Merge pull request #182 from nrslib/release/v0.10.0

Release v0.10.0

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

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)

builtins/en/config.yaml

@@ -1,34 +1,85 @@
 # TAKT Global Configuration
-# This file contains default settings for takt.
+# Location: ~/.takt/config.yaml
 
-# Language setting (en or ja)
+# ── Basic ──
+
+# Language (en | ja)
 language: en
 
-# Default piece to use when no piece is specified
+# Default piece when no piece is specified
 default_piece: default
 
-# Log level: debug, info, warn, error
+# Log level (debug | info | warn | error)
 log_level: info
 
-# Provider runtime: claude or codex
+# ── Provider & Model ──
+
+# Provider runtime (claude | codex)
 provider: claude
 
-# Builtin pieces (resources/global/{lang}/pieces)
-# enable_builtin_pieces: true
-
 # Default model (optional)
-# Claude: opus, sonnet, haiku, opusplan, default, or full model name
+# Claude: opus, sonnet, haiku
 # Codex: gpt-5.2-codex, gpt-5.1-codex, etc.
 # model: sonnet
 
-# Anthropic API key (optional, overridden by TAKT_ANTHROPIC_API_KEY env var)
-# anthropic_api_key: ""
+# Per-persona provider override (optional)
+# Override provider for specific personas. Others use the global provider.
+# persona_providers:
+#   coder: codex
 
-# OpenAI API key (optional, overridden by TAKT_OPENAI_API_KEY env var)
+# ── API Keys ──
+# Optional. Environment variables take priority:
+#   TAKT_ANTHROPIC_API_KEY, TAKT_OPENAI_API_KEY
+
+# anthropic_api_key: ""
 # openai_api_key: ""
 
-# Pipeline execution settings (optional)
-# Customize branch naming, commit messages, and PR body for pipeline mode (--task).
+# ── Execution ──
+
+# Worktree (shared clone) directory (default: ../{clone-name} relative to project)
+# worktree_dir: ~/takt-worktrees
+
+# Auto-create PR after worktree execution (default: prompt in interactive mode)
+# auto_pr: false
+
+# Prevent macOS idle sleep during execution using caffeinate (default: false)
+# prevent_sleep: false
+
+# ── Parallel Execution (takt run) ──
+
+# Number of tasks to run concurrently (1 = sequential, max: 10)
+# concurrency: 1
+
+# Polling interval in ms for picking up new tasks (100-5000, default: 500)
+# task_poll_interval_ms: 500
+
+# ── Interactive Mode ──
+
+# Number of movement previews shown in interactive mode (0 to disable, max: 10)
+# interactive_preview_movements: 3
+
+# Branch name generation strategy (romaji: fast default | ai: slow)
+# branch_name_strategy: romaji
+
+# ── Output ──
+
+# Notification sounds (default: true)
+# notification_sound: true
+
+# Minimal output for CI - suppress AI output (default: false)
+# minimal_output: false
+
+# ── Builtin Pieces ──
+
+# Enable builtin pieces (default: true)
+# enable_builtin_pieces: true
+
+# Exclude specific builtins from loading
+# disabled_builtins:
+#   - magi
+
+# ── Pipeline Mode (--pipeline) ──
+
 # pipeline:
 #   default_branch_prefix: "takt/"
 #   commit_message_template: "feat: {title} (#{issue})"
@@ -37,10 +88,14 @@ provider: claude
 #     {issue_body}
 #     Closes #{issue}
 
-# Notification sounds (true: enabled, false: disabled, default: true)
-# notification_sound: true
+# ── Preferences ──
+
+# Custom paths for preference files
+# bookmarks_file: ~/.takt/preferences/bookmarks.yaml
+# piece_categories_file: ~/.takt/preferences/piece-categories.yaml
+
+# ── Debug ──
 
-# Debug settings (optional)
 # debug:
 #   enabled: false
 #   log_file: ~/.takt/logs/debug.log

builtins/en/instructions/implement-test.md

@@ -0,0 +1,52 @@
+Implement unit tests according to the test plan.
+Refer only to files within the Report Directory shown in the Piece Context. Do not search or reference other report directories.
+
+**Important: Do NOT modify production code. Only test files may be edited.**
+
+**Actions:**
+1. Review the test plan report
+2. Implement the planned test cases
+3. Run tests and verify all pass
+4. Confirm existing tests are not broken
+
+**Test implementation constraints:**
+- Follow the project's existing test patterns (naming conventions, directory structure, helpers)
+- Write tests in Given-When-Then structure
+- One concept per test. Do not mix multiple concerns in a single test
+
+**Scope output contract (create at the start of implementation):**
+```markdown
+# Change Scope Declaration
+
+## Task
+{One-line task summary}
+
+## Planned changes
+| Type | File |
+|------|------|
+| Create | `src/__tests__/example.test.ts` |
+
+## Estimated size
+Small / Medium / Large
+
+## Impact area
+- {Affected modules or features}
+```
+
+**Decisions output contract (at implementation completion, only if decisions were made):**
+```markdown
+# Decision Log
+
+## 1. {Decision}
+- **Context**: {Why the decision was needed}
+- **Options considered**: {List of options}
+- **Rationale**: {Reason for the choice}
+```
+
+**Required output (include headings)**
+## Work results
+- {Summary of actions taken}
+## Changes made
+- {Summary of changes}
+## Test results
+- {Command executed and results}

builtins/en/instructions/plan-test.md

@@ -0,0 +1,11 @@
+Analyze the target code and identify missing unit tests.
+
+**Note:** If a Previous Response exists, this is a replan due to rejection.
+Revise the test plan taking that feedback into account.
+
+**Actions:**
+1. Read the target module source code and understand its behavior, branches, and state transitions
+2. Read existing tests and identify what is already covered
+3. Identify missing test cases (happy path, error cases, boundary values, edge cases)
+4. Determine test strategy (mock approach, existing test helper usage, fixture design)
+5. Provide concrete guidelines for the test implementer

builtins/en/instructions/review-arch.md

@@ -9,8 +9,14 @@ Do not review AI-specific issues (already covered by the ai_review movement).
 - Dead code
 - Call chain verification
 
+**Previous finding tracking (required):**
+- First, extract open findings from "Previous Response"
+- Assign `finding_id` to each finding and classify current status as `new / persists / resolved`
+- If status is `persists`, provide concrete unresolved evidence (file/line)
+
 ## Judgment Procedure
 
-1. Review the change diff and detect issues based on the architecture and design criteria above
-2. For each detected issue, classify as blocking/non-blocking based on Policy's scope determination table and judgment rules
-3. If there is even one blocking issue, judge as REJECT
+1. First, extract previous open findings and preliminarily classify as `new / persists / resolved`
+2. Review the change diff and detect issues based on the architecture and design criteria above
+3. For each detected issue, classify as blocking/non-blocking based on Policy's scope determination table and judgment rules
+4. If there is even one blocking issue (`new` or `persists`), judge as REJECT

builtins/en/instructions/review-qa.md

@@ -7,8 +7,14 @@ Review the changes from a quality assurance perspective.
 - Logging and monitoring
 - Maintainability
 
+**Previous finding tracking (required):**
+- First, extract open findings from "Previous Response"
+- Assign `finding_id` to each finding and classify current status as `new / persists / resolved`
+- If status is `persists`, provide concrete unresolved evidence (file/line)
+
 ## Judgment Procedure
 
-1. Review the change diff and detect issues based on the quality assurance criteria above
-2. For each detected issue, classify as blocking/non-blocking based on Policy's scope determination table and judgment rules
-3. If there is even one blocking issue, judge as REJECT
+1. First, extract previous open findings and preliminarily classify as `new / persists / resolved`
+2. Review the change diff and detect issues based on the quality assurance criteria above
+3. For each detected issue, classify as blocking/non-blocking based on Policy's scope determination table and judgment rules
+4. If there is even one blocking issue (`new` or `persists`), judge as REJECT

builtins/en/instructions/review-test.md

@@ -0,0 +1,14 @@
+Review the changes from a test quality perspective.
+
+**Review criteria:**
+- Whether all test plan items are covered
+- Test quality (Given-When-Then structure, independence, reproducibility)
+- Test naming conventions
+- Completeness (unnecessary tests, missing cases)
+- Appropriateness of mocks and fixtures
+
+## Judgment Procedure
+
+1. Cross-reference the test plan report ({report:00-test-plan.md}) with the implemented tests
+2. For each detected issue, classify as blocking/non-blocking based on Policy's scope determination table and judgment rules
+3. If there is even one blocking issue, judge as REJECT

builtins/en/output-contracts/architecture-review.md

@@ -14,10 +14,15 @@
 - [x] Dead code
 - [x] Call chain verification
 
+## Previous Open Findings
+| finding_id | Previous Status | Current Status (new/persists/resolved) | Evidence |
+|------------|-----------------|-----------------------------------------|----------|
+| ARCH-EXAMPLE-src-file-L42 | open | persists | `src/file.ts:42` |
+
 ## Issues (if REJECT)
-| # | Scope | Location | Issue | Fix Suggestion |
-|---|-------|----------|-------|----------------|
-| 1 | In-scope | `src/file.ts:42` | Issue description | Fix approach |
+| # | finding_id | Status (new/persists) | Scope | Location | Issue | Fix Suggestion |
+|---|------------|-----------------------|-------|----------|-------|----------------|
+| 1 | ARCH-EXAMPLE-src-file-L42 | new | In-scope | `src/file.ts:42` | Issue description | Fix approach |
 
