diff --git a/CHANGELOG.md b/CHANGELOG.md
index 4321142..14ff812 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -4,6 +4,46 @@ All notable changes to this project will be documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
+## [0.10.0] - 2026-02-09
+
+### Added
+
+- **`structural-reform` builtin piece**: Full project review and structural reform — iterative codebase restructuring with staged file splits, powered by `loop_monitors`
+- **`unit-test` builtin piece**: Unit test focused piece — test analysis → test implementation → review → fix, with `loop_monitors` for cycle control
+- **`test-planner` persona**: Specialized persona for analyzing codebase and planning comprehensive test strategies
+- **Interactive mode variants**: Four selectable modes after piece selection — `assistant` (default: AI-guided requirement refinement), `persona` (conversation with first movement's persona), `quiet` (generate instructions without questions), `passthrough` (user input used as-is)
+- **`persona_providers` config**: Per-persona provider overrides (e.g., `{ coder: 'codex' }`) — route specific personas to different providers without creating hybrid pieces
+- **`task_poll_interval_ms` config**: Configurable polling interval for `takt run` to detect new tasks during execution (default: 500ms, range: 100–5000ms)
+- **`interactive_mode` piece field**: Piece-level default interactive mode override (e.g., set `passthrough` for pieces that don't benefit from AI planning)
+- **Task-level output prefixing**: Colored `[taskName]` prefix on all output lines during parallel `takt run` execution, preventing mid-line interleaving between concurrent tasks
+- **Review policy facet**: Shared review policy (`builtins/{lang}/policies/review.md`) for consistent review criteria across pieces
+
+### Changed
+
+- **BREAKING:** Removed all Hybrid Codex pieces (`*-hybrid-codex`) — replaced by `persona_providers` config which achieves the same result without duplicating piece files
+- Removed `tools/generate-hybrid-codex.mjs` (no longer needed with `persona_providers`)
+- Improved parallel execution output: movement-level prefix now includes task context and iteration info in concurrent runs
+- Codex client now detects stream hangs (10-minute idle timeout) and distinguishes timeout vs external abort in error messages
+- Parallel task execution (`takt run`) now polls for newly added tasks during execution instead of only checking between task completions
+- Parallel task execution no longer enforces per-task time limits (previously had a timeout)
+- Issue references now routed through interactive mode (as initial input) instead of skipping interactive mode entirely
+- Builtin `config.yaml` updated to document all GlobalConfig fields
+- Extracted `conversationLoop.ts` for shared conversation logic across interactive mode variants
+- Line editor improvements: additional key bindings and edge case fixes
+
+### Fixed
+
+- Codex processes hanging indefinitely when stream becomes idle — now aborted after 10 minutes of inactivity, releasing worker pool slots
+
+### Internal
+
+- New test coverage: engine-persona-providers, interactive-mode (532 lines), task-prefix-writer, workerPool expansion, pieceResolver expansion, lineEditor expansion, parallel-logger expansion, globalConfig-defaults expansion, pieceExecution-debug-prompts expansion, it-piece-loader expansion, runAllTasks-concurrency expansion, engine-parallel
+- Extracted `TaskPrefixWriter` for task-level parallel output management
+- Extracted `modeSelection.ts`, `passthroughMode.ts`, `personaMode.ts`, `quietMode.ts` from interactive module
+- `InteractiveMode` type model added (`src/core/models/interactive-mode.ts`)
+- `PieceEngine` validates `taskPrefix`/`taskColorIndex` pair consistency at construction
+- Implementation notes document added (`docs/implements/retry-and-session.ja.md`)
+
 ## [0.9.0] - 2026-02-08
 
