmap management. It also adds corresponding skills and interface definitions to support these agents, ensuring structured collaboration and robust quality control.

Agent Definitions and Processes

• Added detailed agent definitions for orchestrator, explorer, planner, worker, reviewer, tester, and roadmap curator, each with their roles, rules, input/output, and prohibited actions, located in .agent/agents/*.md. This establishes a clear separation of concerns and standardized procedures for multi-agent collaboration. [1] [2] [3] [4] [5] [6] [7] [8]

Skill and Interface Definitions

• Introduced the agent-orchestrator skill, providing a structured approach to task orchestration, agent selection, and workflow management, including command definitions and execution rules in .codex/skills/agent-orchestrator/SKILL.md and its interface in .codex/skills/agent-orchestrator/agents/openai.yaml. [1] [2]
• Added the review-agent skill, focusing on Japanese-language code review with severity ordering, file/line references, and clear output formats, supported by its interface in .codex/skills/review-agent/SKILL.md and .codex/skills/review-agent/agents/openai.yaml. [1] [2]

Workflow and Collaboration Policies

• Defined policies for agent collaboration, including how to declare task start, select and exclude agents, coordinate file editing boundaries, and ensure documentation and plan consistency across all stages of the workflow. [1] [2] [3]

Quality and Safety Guidelines

• Embedded rules in each agent and skill to prioritize safety, security, cache consistency, and single source of truth (SSOT) principles, with explicit prohibited actions to prevent common pitfalls and ensure robust review and testing. [1] [2] [3] [4] [5]

Documentation and Traceability

• Established conventions for documenting agent runs, investigation results, plans, reviews, and roadmap updates to ensure traceability and facilitate future reuse of knowledge and decisions. [1] [2] [3] [4]

These changes lay the foundation for a scalable, auditable, and high-quality multi-agent development process.

[Copilot]

Pull request overview

開発ワークフロー向けのマルチエージェント構成（役割定義と導線）を追加し、調査・計画・実装・レビュー・検証・ロードマップ管理の分担ルールを明文化するPR

Changes:

• .agent/agents/ に orchestrator / explorer / planner / worker / reviewer / tester / roadmap-curator のエージェント定義を追加
• .codex/skills/agent-orchestrator を追加し、エージェント選定・書き込み範囲分離・進行宣言の手順を定義
• .codex/skills/review-agent を追加し、差分レビュー手順と出力フォーマットを定義

Reviewed changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
.codex/skills/agent-orchestrator/SKILL.md	オーケストレーション手順・開始宣言・コマンドを定義
.codex/skills/agent-orchestrator/agents/openai.yaml	agent-orchestrator のインターフェース定義と default_prompt を追加
.codex/skills/review-agent/SKILL.md	reviewer 方針に沿ったレビュー手順・出力形式を定義
.codex/skills/review-agent/agents/openai.yaml	review-agent のインターフェース定義と default_prompt を追加
.agent/agents/README.md	エージェント定義ディレクトリの運用方針と呼び出し方を追加
.agent/agents/orchestrator.md	タスク分解・担当割り当て・統合の責務を定義
.agent/agents/explorer.md	調査専任の責務・入力・禁止事項を定義
.agent/agents/planner.md	ExecPlan 作成・更新の責務とルールを定義
.agent/agents/worker.md	実装担当の規約（SSOT/DB/キャッシュ等）を定義
.agent/agents/reviewer.md	レビュー観点・成果物・禁止事項を定義
.agent/agents/tester.md	検証・再現・テスト失敗切り分けの責務を定義
.agent/agents/roadmap-curator.md	ロードマップと ExecPlan 整合性管理の責務を定義