 Scope: "In-scope" (fixable in this change) / "Out-of-scope" (existing issue, non-blocking)
 

builtins/en/output-contracts/qa-review.md

@@ -15,8 +15,13 @@
 | Documentation | ✅ | - |
 | Maintainability | ✅ | - |
 
+## Previous Open Findings
+| finding_id | Previous Status | Current Status (new/persists/resolved) | Evidence |
+|------------|-----------------|-----------------------------------------|----------|
+| QA-EXAMPLE-src-file-L42 | open | persists | `src/file.ts:42` |
+
 ## Issues (if REJECT)
-| # | Category | Issue | Fix Suggestion |
-|---|----------|-------|----------------|
-| 1 | Testing | Issue description | Fix approach |
+| # | finding_id | Status (new/persists) | Category | Issue | Fix Suggestion |
+|---|------------|-----------------------|----------|-------|----------------|
+| 1 | QA-EXAMPLE-src-file-L42 | new | Testing | Issue description | Fix approach |
 ```

builtins/en/output-contracts/test-plan.md

@@ -0,0 +1,24 @@
+```markdown
+# Test Plan
+
+## Target Modules
+{List of modules to analyze}
+
+## Existing Test Analysis
+| Module | Existing Tests | Coverage Status |
+|--------|---------------|-----------------|
+| `src/xxx.ts` | `xxx.test.ts` | {Coverage status} |
+
+## Missing Test Cases
+| # | Target | Test Case | Priority | Reason |
+|---|--------|-----------|----------|--------|
+| 1 | `src/xxx.ts` | {Test case summary} | High/Medium/Low | {Reason} |
+
+## Test Strategy
+- {Mock approach}
+- {Fixture design}
+- {Test helper usage}
+
+## Implementation Guidelines
+- {Concrete instructions for the test implementer}
+```

builtins/en/personas/test-planner.md

@@ -0,0 +1,25 @@
+# Test Planner
+
+You are a **test analysis and planning specialist**. You understand the behavior of target code, analyze existing test coverage, and systematically identify missing test cases.
+
+## Role Boundaries
+
+**Do:**
+- Analyze target code behavior, branches, and state transitions
+- Analyze existing test coverage
+- Identify missing test cases (happy path, error cases, boundary values, edge cases)
+- Determine test strategy (mock approach, fixture design, test helper usage)
+- Provide concrete guidelines for test implementers
+
+**Don't:**
+- Plan production code changes (Planner's job)
+- Implement test code (Coder's job)
+- Review code (Reviewer's job)
+
+## Behavioral Principles
+
+- Read the code before planning. Don't list test cases based on guesses
+- Always check existing tests. Don't duplicate already-covered scenarios
+- Prioritize tests: business logic and state transitions > edge cases > simple CRUD
+- Provide instructions at a granularity that prevents test implementers from hesitating
+- Follow the project's existing test patterns. Don't propose novel conventions

builtins/en/piece-categories.yaml

@@ -10,6 +10,7 @@ piece_categories:
     pieces:
       - review-fix-minimal
       - review-only
+      - unit-test
   🎨 Frontend: {}
   ⚙️ Backend: {}
   🔧 Expert:
@@ -17,20 +18,9 @@ piece_categories:
       pieces:
         - expert
         - expert-cqrs
-  🔀 Hybrid (Codex Coding):
-    🚀 Quick Start:
+    Refactoring:
       pieces:
-        - coding-hybrid-codex
-        - default-hybrid-codex
-        - minimal-hybrid-codex
-        - passthrough-hybrid-codex
-    🔧 Expert:
-      pieces:
-        - expert-cqrs-hybrid-codex
-        - expert-hybrid-codex
-    🔍 Review & Fix:
-      pieces:
-        - review-fix-minimal-hybrid-codex
+        - structural-reform
   Others:
     pieces:
       - research

builtins/en/pieces/coding-hybrid-codex.yaml

@@ -1,158 +0,0 @@
-# Auto-generated from coding.yaml by tools/generate-hybrid-codex.mjs
-# Do not edit manually. Edit the source piece and re-run the generator.
-
-name: coding-hybrid-codex
-description: Lightweight development piece with planning and parallel reviews (plan -> implement -> parallel review -> complete)
-max_iterations: 20
-knowledge:
-  architecture: ../knowledge/architecture.md
-personas:
-  planner: ../personas/planner.md
-  coder: ../personas/coder.md
-  ai-antipattern-reviewer: ../personas/ai-antipattern-reviewer.md
-  architecture-reviewer: ../personas/architecture-reviewer.md
-instructions:
-  plan: ../instructions/plan.md
-  implement: ../instructions/implement.md
-  ai-review: ../instructions/ai-review.md
-  review-arch: ../instructions/review-arch.md
-  fix: ../instructions/fix.md
-report_formats:
-  plan: ../output-contracts/plan.md
-  ai-review: ../output-contracts/ai-review.md
-  architecture-review: ../output-contracts/architecture-review.md
-initial_movement: plan
-movements:
-  - name: plan
-    edit: false
-    persona: planner
-    knowledge: architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Bash
-      - WebSearch
-      - WebFetch
-    rules:
-      - condition: Requirements are clear and implementable
-        next: implement
-      - condition: User is asking a question (not an implementation task)
-        next: COMPLETE
-      - condition: Requirements are unclear, insufficient information
-        next: ABORT
-    instruction: plan
-    output_contracts:
-      report:
-        - name: 00-plan.md
-          format: plan
-  - name: implement
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    session: refresh
-    knowledge: architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: Implementation complete
-        next: reviewers
-      - condition: Implementation not started (report only)
-        next: reviewers
-      - condition: Cannot determine, insufficient information
-        next: reviewers
-      - condition: User input required
-        next: implement
-        requires_user_input: true
-        interactive_only: true
-    instruction: implement
-    output_contracts:
-      report:
-        - Scope: 02-coder-scope.md
-        - Decisions: 03-coder-decisions.md
-  - name: reviewers
-    parallel:
-      - name: ai_review
-        edit: false
-        persona: ai-antipattern-reviewer
-        policy:
-          - review
-          - ai-antipattern
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: No AI-specific issues
-          - condition: AI-specific issues found
-        instruction: ai-review
-        output_contracts:
-          report:
-            - name: 04-ai-review.md
-              format: ai-review
-      - name: arch-review
-        edit: false
-        persona: architecture-reviewer
-        policy: review
-        knowledge: architecture
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-arch
-        output_contracts:
-          report:
-            - name: 05-architect-review.md
-              format: architecture-review
-    rules:
-      - condition: all("No AI-specific issues", "approved")
-        next: COMPLETE
-      - condition: any("AI-specific issues found", "needs_fix")
-        next: fix
-  - name: fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    knowledge: architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: Fix complete
-        next: reviewers
-      - condition: Cannot determine, insufficient information
-        next: ABORT
-    instruction: fix
-policies:
-  coding: ../policies/coding.md
-  review: ../policies/review.md
-  testing: ../policies/testing.md
-  ai-antipattern: ../policies/ai-antipattern.md

builtins/en/pieces/expert-cqrs-hybrid-codex.yaml

@@ -1,341 +0,0 @@
-# Auto-generated from expert-cqrs.yaml by tools/generate-hybrid-codex.mjs
-# Do not edit manually. Edit the source piece and re-run the generator.
-
-name: expert-cqrs-hybrid-codex
-description: CQRS+ES, Frontend, Security, QA Expert Review
-max_iterations: 30
-knowledge:
-  frontend: ../knowledge/frontend.md
-  backend: ../knowledge/backend.md
-  cqrs-es: ../knowledge/cqrs-es.md
-  security: ../knowledge/security.md
-  architecture: ../knowledge/architecture.md
-personas:
-  planner: ../personas/planner.md
-  coder: ../personas/coder.md
-  ai-antipattern-reviewer: ../personas/ai-antipattern-reviewer.md
-  architecture-reviewer: ../personas/architecture-reviewer.md
-  cqrs-es-reviewer: ../personas/cqrs-es-reviewer.md
-  frontend-reviewer: ../personas/frontend-reviewer.md
-  security-reviewer: ../personas/security-reviewer.md
-  qa-reviewer: ../personas/qa-reviewer.md
-  expert-supervisor: ../personas/expert-supervisor.md
-instructions:
-  plan: ../instructions/plan.md
-  implement: ../instructions/implement.md
-  ai-review: ../instructions/ai-review.md
-  ai-fix: ../instructions/ai-fix.md
-  arbitrate: ../instructions/arbitrate.md
-  review-cqrs-es: ../instructions/review-cqrs-es.md
-  review-frontend: ../instructions/review-frontend.md
-  review-security: ../instructions/review-security.md
-  review-qa: ../instructions/review-qa.md
-  fix: ../instructions/fix.md
-  supervise: ../instructions/supervise.md
-  fix-supervisor: ../instructions/fix-supervisor.md
-report_formats:
-  plan: ../output-contracts/plan.md
-  ai-review: ../output-contracts/ai-review.md
-  cqrs-es-review: ../output-contracts/cqrs-es-review.md
-  frontend-review: ../output-contracts/frontend-review.md
-  security-review: ../output-contracts/security-review.md
-  qa-review: ../output-contracts/qa-review.md
-  validation: ../output-contracts/validation.md
-  summary: ../output-contracts/summary.md
-initial_movement: plan
-movements:
-  - name: plan
-    edit: false
-    persona: planner
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Bash
-      - WebSearch
-      - WebFetch
-    instruction: plan
-    rules:
-      - condition: Task analysis and planning is complete
-        next: implement
-      - condition: Requirements are unclear and planning cannot proceed
-        next: ABORT
-    output_contracts:
-      report:
-        - name: 00-plan.md
-          format: plan
-  - name: implement
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    session: refresh
-    knowledge:
-      - frontend
-      - backend
-      - cqrs-es
-      - security
-      - architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    instruction: implement
-    rules:
-      - condition: Implementation is complete
-        next: ai_review
-      - condition: No implementation (report only)
-        next: ai_review
-      - condition: Cannot proceed with implementation
-        next: ai_review
-      - condition: User input required
-        next: implement
-        requires_user_input: true
-        interactive_only: true
-    output_contracts:
-      report:
-        - Scope: 01-coder-scope.md
-        - Decisions: 02-coder-decisions.md
-  - name: ai_review
-    edit: false
-    persona: ai-antipattern-reviewer
-    policy:
-      - review
-      - ai-antipattern
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - WebSearch
-      - WebFetch
-    instruction: ai-review
-    rules:
-      - condition: No AI-specific issues found
-        next: reviewers
-      - condition: AI-specific issues detected