 ### Added
diff --git a/README.md b/README.md
index 59c39ba..602309b 100644
--- a/README.md
+++ b/README.md
@@ -92,13 +92,25 @@ takt
 takt hello
 ```
 
-**Note:** Issue references (`#6`) and `--task` / `--issue` options skip interactive mode and execute the task directly. All other inputs (including text with spaces) enter interactive mode for requirement refinement.
+**Note:** `--task` option skips interactive mode and executes the task directly. Issue references (`#6`, `--issue`) are used as initial input in interactive mode.
 
 **Flow:**
 1. Select piece
-2. Refine task content through conversation with AI
-3. Finalize task instructions with `/go` (you can also add additional instructions like `/go additional instructions`), or use `/play <task>` to execute a task immediately
-4. Execute (create worktree, run piece, create PR)
+2. Select interactive mode (assistant / persona / quiet / passthrough)
+3. Refine task content through conversation with AI
+4. Finalize task instructions with `/go` (you can also add additional instructions like `/go additional instructions`), or use `/play <task>` to execute a task immediately
+5. Execute (create worktree, run piece, create PR)
+
+#### Interactive Mode Variants
+
+| Mode | Description |
+|------|-------------|
+| `assistant` | Default. AI asks clarifying questions before generating task instructions. |
+| `persona` | Conversation with the first movement's persona (uses its system prompt and tools). |
+| `quiet` | Generates task instructions without asking questions (best-effort). |
+| `passthrough` | Passes user input directly as task text without AI processing. |
+
+Pieces can set a default mode via the `interactive_mode` field in YAML.
 
 #### Execution Example
 
@@ -451,8 +463,10 @@ TAKT includes multiple builtin pieces:
 | `passthrough` | Thinnest wrapper. Pass task directly to coder as-is. No review. |
 | `compound-eye` | Multi-model review: sends the same instruction to Claude and Codex simultaneously, then synthesizes both responses. |
 | `review-only` | Read-only code review piece that makes no changes. |
+| `structural-reform` | Full project review and structural reform: iterative codebase restructuring with staged file splits. |
+| `unit-test` | Unit test focused piece: test analysis → test implementation → review → fix. |
 
-**Hybrid Codex variants** (`*-hybrid-codex`): Each major piece has a Codex variant where the coder agent runs on Codex while reviewers use Claude. Available for: default, minimal, expert, expert-cqrs, passthrough, review-fix-minimal, coding.
+**Per-persona provider overrides:** Use `persona_providers` in config to route specific personas to different providers (e.g., coder on Codex, reviewers on Claude) without duplicating pieces.
 
 Use `takt switch` to switch pieces.
 
@@ -475,6 +489,7 @@ Use `takt switch` to switch pieces.
 | **research-planner** | Research task planning and scope definition |
 | **research-digger** | Deep investigation and information gathering |
 | **research-supervisor** | Research quality validation and completeness assessment |
+| **test-planner** | Test strategy analysis and comprehensive test planning |
 | **pr-commenter** | Posts review findings as GitHub PR comments |
 
 ## Custom Personas
@@ -543,8 +558,15 @@ branch_name_strategy: romaji  # Branch name generation: 'romaji' (fast) or 'ai'
 prevent_sleep: false     # Prevent macOS idle sleep during execution (caffeinate)
 notification_sound: true # Enable/disable notification sounds
 concurrency: 1           # Parallel task count for takt run (1-10, default: 1 = sequential)
+task_poll_interval_ms: 500  # Polling interval for new tasks during takt run (100-5000, default: 500)
 interactive_preview_movements: 3  # Movement previews in interactive mode (0-10, default: 3)
 
+# Per-persona provider overrides (optional)
+# Route specific personas to different providers without duplicating pieces
+# persona_providers:
+#   coder: codex             # Run coder on Codex
+#   ai-antipattern-reviewer: claude  # Keep reviewers on Claude
+
 # API Key configuration (optional)
 # Can be overridden by environment variables TAKT_ANTHROPIC_API_KEY / TAKT_OPENAI_API_KEY
 anthropic_api_key: sk-ant-...  # For Claude (Anthropic)
diff --git a/docs/README.ja.md b/docs/README.ja.md
index 3395595..e492d77 100644
--- a/docs/README.ja.md
+++ b/docs/README.ja.md
@@ -92,13 +92,25 @@ takt
 takt hello
 ```
 
-**注意:** Issue 参照（`#6`）や `--task` / `--issue` オプションを指定すると、対話モードをスキップして直接タスク実行されます。それ以外の入力（スペースを含む文字列を含む）はすべて対話モードに入ります。
+**注意:** `--task` オプションを指定すると対話モードをスキップして直接タスク実行されます。Issue 参照（`#6`、`--issue`）は対話モードの初期入力として使用されます。
 