-        next: ai_fix
-    output_contracts:
-      report:
-        - name: 03-ai-review.md
-          format: ai-review
-  - name: ai_fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    session: refresh
-    knowledge:
-      - frontend
-      - backend
-      - cqrs-es
-      - security
-      - architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    instruction: ai-fix
-    rules:
-      - condition: AI Reviewer's issues have been fixed
-        next: ai_review
-      - condition: No fix needed (verified target files/spec)
-        next: ai_no_fix
-      - condition: Unable to proceed with fixes
-        next: ai_no_fix
-  - name: ai_no_fix
-    edit: false
-    persona: architecture-reviewer
-    policy: review
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-    rules:
-      - condition: ai_review's findings are valid (fix required)
-        next: ai_fix
-      - condition: ai_fix's judgment is valid (no fix needed)
-        next: reviewers
-    instruction: arbitrate
-  - name: reviewers
-    parallel:
-      - name: cqrs-es-review
-        edit: false
-        persona: cqrs-es-reviewer
-        policy: review
-        knowledge:
-          - cqrs-es
-          - backend
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-cqrs-es
-        output_contracts:
-          report:
-            - name: 04-cqrs-es-review.md
-              format: cqrs-es-review
-      - name: frontend-review
-        edit: false
-        persona: frontend-reviewer
-        policy: review
-        knowledge: frontend
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-frontend
-        output_contracts:
-          report:
-            - name: 05-frontend-review.md
-              format: frontend-review
-      - name: security-review
-        edit: false
-        persona: security-reviewer
-        policy: review
-        knowledge: security
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-security
-        output_contracts:
-          report:
-            - name: 06-security-review.md
-              format: security-review
-      - name: qa-review
-        edit: false
-        persona: qa-reviewer
-        policy:
-          - review
-          - qa
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-qa
-        output_contracts:
-          report:
-            - name: 07-qa-review.md
-              format: qa-review
-    rules:
-      - condition: all("approved")
-        next: supervise
-      - condition: any("needs_fix")
-        next: fix
-  - name: fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    knowledge:
-      - frontend
-      - backend
-      - cqrs-es
-      - security
-      - architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: Fix complete
-        next: reviewers
-      - condition: Cannot proceed, insufficient info
-        next: plan
-    instruction: fix
-  - name: supervise
-    edit: false
-    persona: expert-supervisor
-    policy: review
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - WebSearch
-      - WebFetch
-    instruction: supervise
-    rules:
-      - condition: All validations pass and ready to merge
-        next: COMPLETE
-      - condition: Issues detected during final review
-        next: fix_supervisor
-    output_contracts:
-      report:
-        - Validation: 08-supervisor-validation.md
-        - Summary: summary.md
-  - name: fix_supervisor
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    knowledge:
-      - frontend
-      - backend
-      - cqrs-es
-      - security
-      - architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    instruction: fix-supervisor
-    rules:
-      - condition: Supervisor's issues have been fixed
-        next: supervise
-      - condition: Unable to proceed with fixes
-        next: plan
-policies:
-  coding: ../policies/coding.md
-  review: ../policies/review.md
-  testing: ../policies/testing.md
-  ai-antipattern: ../policies/ai-antipattern.md
-  qa: ../policies/qa.md

builtins/en/pieces/expert-hybrid-codex.yaml

@@ -1,335 +0,0 @@
-# Auto-generated from expert.yaml by tools/generate-hybrid-codex.mjs
-# Do not edit manually. Edit the source piece and re-run the generator.
-
-name: expert-hybrid-codex
-description: Architecture, Frontend, Security, QA Expert Review
-max_iterations: 30
-knowledge:
-  frontend: ../knowledge/frontend.md
-  backend: ../knowledge/backend.md
-  security: ../knowledge/security.md
-  architecture: ../knowledge/architecture.md
-personas:
-  planner: ../personas/planner.md
-  coder: ../personas/coder.md
-  ai-antipattern-reviewer: ../personas/ai-antipattern-reviewer.md
-  architecture-reviewer: ../personas/architecture-reviewer.md
-  frontend-reviewer: ../personas/frontend-reviewer.md
-  security-reviewer: ../personas/security-reviewer.md
-  qa-reviewer: ../personas/qa-reviewer.md
-  expert-supervisor: ../personas/expert-supervisor.md
-instructions:
-  plan: ../instructions/plan.md
-  implement: ../instructions/implement.md
-  ai-review: ../instructions/ai-review.md
-  ai-fix: ../instructions/ai-fix.md
-  arbitrate: ../instructions/arbitrate.md
-  review-arch: ../instructions/review-arch.md
-  review-frontend: ../instructions/review-frontend.md
-  review-security: ../instructions/review-security.md
-  review-qa: ../instructions/review-qa.md
-  fix: ../instructions/fix.md
-  supervise: ../instructions/supervise.md
-  fix-supervisor: ../instructions/fix-supervisor.md
-report_formats:
-  plan: ../output-contracts/plan.md
-  ai-review: ../output-contracts/ai-review.md
-  architecture-review: ../output-contracts/architecture-review.md
-  frontend-review: ../output-contracts/frontend-review.md
-  security-review: ../output-contracts/security-review.md
-  qa-review: ../output-contracts/qa-review.md
-  validation: ../output-contracts/validation.md
-  summary: ../output-contracts/summary.md
-initial_movement: plan
-movements:
-  - name: plan
-    edit: false
-    persona: planner
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Bash
-      - WebSearch
-      - WebFetch
-    instruction: plan
-    rules:
-      - condition: Task analysis and planning is complete
-        next: implement
-      - condition: Requirements are unclear and planning cannot proceed
-        next: ABORT
-    output_contracts:
-      report:
-        - name: 00-plan.md
-          format: plan
-  - name: implement
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    session: refresh
-    knowledge:
-      - frontend
-      - backend
-      - security
-      - architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    instruction: implement
-    rules:
-      - condition: Implementation is complete
-        next: ai_review
-      - condition: No implementation (report only)
-        next: ai_review
-      - condition: Cannot proceed with implementation
-        next: ai_review
-      - condition: User input required
-        next: implement
-        requires_user_input: true
-        interactive_only: true
-    output_contracts:
-      report:
-        - Scope: 01-coder-scope.md
-        - Decisions: 02-coder-decisions.md
-  - name: ai_review
-    edit: false
-    persona: ai-antipattern-reviewer
-    policy:
-      - review
-      - ai-antipattern
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - WebSearch
-      - WebFetch
-    instruction: ai-review
-    rules:
-      - condition: No AI-specific issues found
-        next: reviewers
-      - condition: AI-specific issues detected
-        next: ai_fix
-    output_contracts:
-      report:
-        - name: 03-ai-review.md
-          format: ai-review
-  - name: ai_fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    session: refresh
-    knowledge:
-      - frontend
-      - backend
-      - security
-      - architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    instruction: ai-fix
-    rules:
-      - condition: AI Reviewer's issues have been fixed
-        next: ai_review
-      - condition: No fix needed (verified target files/spec)
-        next: ai_no_fix
-      - condition: Unable to proceed with fixes
-        next: ai_no_fix
-  - name: ai_no_fix
-    edit: false
-    persona: architecture-reviewer
-    policy: review
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-    rules:
-      - condition: ai_review's findings are valid (fix required)
-        next: ai_fix
-      - condition: ai_fix's judgment is valid (no fix needed)
-        next: reviewers
-    instruction: arbitrate
-  - name: reviewers
-    parallel:
-      - name: arch-review
-        edit: false
-        persona: architecture-reviewer
-        policy: review
-        knowledge:
-          - architecture
-          - backend
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-arch
-        output_contracts:
-          report:
-            - name: 04-architect-review.md
-              format: architecture-review
-      - name: frontend-review
-        edit: false
-        persona: frontend-reviewer
-        policy: review
-        knowledge: frontend
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-frontend
-        output_contracts:
-          report:
-            - name: 05-frontend-review.md
-              format: frontend-review
-      - name: security-review
-        edit: false
-        persona: security-reviewer
-        policy: review
-        knowledge: security
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-security
-        output_contracts:
-          report:
-            - name: 06-security-review.md
-              format: security-review
-      - name: qa-review
-        edit: false
-        persona: qa-reviewer
-        policy:
-          - review
-          - qa
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-qa
-        output_contracts:
-          report:
-            - name: 07-qa-review.md
-              format: qa-review
-    rules:
-      - condition: all("approved")
-        next: supervise
-      - condition: any("needs_fix")
-        next: fix
-  - name: fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    knowledge:
-      - frontend
-      - backend
-      - security
-      - architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: Fix complete
-        next: reviewers
-      - condition: Cannot proceed, insufficient info
-        next: plan
-    instruction: fix
-  - name: supervise
-    edit: false
-    persona: expert-supervisor
-    policy: review
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - WebSearch
-      - WebFetch
-    instruction: supervise
-    rules:
-      - condition: All validations pass and ready to merge
-        next: COMPLETE
-      - condition: Issues detected during final review
-        next: fix_supervisor
-    output_contracts:
-      report:
-        - Validation: 08-supervisor-validation.md
-        - Summary: summary.md
-  - name: fix_supervisor
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    knowledge:
-      - frontend
-      - backend
-      - security
-      - architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    instruction: fix-supervisor
-    rules:
-      - condition: Supervisor's issues have been fixed
-        next: supervise
-      - condition: Unable to proceed with fixes
-        next: plan
-policies:
-  coding: ../policies/coding.md
-  review: ../policies/review.md
-  testing: ../policies/testing.md
-  ai-antipattern: ../policies/ai-antipattern.md
-  qa: ../policies/qa.md

builtins/en/pieces/minimal-hybrid-codex.yaml

@@ -1,202 +0,0 @@
-# Auto-generated from minimal.yaml by tools/generate-hybrid-codex.mjs
-# Do not edit manually. Edit the source piece and re-run the generator.
-
-name: minimal-hybrid-codex
-description: Minimal development piece (implement -> parallel review -> fix if needed -> complete)
-max_iterations: 20
-personas:
-  coder: ../personas/coder.md
-  ai-antipattern-reviewer: ../personas/ai-antipattern-reviewer.md
-  supervisor: ../personas/supervisor.md
-instructions:
-  implement: ../instructions/implement.md
-  review-ai: ../instructions/review-ai.md
-  ai-fix: ../instructions/ai-fix.md
-  supervise: ../instructions/supervise.md
-  fix-supervisor: ../instructions/fix-supervisor.md
-report_formats:
-  ai-review: ../output-contracts/ai-review.md
-initial_movement: implement
-movements:
-  - name: implement
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    instruction: implement
-    rules:
-      - condition: Implementation complete
-        next: reviewers
-      - condition: Cannot proceed, insufficient info
-        next: ABORT
-      - condition: User input required because there are items to confirm with the user
-        next: implement
-        requires_user_input: true
-        interactive_only: true
-    output_contracts:
-      report:
-        - Scope: 01-coder-scope.md
-        - Decisions: 02-coder-decisions.md
-  - name: reviewers
-    parallel:
-      - name: ai_review
-        edit: false
-        persona: ai-antipattern-reviewer
-        policy:
-          - review
-          - ai-antipattern
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        instruction: review-ai
-        rules:
-          - condition: No AI-specific issues
-          - condition: AI-specific issues found
-        output_contracts:
-          report:
-            - name: 03-ai-review.md
-              format: ai-review
-      - name: supervise
-        edit: false
-        persona: supervisor
-        policy: review
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - Bash
-          - WebSearch
-          - WebFetch
-        instruction: supervise
-        rules:
-          - condition: All checks passed
-          - condition: Requirements unmet, tests failing
-        output_contracts:
-          report:
-            - Validation: 05-supervisor-validation.md
-            - Summary: summary.md
-    rules:
-      - condition: all("No AI-specific issues", "All checks passed")
-        next: COMPLETE
-      - condition: all("AI-specific issues found", "Requirements unmet, tests failing")
-        next: fix_both
-      - condition: any("AI-specific issues found")
-        next: ai_fix
-      - condition: any("Requirements unmet, tests failing")
-        next: supervise_fix
-  - name: fix_both
-    parallel:
-      - name: ai_fix_parallel
-        edit: true
-        persona: coder
-        provider: codex
-        policy:
-          - coding
-          - testing
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - Edit
-          - Bash
-          - WebSearch
-          - WebFetch
-        permission_mode: edit
-        rules:
-          - condition: AI Reviewer's issues fixed
-          - condition: No fix needed (verified target files/spec)
-          - condition: Cannot proceed, insufficient info
-        instruction: ai-fix
-      - name: supervise_fix_parallel
-        edit: true
-        persona: coder
-        provider: codex
-        policy:
-          - coding
-          - testing
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - Edit
-          - Bash
-          - WebSearch
-          - WebFetch
-        permission_mode: edit
-        rules:
-          - condition: Supervisor's issues fixed
-          - condition: Cannot proceed, insufficient info
-        instruction: fix-supervisor
-    rules:
-      - condition: all("AI Reviewer's issues fixed", "Supervisor's issues fixed")
-        next: reviewers
-      - condition: any("No fix needed (verified target files/spec)", "Cannot proceed, insufficient info")
-        next: implement
-  - name: ai_fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: AI Reviewer's issues fixed
-        next: reviewers
-      - condition: No fix needed (verified target files/spec)
-        next: implement
-      - condition: Cannot proceed, insufficient info
-        next: implement
-    instruction: ai-fix
-  - name: supervise_fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: Supervisor's issues fixed
-        next: reviewers
-      - condition: Cannot proceed, insufficient info
-        next: implement
-    instruction: fix-supervisor
-policies:
-  coding: ../policies/coding.md
-  review: ../policies/review.md
-  testing: ../policies/testing.md
-  ai-antipattern: ../policies/ai-antipattern.md

builtins/en/pieces/passthrough-hybrid-codex.yaml

@@ -1,44 +0,0 @@
-# Auto-generated from passthrough.yaml by tools/generate-hybrid-codex.mjs
-# Do not edit manually. Edit the source piece and re-run the generator.
-
-name: passthrough-hybrid-codex
-description: Single-agent thin wrapper. Pass task directly to coder as-is.
-max_iterations: 10
-personas:
-  coder: ../personas/coder.md
-initial_movement: execute
-movements:
-  - name: execute
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: Task complete
-        next: COMPLETE
-      - condition: Cannot proceed
-        next: ABORT
-      - condition: User input required
-        next: execute
-        requires_user_input: true
-        interactive_only: true
-    instruction_template: |
-      Do the task.
-    output_contracts:
-      report:
-        - Summary: summary.md
-policies:
-  coding: ../policies/coding.md
-  testing: ../policies/testing.md

builtins/en/pieces/review-fix-minimal-hybrid-codex.yaml

@@ -1,202 +0,0 @@
-# Auto-generated from review-fix-minimal.yaml by tools/generate-hybrid-codex.mjs
-# Do not edit manually. Edit the source piece and re-run the generator.
-
-name: review-fix-minimal-hybrid-codex
-description: Review and fix piece for existing code (starts with review, no implementation)
-max_iterations: 20
-personas:
-  coder: ../personas/coder.md
-  ai-antipattern-reviewer: ../personas/ai-antipattern-reviewer.md
-  supervisor: ../personas/supervisor.md
-instructions:
-  implement: ../instructions/implement.md
-  review-ai: ../instructions/review-ai.md
-  ai-fix: ../instructions/ai-fix.md
-  supervise: ../instructions/supervise.md
-  fix-supervisor: ../instructions/fix-supervisor.md
-report_formats:
-  ai-review: ../output-contracts/ai-review.md
-initial_movement: reviewers
-movements:
-  - name: implement
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    instruction: implement
-    rules:
-      - condition: Implementation complete
-        next: reviewers
-      - condition: Cannot proceed, insufficient info
-        next: ABORT
-      - condition: User input required because there are items to confirm with the user
-        next: implement
-        requires_user_input: true
-        interactive_only: true
-    output_contracts:
-      report:
-        - Scope: 01-coder-scope.md
-        - Decisions: 02-coder-decisions.md
-  - name: reviewers
-    parallel:
-      - name: ai_review
-        edit: false
-        persona: ai-antipattern-reviewer
-        policy:
-          - review
-          - ai-antipattern
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        instruction: review-ai
-        rules:
-          - condition: No AI-specific issues
-          - condition: AI-specific issues found
-        output_contracts:
-          report:
-            - name: 03-ai-review.md
-              format: ai-review
-      - name: supervise
-        edit: false
-        persona: supervisor
-        policy: review
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - Bash
-          - WebSearch
-          - WebFetch
-        instruction: supervise
-        rules:
-          - condition: All checks passed
-          - condition: Requirements unmet, tests failing
-        output_contracts:
-          report:
-            - Validation: 05-supervisor-validation.md
-            - Summary: summary.md
-    rules:
-      - condition: all("No AI-specific issues", "All checks passed")
-        next: COMPLETE
-      - condition: all("AI-specific issues found", "Requirements unmet, tests failing")
-        next: fix_both
-      - condition: any("AI-specific issues found")
-        next: ai_fix
-      - condition: any("Requirements unmet, tests failing")
-        next: supervise_fix
-  - name: fix_both
-    parallel:
-      - name: ai_fix_parallel
-        edit: true
-        persona: coder
-        provider: codex
-        policy:
-          - coding
-          - testing
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - Edit
-          - Bash
-          - WebSearch
-          - WebFetch
-        permission_mode: edit
-        rules:
-          - condition: AI Reviewer's issues fixed
-          - condition: No fix needed (verified target files/spec)
-          - condition: Cannot proceed, insufficient info
-        instruction: ai-fix
-      - name: supervise_fix_parallel
-        edit: true
-        persona: coder
-        provider: codex
-        policy:
-          - coding
-          - testing
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - Edit
-          - Bash
-          - WebSearch
-          - WebFetch
-        permission_mode: edit
-        rules:
-          - condition: Supervisor's issues fixed
-          - condition: Cannot proceed, insufficient info
-        instruction: fix-supervisor
-    rules:
-      - condition: all("AI Reviewer's issues fixed", "Supervisor's issues fixed")
-        next: reviewers
-      - condition: any("No fix needed (verified target files/spec)", "Cannot proceed, insufficient info")
-        next: implement
-  - name: ai_fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: AI Reviewer's issues fixed
-        next: reviewers
-      - condition: No fix needed (verified target files/spec)
-        next: implement
-      - condition: Cannot proceed, insufficient info
-        next: implement
-    instruction: ai-fix
-  - name: supervise_fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: Supervisor's issues fixed
-        next: reviewers
-      - condition: Cannot proceed, insufficient info