 **フロー:**
 1. ピース選択
-2. AI との会話でタスク内容を整理
-3. `/go` でタスク指示を確定（`/go 追加の指示` のように指示を追加することも可能）、または `/play <タスク>` で即座に実行
-4. 実行（worktree 作成、ピース実行、PR 作成）
+2. 対話モード選択（assistant / persona / quiet / passthrough）
+3. AI との会話でタスク内容を整理
+4. `/go` でタスク指示を確定（`/go 追加の指示` のように指示を追加することも可能）、または `/play <タスク>` で即座に実行
+5. 実行（worktree 作成、ピース実行、PR 作成）
+
+#### 対話モードの種類
+
+| モード | 説明 |
+|--------|------|
+| `assistant` | デフォルト。AI が質問を通じてタスク要件を明確にしてから指示を生成。 |
+| `persona` | 最初のムーブメントのペルソナとの会話（ペルソナのシステムプロンプトとツールを使用）。 |
+| `quiet` | 質問なしでタスク指示を生成（ベストエフォート）。 |
+| `passthrough` | ユーザー入力をそのままタスクテキストとして使用。AI 処理なし。 |
+
+ピースの `interactive_mode` フィールドでデフォルトモードを設定可能。
 
 #### 実行例
 
@@ -451,8 +463,10 @@ TAKTには複数のビルトインピースが同梱されています:
 | `passthrough` | 最小構成。タスクをそのまま coder に渡す薄いラッパー。レビューなし。 |
 | `compound-eye` | マルチモデルレビュー: Claude と Codex に同じ指示を同時送信し、両方の回答を統合。 |
 | `review-only` | 変更を加えない読み取り専用のコードレビューピース。 |
+| `structural-reform` | プロジェクト全体の構造改革: 段階的なファイル分割を伴う反復的なコードベース再構成。 |
+| `unit-test` | ユニットテスト重視ピース: テスト分析 → テスト実装 → レビュー → 修正。 |
 
-**Hybrid Codex バリアント** (`*-hybrid-codex`): 主要ピースごとに、coder エージェントを Codex で実行しレビュアーは Claude を使うハイブリッド構成が用意されています。対象: default, minimal, expert, expert-cqrs, passthrough, review-fix-minimal, coding。
+**ペルソナ別プロバイダー設定:** 設定ファイルの `persona_providers` で、特定のペルソナを異なるプロバイダーにルーティングできます（例: coder は Codex、レビュアーは Claude）。ピースを複製する必要はありません。
 
 `takt switch` でピースを切り替えられます。
 
@@ -475,6 +489,7 @@ TAKTには複数のビルトインピースが同梱されています:
 | **research-planner** | リサーチタスクの計画・スコープ定義 |
 | **research-digger** | 深掘り調査と情報収集 |
 | **research-supervisor** | リサーチ品質の検証と網羅性の評価 |
+| **test-planner** | テスト戦略分析と包括的なテスト計画 |
 | **pr-commenter** | レビュー結果を GitHub PR にコメントとして投稿 |
 
 ## カスタムペルソナ
@@ -543,8 +558,15 @@ branch_name_strategy: romaji  # ブランチ名生成: 'romaji'（高速）ま
 prevent_sleep: false     # macOS の実行中スリープ防止（caffeinate）
 notification_sound: true # 通知音の有効/無効
 concurrency: 1           # takt run の並列タスク数（1-10、デフォルト: 1 = 逐次実行）
+task_poll_interval_ms: 500  # takt run 中の新タスク検出ポーリング間隔（100-5000、デフォルト: 500）
 interactive_preview_movements: 3  # 対話モードでのムーブメントプレビュー数（0-10、デフォルト: 3）
 
+# ペルソナ別プロバイダー設定（オプション）
+# ピースを複製せずに特定のペルソナを異なるプロバイダーにルーティング
+# persona_providers:
+#   coder: codex             # coder を Codex で実行
+#   ai-antipattern-reviewer: claude  # レビュアーは Claude のまま
+
 # API Key 設定（オプション）
 # 環境変数 TAKT_ANTHROPIC_API_KEY / TAKT_OPENAI_API_KEY で上書き可能
 anthropic_api_key: sk-ant-...  # Claude (Anthropic) を使う場合
diff --git a/package-lock.json b/package-lock.json
index f0ab485..4e2d30a 100644
--- a/package-lock.json
+++ b/package-lock.json
@@ -1,12 +1,12 @@
 {
   "name": "takt",
-  "version": "0.9.0",
+  "version": "0.10.0",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "takt",
-      "version": "0.9.0",
+      "version": "0.10.0",
       "license": "MIT",
       "dependencies": {
         "@anthropic-ai/claude-agent-sdk": "^0.2.37",
diff --git a/package.json b/package.json
index 46bd391..5c03c7b 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "takt",
-  "version": "0.9.0",
+  "version": "0.10.0",
   "description": "TAKT: Task Agent Koordination Tool - AI Agent Piece Orchestration",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",