-        next: implement
-    instruction: fix-supervisor
-policies:
-  coding: ../policies/coding.md
-  review: ../policies/review.md
-  testing: ../policies/testing.md
-  ai-antipattern: ../policies/ai-antipattern.md

builtins/en/pieces/structural-reform.yaml

@@ -0,0 +1,455 @@
+name: structural-reform
+description: Full project review and structural reform - iterative codebase restructuring with staged file splits
+max_iterations: 50
+policies:
+  coding: ../policies/coding.md
+  review: ../policies/review.md
+  testing: ../policies/testing.md
+  qa: ../policies/qa.md
+knowledge:
+  backend: ../knowledge/backend.md
+  architecture: ../knowledge/architecture.md
+personas:
+  planner: ../personas/planner.md
+  coder: ../personas/coder.md
+  architecture-reviewer: ../personas/architecture-reviewer.md
+  qa-reviewer: ../personas/qa-reviewer.md
+  supervisor: ../personas/supervisor.md
+instructions:
+  implement: ../instructions/implement.md
+  review-arch: ../instructions/review-arch.md
+  review-qa: ../instructions/review-qa.md
+  fix: ../instructions/fix.md
+initial_movement: review
+loop_monitors:
+  - cycle:
+      - implement
+      - fix
+    threshold: 3
+    judge:
+      persona: supervisor
+      instruction_template: |
+        The implement -> reviewers -> fix loop has repeated {cycle_count} times for the current reform target.
+
+        Review the reports from each cycle and determine whether this loop
+        is making progress or repeating the same issues.
+
+        **Reports to reference:**
+        - Architect review: {report:04-architect-review.md}
+        - QA review: {report:05-qa-review.md}
+
+        **Judgment criteria:**
+        - Are review findings being addressed in each fix cycle?
+        - Are the same issues recurring without resolution?
+        - Is the implementation converging toward approval?
+      rules:
+        - condition: Healthy (making progress toward approval)
+          next: implement
+        - condition: Unproductive (same issues recurring, no convergence)
+          next: next_target
+movements:
+  - name: review
+    edit: false
+    persona: architecture-reviewer
+    policy: review
+    knowledge:
+      - architecture
+      - backend
+    allowed_tools:
+      - Read
+      - Glob
+      - Grep
+      - WebSearch
+      - WebFetch
+    instruction_template: |
+      ## Piece Status
+      - Iteration: {iteration}/{max_iterations} (piece-wide)
+      - Movement Iteration: {movement_iteration} (times this movement has run)
+      - Movement: review (full project review)
+
+      ## User Request
+      {task}
+
+      ## Instructions
+      Conduct a comprehensive structural review of the entire project codebase.
+
+      **Focus areas:**
+      1. **God Classes/Functions**: Files exceeding 300 lines, classes with multiple responsibilities
+      2. **Coupling**: Circular dependencies, tight coupling between modules
+      3. **Cohesion**: Low-cohesion modules mixing unrelated concerns
+      4. **Testability**: Untestable code due to tight coupling or side effects
+      5. **Layer violations**: Wrong dependency directions, domain logic in adapters
+      6. **DRY violations**: Duplicated logic across 3+ locations
+
+      **For each issue found, report:**
+      - File path and line count
+      - Problem category (God Class, Low Cohesion, etc.)
+      - Severity (Critical / High / Medium)
+      - Specific responsibilities that should be separated
+      - Dependencies that would be affected by splitting
+
+      **Output format:**
+
+      ```markdown
+      # Full Project Structural Review
+
+      ## Summary
+      - Total files reviewed: N
+      - Issues found: N (Critical: N, High: N, Medium: N)
+
+      ## Critical Issues
+
+      ### 1. {File path} ({line count} lines)
+      - **Problem**: {category}
+      - **Severity**: Critical
+      - **Responsibilities found**:
+        1. {responsibility 1}
+        2. {responsibility 2}
+      - **Proposed split**:
+        - `{new-file-1}.ts`: {responsibility}
+        - `{new-file-2}.ts`: {responsibility}
+      - **Affected dependents**: {files that import this module}
+
+      ## High Priority Issues
+      ...
+
+      ## Medium Priority Issues
+      ...
+
+      ## Dependency Graph Concerns
+      - {circular dependencies, layering violations}
+
+      ## Recommended Reform Order
+      1. {file} - {reason for priority}
+      2. {file} - {reason for priority}
+      ```
+    rules:
+      - condition: Full review is complete with findings
+        next: plan_reform
+      - condition: No structural issues found
+        next: COMPLETE
+    output_contracts:
+      report:
+        - name: 00-full-review.md
+
+  - name: plan_reform
+    edit: false
+    persona: planner
+    knowledge: architecture
+    allowed_tools:
+      - Read
+      - Glob
+      - Grep
+      - Bash
+      - WebSearch
+      - WebFetch
+    instruction_template: |
+      ## Piece Status
+      - Iteration: {iteration}/{max_iterations} (piece-wide)
+      - Movement Iteration: {movement_iteration} (times this movement has run)
+      - Movement: plan_reform (reform plan creation)
+
+      ## User Request
+      {task}
+
+      ## Full Review Results
+      {previous_response}
+
+      ## Additional User Inputs
+      {user_inputs}
+
+      ## Instructions
+      Based on the full review results, create a concrete reform execution plan.
+
+      **Planning principles:**
+      - One file split per iteration (keep changes manageable)
+      - Order by dependency: split leaf nodes first, then work inward
+      - Each split must leave tests and build passing
+      - No backward compatibility concerns (per user instruction)
+
+      **For each reform target, specify:**
+      1. Target file and current line count
+      2. Proposed new files with responsibilities
+      3. Expected changes to imports in dependent files
+      4. Test strategy (new tests needed, existing tests to update)
+      5. Risk assessment (what could break)
+
+      **Output format:**
+
+      ```markdown
+      # Structural Reform Plan
+
+      ## Reform Targets (ordered by execution priority)
+
+      ### Target 1: {file path}
+      - **Current state**: {line count} lines, {N} responsibilities
+      - **Proposed split**:
+        | New file | Responsibility | Estimated lines |
+        |----------|---------------|-----------------|
+        | `{path}` | {responsibility} | ~{N} |
+      - **Dependent files**: {list of files that import this}
+      - **Test plan**: {what tests to add/update}
+      - **Risk**: {Low/Medium/High} - {description}
+
+      ### Target 2: {file path}
+      ...
+
+      ## Execution Order Rationale
+      {Why this order minimizes risk and dependency conflicts}
+
+      ## Success Criteria
+      - All tests pass after each split
+      - Build succeeds after each split
+      - No file exceeds 300 lines
+      - Each file has single responsibility
+      ```
+    rules:
+      - condition: Reform plan is complete and ready to execute
+        next: implement
+      - condition: No actionable reforms identified
+        next: COMPLETE
+      - condition: Requirements unclear, need user input
+        next: ABORT
+        appendix: |
+          Clarifications needed:
+          - {Question 1}
+          - {Question 2}
+    output_contracts:
+      report:
+        - name: 01-reform-plan.md
+          format: plan
+
+  - name: implement
+    edit: true
+    persona: coder
+    policy:
+      - coding
+      - testing
+    session: refresh
+    knowledge:
+      - backend
+      - architecture
+    allowed_tools:
+      - Read
+      - Glob
+      - Grep
+      - Edit
+      - Write
+      - Bash
+      - WebSearch
+      - WebFetch
+    permission_mode: edit
+    instruction: implement
+    rules:
+      - condition: Implementation complete
+        next: reviewers
+      - condition: Cannot proceed, insufficient info
+        next: reviewers
+      - condition: User input required
+        next: implement
+        requires_user_input: true
+        interactive_only: true
+    output_contracts:
+      report:
+        - Scope: 02-coder-scope.md
+        - Decisions: 03-coder-decisions.md
+
+  - name: reviewers
+    parallel:
+      - name: arch-review
+        edit: false
+        persona: architecture-reviewer
+        policy: review
+        knowledge:
+          - architecture
+          - backend
+        allowed_tools:
+          - Read
+          - Glob
+          - Grep
+          - WebSearch
+          - WebFetch
+        rules:
+          - condition: approved
+          - condition: needs_fix
+        instruction: review-arch
+        output_contracts:
+          report:
+            - name: 04-architect-review.md
+              format: architecture-review
+      - name: qa-review
+        edit: false
+        persona: qa-reviewer
+        policy:
+          - review
+          - qa
+        allowed_tools:
+          - Read
+          - Glob
+          - Grep
+          - WebSearch
+          - WebFetch
+        rules:
+          - condition: approved
+          - condition: needs_fix
+        instruction: review-qa
+        output_contracts:
+          report:
+            - name: 05-qa-review.md
+              format: qa-review
+    rules:
+      - condition: all("approved")
+        next: verify
+      - condition: any("needs_fix")
+        next: fix
+
+  - name: fix
+    edit: true
+    persona: coder
+    policy:
+      - coding
+      - testing
+    knowledge:
+      - backend
+      - architecture
+    allowed_tools:
+      - Read
+      - Glob
+      - Grep
+      - Edit
+      - Write
+      - Bash
+      - WebSearch
+      - WebFetch
+    permission_mode: edit
+    rules:
+      - condition: Fix complete
+        next: reviewers
+      - condition: Cannot proceed, insufficient info
+        next: plan_reform
+    instruction: fix
+
+  - name: verify
+    edit: false
+    persona: supervisor
+    policy: review
+    allowed_tools:
+      - Read
+      - Glob
+      - Grep
+      - Bash
+      - WebSearch
+      - WebFetch
+    instruction_template: |
+      ## Piece Status
+      - Iteration: {iteration}/{max_iterations} (piece-wide)
+      - Movement Iteration: {movement_iteration} (times this movement has run)
+      - Movement: verify (build and test verification)
+
+      ## Instructions
+      Verify that the current reform step has been completed successfully.
+
+      **Verification checklist:**
+      1. **Build**: Run the build command and confirm it passes
+      2. **Tests**: Run the test suite and confirm all tests pass
+      3. **File sizes**: Confirm no new file exceeds 300 lines
+      4. **Single responsibility**: Confirm each new file has a clear, single purpose
+      5. **Import consistency**: Confirm all imports are updated correctly
+
+      **Report format:**
+
+      ```markdown
+      # Verification Results
+
+      ## Result: PASS / FAIL
+
+      | Check | Status | Details |
+      |-------|--------|---------|
+      | Build | PASS/FAIL | {output summary} |
+      | Tests | PASS/FAIL | {N passed, N failed} |
+      | File sizes | PASS/FAIL | {any file > 300 lines} |
+      | Single responsibility | PASS/FAIL | {assessment} |
+      | Import consistency | PASS/FAIL | {any broken imports} |
+
+      ## Issues (if FAIL)
+      1. {issue description}
+      ```
+    rules:
+      - condition: All verifications passed
+        next: next_target
+      - condition: Verification failed
+        next: fix
+    output_contracts:
+      report:
+        - name: 06-verification.md
+          format: validation
+
+  - name: next_target
+    edit: false
+    persona: planner
+    knowledge: architecture
+    allowed_tools:
+      - Read
+      - Glob
+      - Grep
+      - Bash
+      - WebSearch
+      - WebFetch
+    instruction_template: |
+      ## Piece Status
+      - Iteration: {iteration}/{max_iterations} (piece-wide)
+      - Movement Iteration: {movement_iteration} (times this movement has run)
+      - Movement: next_target (progress check and next target selection)
+
+      ## Original Reform Plan
+      {report:01-reform-plan.md}
+
+      ## Latest Verification
+      {previous_response}
+
+      ## Instructions
+      Assess the progress of the structural reform and determine the next action.
+
+      **Steps:**
+      1. Review the reform plan and identify which targets have been completed
+      2. Check the current codebase state against the plan
+      3. Determine if there are remaining reform targets
+
+      **Output format:**
+
+      ```markdown
+      # Reform Progress
+
+      ## Completed Targets
+      | # | Target | Status |
+      |---|--------|--------|
+      | 1 | {file} | Completed |
+      | 2 | {file} | Completed |
+
+      ## Remaining Targets
+      | # | Target | Priority |
+      |---|--------|----------|
+      | 3 | {file} | Next |
+      | 4 | {file} | Pending |
+
+      ## Next Action
+      - **Target**: {next file to reform}
+      - **Plan**: {brief description of the split}
+
+      ## Overall Progress
+      {N}/{total} targets completed. Estimated remaining iterations: {N}
+      ```
+    rules:
+      - condition: More reform targets remain
+        next: implement
+      - condition: All reform targets completed
+        next: COMPLETE
+    output_contracts:
+      report:
+        - name: 07-progress.md
+report_formats:
+  plan: ../output-contracts/plan.md
+  architecture-review: ../output-contracts/architecture-review.md
+  qa-review: ../output-contracts/qa-review.md
+  validation: ../output-contracts/validation.md
+  summary: ../output-contracts/summary.md

builtins/en/pieces/unit-test.yaml

@@ -1,37 +1,32 @@
-# Auto-generated from default.yaml by tools/generate-hybrid-codex.mjs
-# Do not edit manually. Edit the source piece and re-run the generator.
-
-name: default-hybrid-codex
-description: Standard development piece with planning and specialized reviews
-max_iterations: 30
+name: unit-test
+description: Unit test focused piece (test analysis → test implementation → review → fix)
+max_iterations: 20
+policies:
+  coding: ../policies/coding.md
+  review: ../policies/review.md
+  testing: ../policies/testing.md
+  ai-antipattern: ../policies/ai-antipattern.md
+  qa: ../policies/qa.md
 knowledge:
-  backend: ../knowledge/backend.md
   architecture: ../knowledge/architecture.md
+  backend: ../knowledge/backend.md
 personas:
-  planner: ../personas/planner.md
+  test-planner: ../personas/test-planner.md
   coder: ../personas/coder.md
   ai-antipattern-reviewer: ../personas/ai-antipattern-reviewer.md
   architecture-reviewer: ../personas/architecture-reviewer.md
   qa-reviewer: ../personas/qa-reviewer.md
   supervisor: ../personas/supervisor.md
 instructions:
-  plan: ../instructions/plan.md
-  implement: ../instructions/implement.md
+  plan-test: ../instructions/plan-test.md
+  implement-test: ../instructions/implement-test.md
   ai-review: ../instructions/ai-review.md
   ai-fix: ../instructions/ai-fix.md
   arbitrate: ../instructions/arbitrate.md
-  review-arch: ../instructions/review-arch.md
-  review-qa: ../instructions/review-qa.md
+  review-test: ../instructions/review-test.md
   fix: ../instructions/fix.md
   supervise: ../instructions/supervise.md
-report_formats:
-  plan: ../output-contracts/plan.md
-  ai-review: ../output-contracts/ai-review.md
-  architecture-review: ../output-contracts/architecture-review.md
-  qa-review: ../output-contracts/qa-review.md
-  validation: ../output-contracts/validation.md
-  summary: ../output-contracts/summary.md
-initial_movement: plan
+initial_movement: plan_test
 loop_monitors:
   - cycle:
       - ai_review
@@ -56,12 +51,15 @@ loop_monitors:
         - condition: Healthy (making progress)
           next: ai_review
         - condition: Unproductive (no improvement)
-          next: reviewers
+          next: review_test
 movements:
-  - name: plan
+  - name: plan_test
     edit: false
-    persona: planner
-    knowledge: architecture
+    persona: test-planner
+    policy: testing
+    knowledge:
+      - architecture
+      - backend
     allowed_tools:
       - Read
       - Glob
@@ -70,9 +68,9 @@ movements:
       - WebSearch
       - WebFetch
     rules:
-      - condition: Requirements are clear and implementable
-        next: implement
-      - condition: User is asking a question (not an implementation task)
+      - condition: Test plan complete
+        next: implement_test
+      - condition: User is asking a question (not a test task)
         next: COMPLETE
       - condition: Requirements unclear, insufficient info
         next: ABORT
@@ -80,15 +78,15 @@ movements:
           Clarifications needed:
           - {Question 1}
           - {Question 2}
-    instruction: plan
+    instruction: plan-test
     output_contracts:
       report:
-        - name: 00-plan.md
-          format: plan
-  - name: implement
+        - name: 00-test-plan.md
+          format: test-plan
+
+  - name: implement_test
     edit: true
     persona: coder
-    provider: codex
     policy:
       - coding
       - testing
@@ -107,21 +105,22 @@ movements:
       - WebFetch
     permission_mode: edit
     rules:
-      - condition: Implementation complete
+      - condition: Test implementation complete
         next: ai_review
       - condition: No implementation (report only)
         next: ai_review
       - condition: Cannot proceed, insufficient info
         next: ai_review
       - condition: User input required
-        next: implement
+        next: implement_test
         requires_user_input: true
         interactive_only: true
-    instruction: implement
+    instruction: implement-test
     output_contracts:
       report:
         - Scope: 02-coder-scope.md
         - Decisions: 03-coder-decisions.md
+
   - name: ai_review
     edit: false
     persona: ai-antipattern-reviewer
@@ -136,7 +135,7 @@ movements:
       - WebFetch
     rules:
       - condition: No AI-specific issues
-        next: reviewers
+        next: review_test
       - condition: AI-specific issues found
         next: ai_fix
     instruction: ai-review
@@ -144,10 +143,10 @@ movements:
       report:
         - name: 04-ai-review.md
           format: ai-review
+
   - name: ai_fix
     edit: true
     persona: coder
-    provider: codex
     policy:
       - coding
       - testing
@@ -173,6 +172,7 @@ movements:
       - condition: Cannot proceed, insufficient info
         next: ai_no_fix
     instruction: ai-fix
+
   - name: ai_no_fix
     edit: false
     persona: architecture-reviewer
@@ -185,63 +185,39 @@ movements:
       - condition: ai_review's findings are valid (fix required)
         next: ai_fix
       - condition: ai_fix's judgment is valid (no fix needed)
-        next: reviewers
+        next: review_test
     instruction: arbitrate
-  - name: reviewers
-    parallel:
-      - name: arch-review
-        edit: false
-        persona: architecture-reviewer
-        policy: review
-        knowledge:
-          - architecture
-          - backend
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-arch
-        output_contracts:
-          report:
-            - name: 05-architect-review.md
-              format: architecture-review
-      - name: qa-review
-        edit: false
-        persona: qa-reviewer
-        policy:
-          - review
-          - qa
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-qa
-        output_contracts:
-          report:
-            - name: 06-qa-review.md
-              format: qa-review
+
+  - name: review_test
+    edit: false
+    persona: qa-reviewer
+    policy:
+      - review
+      - qa
+    allowed_tools:
+      - Read
+      - Glob
+      - Grep
+      - WebSearch
+      - WebFetch
     rules:
-      - condition: all("approved")
+      - condition: approved
         next: supervise
-      - condition: any("needs_fix")
+      - condition: needs_fix
         next: fix
+    instruction: review-test
+    output_contracts:
+      report:
+        - name: 05-qa-review.md
+          format: qa-review
+
   - name: fix
     edit: true
     persona: coder
-    provider: codex
     policy:
       - coding
       - testing
+    session: refresh
     knowledge:
       - backend
       - architecture
@@ -257,10 +233,11 @@ movements:
     permission_mode: edit
     rules:
       - condition: Fix complete
-        next: reviewers
+        next: review_test
       - condition: Cannot proceed, insufficient info
-        next: plan
+        next: plan_test
     instruction: fix
+
   - name: supervise
     edit: false
     persona: supervisor
@@ -276,15 +253,15 @@ movements:
       - condition: All checks passed
         next: COMPLETE
       - condition: Requirements unmet, tests failing, build errors
-        next: plan
+        next: plan_test
     instruction: supervise
     output_contracts:
       report:
-        - Validation: 07-supervisor-validation.md
+        - Validation: 06-supervisor-validation.md
         - Summary: summary.md
-policies:
-  coding: ../policies/coding.md
-  review: ../policies/review.md
-  testing: ../policies/testing.md
-  ai-antipattern: ../policies/ai-antipattern.md
-  qa: ../policies/qa.md
+report_formats:
+  test-plan: ../output-contracts/test-plan.md
+  ai-review: ../output-contracts/ai-review.md
+  qa-review: ../output-contracts/qa-review.md
+  validation: ../output-contracts/validation.md
+  summary: ../output-contracts/summary.md

builtins/en/policies/review.md

@@ -86,6 +86,18 @@ Every issue raised must include the following.
      Extract into a shared function."
 ```
 
+## Finding ID Tracking (`finding_id`)
+
+To prevent circular rejections, track findings by ID.
+
+- Every issue raised in a REJECT must include a `finding_id`
+- If the same issue is raised again, reuse the same `finding_id`
+- For repeated issues, set status to `persists` and include concrete evidence (file/line) that it remains unresolved
+- New issues must use status `new`
+- Resolved issues must be listed with status `resolved`
+- Issues without `finding_id` are invalid (cannot be used as rejection grounds)
+- REJECT is allowed only when there is at least one `new` or `persists` issue
+
 ## Boy Scout Rule
 
 Leave it better than you found it.

builtins/ja/config.yaml

@@ -1,34 +1,85 @@
 # TAKT グローバル設定
-# takt のデフォルト設定ファイルです。
+# 配置場所: ~/.takt/config.yaml
 
-# 言語設定 (en または ja)
+# ── 基本設定 ──
+
+# 言語 (en | ja)
 language: ja
 
-# デフォルトのピース - 指定がない場合に使用します
+# デフォルトピース（指定なし時に使用）
 default_piece: default
 
-# ログレベル: debug, info, warn, error
+# ログレベル (debug | info | warn | error)
 log_level: info
 
-# プロバイダー: claude または codex
+# ── プロバイダー & モデル ──
+
+# プロバイダー (claude | codex)
 provider: claude
 
-# ビルトインピースの読み込み (resources/global/{lang}/pieces)
-# enable_builtin_pieces: true
-
-# デフォルトモデル (オプション)
-# Claude: opus, sonnet, haiku, opusplan, default, またはフルモデル名
+# デフォルトモデル（オプション）
+# Claude: opus, sonnet, haiku
 # Codex: gpt-5.2-codex, gpt-5.1-codex など
 # model: sonnet
 
-# Anthropic APIキー (オプション、環境変数 TAKT_ANTHROPIC_API_KEY で上書き可能)
-# anthropic_api_key: ""
+# ペルソナ単位のプロバイダー上書き（オプション）
+# 特定ペルソナだけプロバイダーを変更。未指定のペルソナはグローバル設定を使用。
+# persona_providers:
+#   coder: codex
 
-# OpenAI APIキー (オプション、環境変数 TAKT_OPENAI_API_KEY で上書き可能)
+# ── APIキー ──
+# オプション。環境変数が優先:
+#   TAKT_ANTHROPIC_API_KEY, TAKT_OPENAI_API_KEY
+
+# anthropic_api_key: ""
 # openai_api_key: ""
 
-# パイプライン実行設定 (オプション)
-# パイプラインモード (--task) のブランチ名、コミットメッセージ、PRの本文をカスタマイズできます。
+# ── 実行設定 ──
+
+# ワークツリー（shared clone）ディレクトリ（デフォルト: プロジェクトの ../{clone-name}）
+# worktree_dir: ~/takt-worktrees
+
+# ワークツリー実行後に自動PR作成（デフォルト: 対話モードで確認）
+# auto_pr: false
+
+# macOS のアイドルスリープを防止（デフォルト: false）
+# prevent_sleep: false
+
+# ── 並列実行 (takt run) ──
+
+# タスクの同時実行数（1 = 逐次実行、最大: 10）
+# concurrency: 1
+
+# 新規タスクのポーリング間隔 ms（100-5000、デフォルト: 500）
+# task_poll_interval_ms: 500
+
+# ── 対話モード ──
+
+# ムーブメントプレビューの表示数（0 で無効、最大: 10）
+# interactive_preview_movements: 3
+
+# ブランチ名の生成方式（romaji: 高速デフォルト | ai: 低速）
+# branch_name_strategy: romaji
+
+# ── 出力 ──
+
+# 通知音（デフォルト: true）
+# notification_sound: true
+
+# CI 向け最小出力 - AI 出力を抑制（デフォルト: false）
+# minimal_output: false
+
+# ── ビルトインピース ──
+
+# ビルトインピースの有効化（デフォルト: true）
+# enable_builtin_pieces: true
+
+# 特定のビルトインを除外
+# disabled_builtins:
+#   - magi
+
+# ── パイプラインモード (--pipeline) ──
+
 # pipeline:
 #   default_branch_prefix: "takt/"
 #   commit_message_template: "feat: {title} (#{issue})"
@@ -37,10 +88,14 @@ provider: claude
 #     {issue_body}
 #     Closes #{issue}
 
-# 通知音 (true: 有効 / false: 無効、デフォルト: true)
-# notification_sound: true
+# ── プリファレンス ──
+
+# プリファレンスファイルのカスタムパス
+# bookmarks_file: ~/.takt/preferences/bookmarks.yaml
+# piece_categories_file: ~/.takt/preferences/piece-categories.yaml
+
+# ── デバッグ ──
 
-# デバッグ設定 (オプション)
 # debug:
 #   enabled: false
 #   log_file: ~/.takt/logs/debug.log

builtins/ja/instructions/implement-test.md

@@ -0,0 +1,52 @@
+テスト計画に従って単体テストを実装してください。
+Piece Contextに示されたReport Directory内のファイルのみ参照してください。他のレポートディレクトリは検索/参照しないでください。
+
+**重要: プロダクションコードは変更しないでください。テストファイルのみ編集可能です。**
+
+**やること:**
+1. テスト計画のレポートを確認する
+2. 計画されたテストケースを実装する
+3. テストを実行して全パスを確認する
+4. 既存テストが壊れていないことを確認する
+
+**テスト実装の制約:**
+- プロジェクトの既存テストパターン（命名規約、ディレクトリ構成、ヘルパー）に従う
+- Given-When-Then 構造で記述する
+- 1テスト1概念。複数の関心事を1テストに混ぜない
+
+**Scope出力契約（実装開始時に作成）:**
+```markdown
+# 変更スコープ宣言
+
+## タスク
+{タスクの1行要約}
+
+## 変更予定
+| 種別 | ファイル |
+|------|---------|
+| 作成 | `src/__tests__/example.test.ts` |
+
+## 推定規模
+Small / Medium / Large
+
+## 影響範囲
+- {影響するモジュールや機能}
+```
+
+**Decisions出力契約（実装完了時、決定がある場合のみ）:**
+```markdown
+# 決定ログ
+
+## 1. {決定内容}
+- **背景**: {なぜ決定が必要だったか}
+- **検討した選択肢**: {選択肢リスト}
+- **理由**: {選んだ理由}
+```
+
+**必須出力（見出しを含める）**
+## 作業結果
+- {実施内容の要約}
+## 変更内容
+- {変更内容の要約}
+## テスト結果
+- {実行コマンドと結果}

builtins/ja/instructions/plan-test.md

@@ -0,0 +1,11 @@
+対象コードを分析し、不足している単体テストを洗い出してください。
+
+**注意:** Previous Responseがある場合は差し戻しのため、
+その内容を踏まえてテスト計画を見直してください。
+
+**やること:**
+1. 対象モジュールのソースコードを読み、振る舞い・分岐・状態遷移を理解する
+2. 既存テストを読み、カバーされている観点を把握する
+3. 不足しているテストケース（正常系・異常系・境界値・エッジケース）を洗い出す
+4. テスト方針（モック戦略、既存テストヘルパーの活用、フィクスチャ設計）を決める
+5. テスト実装者向けの具体的なガイドラインを出す

builtins/ja/instructions/review-arch.md

@@ -9,8 +9,14 @@ AI特有の問題はレビューしないでください（ai_reviewムーブメ
 - デッドコード
 - 呼び出しチェーン検証
 
+**前回指摘の追跡（必須）:**
+- まず「Previous Response」から前回の open findings を抽出する
+- 各 finding に `finding_id` を付け、今回の状態を `new / persists / resolved` で判定する
+- `persists` と判定する場合は、未解決である根拠（ファイル/行）を必ず示す
+
 ## 判定手順
 
-1. 変更差分を確認し、構造・設計の観点に基づいて問題を検出する
-2. 検出した問題ごとに、Policyのスコープ判定表と判定ルールに基づいてブロッキング/非ブロッキングを分類する
-3. ブロッキング問題が1件でもあればREJECTと判定する
+1. まず前回open findingsを抽出し、`new / persists / resolved` を仮判定する
+2. 変更差分を確認し、構造・設計の観点に基づいて問題を検出する
+3. 検出した問題ごとに、Policyのスコープ判定表と判定ルールに基づいてブロッキング/非ブロッキングを分類する
+4. ブロッキング問題（`new` または `persists`）が1件でもあればREJECTと判定する

builtins/ja/instructions/review-qa.md

@@ -7,8 +7,14 @@
 - ログとモニタリング
 - 保守性
 
+**前回指摘の追跡（必須）:**
+- まず「Previous Response」から前回の open findings を抽出する
+- 各 finding に `finding_id` を付け、今回の状態を `new / persists / resolved` で判定する
+- `persists` と判定する場合は、未解決である根拠（ファイル/行）を必ず示す
+
 ## 判定手順
 
-1. 変更差分を確認し、品質保証の観点に基づいて問題を検出する
-2. 検出した問題ごとに、Policyのスコープ判定表と判定ルールに基づいてブロッキング/非ブロッキングを分類する
-3. ブロッキング問題が1件でもあればREJECTと判定する
+1. まず前回open findingsを抽出し、`new / persists / resolved` を仮判定する
+2. 変更差分を確認し、品質保証の観点に基づいて問題を検出する
+3. 検出した問題ごとに、Policyのスコープ判定表と判定ルールに基づいてブロッキング/非ブロッキングを分類する
+4. ブロッキング問題（`new` または `persists`）が1件でもあればREJECTと判定する

builtins/ja/instructions/review-test.md

@@ -0,0 +1,14 @@
+テスト品質の観点から変更をレビューしてください。
+
+**レビュー観点:**
+- テスト計画の観点がすべてカバーされているか
+- テスト品質（Given-When-Then構造、独立性、再現性）
+- テスト命名規約
+- 過不足（不要なテスト、足りないケース）
+- モック・フィクスチャの適切さ
+
+## 判定手順
+
+1. テスト計画レポート（{report:00-test-plan.md}）と実装されたテストを突合する
+2. 検出した問題ごとに、Policyのスコープ判定表と判定ルールに基づいてブロッキング/非ブロッキングを分類する
+3. ブロッキング問題が1件でもあればREJECTと判定する

builtins/ja/output-contracts/architecture-review.md

@@ -14,10 +14,15 @@
 - [x] デッドコード
 - [x] 呼び出しチェーン検証
 
+## 前回Open Findings
+| finding_id | 前回状態 | 今回状態(new/persists/resolved) | 根拠 |
+|------------|----------|----------------------------------|------|
+| ARCH-EXAMPLE-src-file-L42 | open | persists | `src/file.ts:42` |
+
 ## 問題点（REJECTの場合）
-| # | スコープ | 場所 | 問題 | 修正案 |
-|---|---------|------|------|--------|
-| 1 | スコープ内 | `src/file.ts:42` | 問題の説明 | 修正方法 |
+| # | finding_id | 状態(new/persists) | スコープ | 場所 | 問題 | 修正案 |
+|---|------------|--------------------|---------|------|------|--------|
+| 1 | ARCH-EXAMPLE-src-file-L42 | new | スコープ内 | `src/file.ts:42` | 問題の説明 | 修正方法 |
 
 スコープ: 「スコープ内」（今回修正可能）/ 「スコープ外」（既存問題・非ブロッキング）
 

builtins/ja/output-contracts/qa-review.md

@@ -15,8 +15,13 @@
 | ドキュメント | ✅ | - |
 | 保守性 | ✅ | - |
 
+## 前回Open Findings
+| finding_id | 前回状態 | 今回状態(new/persists/resolved) | 根拠 |
+|------------|----------|----------------------------------|------|
+| QA-EXAMPLE-src-file-L42 | open | persists | `src/file.ts:42` |
+
 ## 問題点（REJECTの場合）
-| # | カテゴリ | 問題 | 修正案 |
-|---|---------|------|--------|
-| 1 | テスト | 問題の説明 | 修正方法 |
+| # | finding_id | 状態(new/persists) | カテゴリ | 問題 | 修正案 |
+|---|------------|--------------------|---------|------|--------|
+| 1 | QA-EXAMPLE-src-file-L42 | new | テスト | 問題の説明 | 修正方法 |
 ```

builtins/ja/output-contracts/test-plan.md

@@ -0,0 +1,24 @@
+```markdown
+# テスト計画
+
+## 対象モジュール
+{分析対象のモジュール一覧}
+
+## 既存テストの分析
+| モジュール | 既存テスト | カバレッジ状況 |
+|-----------|-----------|--------------|
+| `src/xxx.ts` | `xxx.test.ts` | {カバー状況} |
+
+## 不足テストケース
+| # | 対象 | テストケース | 優先度 | 理由 |
+|---|------|------------|--------|------|
+| 1 | `src/xxx.ts` | {テストケース概要} | 高/中/低 | {理由} |
+
+## テスト方針
+- {モック戦略}
+- {フィクスチャ設計}
+- {テストヘルパー活用}
+
+## 実装ガイドライン
+- {テスト実装者向けの具体的指示}
+```

builtins/ja/personas/test-planner.md

@@ -0,0 +1,25 @@
+# Test Planner
+
+あなたはテスト分析と計画の専門家です。対象コードの振る舞いを理解し、既存テストのカバレッジを分析して、不足しているテストケースを体系的に洗い出す。
+
+## 役割の境界
+
+**やること:**
+- 対象コードの振る舞い・分岐・状態遷移を読み解く
+- 既存テストのカバレッジを分析する
+- 不足しているテストケース（正常系・異常系・境界値・エッジケース）を洗い出す
+- テスト戦略（モック方針、フィクスチャ設計、テストヘルパー活用）を決める
+- テスト実装者への具体的なガイドラインを出す
+
+**やらないこと:**
+- プロダクションコードの変更計画（Plannerの仕事）
+- テストコードの実装（Coderの仕事）
+- コードレビュー（Reviewerの仕事）
+
+## 行動姿勢
+
+- コードを読んでから計画する。推測でテストケースを列挙しない
+- 既存テストを必ず確認する。カバー済みの観点を重複して計画しない
+- テスト優先度を付ける。ビジネスロジック・状態遷移 > エッジケース > 単純なCRUD
+- テスト実装者が迷わない粒度で指示を出す
+- プロジェクトの既存テストパターンに合わせる。独自の書き方を提案しない

builtins/ja/piece-categories.yaml

@@ -10,26 +10,17 @@ piece_categories:
     pieces:
       - review-fix-minimal
       - review-only
+      - unit-test
   🎨 フロントエンド: {}
   ⚙️ バックエンド: {}
-  🔧 フルスタック:
-    pieces:
-      - expert
-      - expert-cqrs
-  🔀 ハイブリッド (Codex Coding):
-    🚀 クイックスタート:
+  🔧 エキスパート:
+    フルスタック:
       pieces:
-        - coding-hybrid-codex
-        - default-hybrid-codex
-        - minimal-hybrid-codex
-        - passthrough-hybrid-codex
-    🔧 フルスタック:
+        - expert
+        - expert-cqrs
+    リファクタリング:
       pieces:
-        - expert-cqrs-hybrid-codex
-        - expert-hybrid-codex
-    🔍 レビュー＆修正:
-      pieces:
-        - review-fix-minimal-hybrid-codex
+        - structural-reform
   その他:
     pieces:
       - research

builtins/ja/pieces/coding-hybrid-codex.yaml

@@ -1,158 +0,0 @@
-# Auto-generated from coding.yaml by tools/generate-hybrid-codex.mjs
-# Do not edit manually. Edit the source piece and re-run the generator.
-
-name: coding-hybrid-codex
-description: Lightweight development piece with planning and parallel reviews (plan -> implement -> parallel review -> complete)
-max_iterations: 20
-knowledge:
-  architecture: ../knowledge/architecture.md
-personas:
-  planner: ../personas/planner.md
-  coder: ../personas/coder.md
-  ai-antipattern-reviewer: ../personas/ai-antipattern-reviewer.md
-  architecture-reviewer: ../personas/architecture-reviewer.md
-instructions:
-  plan: ../instructions/plan.md
-  implement: ../instructions/implement.md
-  ai-review: ../instructions/ai-review.md
-  review-arch: ../instructions/review-arch.md
-  fix: ../instructions/fix.md
-report_formats:
-  plan: ../output-contracts/plan.md
-  ai-review: ../output-contracts/ai-review.md
-  architecture-review: ../output-contracts/architecture-review.md
-initial_movement: plan
-movements:
-  - name: plan
-    edit: false
-    persona: planner
-    knowledge: architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Bash
-      - WebSearch
-      - WebFetch
-    rules:
-      - condition: 要件が明確で実装可能
-        next: implement
-      - condition: ユーザーが質問をしている（実装タスクではない）
-        next: COMPLETE
-      - condition: 要件が不明確、情報不足
-        next: ABORT
-    instruction: plan
-    output_contracts:
-      report:
-        - name: 00-plan.md
-          format: plan
-  - name: implement
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    session: refresh
-    knowledge: architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: 実装完了
-        next: reviewers
-      - condition: 実装未着手（レポートのみ）
-        next: reviewers
-      - condition: 判断できない、情報不足
-        next: reviewers
-      - condition: ユーザー入力が必要
-        next: implement
-        requires_user_input: true
-        interactive_only: true
-    instruction: implement
-    output_contracts:
-      report:
-        - Scope: 02-coder-scope.md
-        - Decisions: 03-coder-decisions.md
-  - name: reviewers
-    parallel:
-      - name: ai_review
-        edit: false
-        persona: ai-antipattern-reviewer
-        policy:
-          - review
-          - ai-antipattern
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: AI特有の問題なし
-          - condition: AI特有の問題あり
-        instruction: ai-review
-        output_contracts:
-          report:
-            - name: 04-ai-review.md
-              format: ai-review
-      - name: arch-review
-        edit: false
-        persona: architecture-reviewer
-        policy: review
-        knowledge: architecture
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-arch
-        output_contracts:
-          report:
-            - name: 05-architect-review.md
-              format: architecture-review
-    rules:
-      - condition: all("AI特有の問題なし", "approved")
-        next: COMPLETE
-      - condition: any("AI特有の問題あり", "needs_fix")
-        next: fix
-  - name: fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    knowledge: architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: 修正完了
-        next: reviewers
-      - condition: 判断できない、情報不足
-        next: ABORT
-    instruction: fix
-policies:
-  coding: ../policies/coding.md
-  review: ../policies/review.md
-  testing: ../policies/testing.md
-  ai-antipattern: ../policies/ai-antipattern.md

builtins/ja/pieces/expert-cqrs-hybrid-codex.yaml

@@ -1,341 +0,0 @@
-# Auto-generated from expert-cqrs.yaml by tools/generate-hybrid-codex.mjs
-# Do not edit manually. Edit the source piece and re-run the generator.
-
-name: expert-cqrs-hybrid-codex
-description: CQRS+ES・フロントエンド・セキュリティ・QA専門家レビュー
-max_iterations: 30
-knowledge:
-  frontend: ../knowledge/frontend.md
-  backend: ../knowledge/backend.md
-  cqrs-es: ../knowledge/cqrs-es.md
-  security: ../knowledge/security.md
-  architecture: ../knowledge/architecture.md
-personas:
-  planner: ../personas/planner.md
-  coder: ../personas/coder.md
-  ai-antipattern-reviewer: ../personas/ai-antipattern-reviewer.md
-  architecture-reviewer: ../personas/architecture-reviewer.md
-  cqrs-es-reviewer: ../personas/cqrs-es-reviewer.md
-  frontend-reviewer: ../personas/frontend-reviewer.md
-  security-reviewer: ../personas/security-reviewer.md
-  qa-reviewer: ../personas/qa-reviewer.md
-  expert-supervisor: ../personas/expert-supervisor.md
-instructions:
-  plan: ../instructions/plan.md
-  implement: ../instructions/implement.md
-  ai-review: ../instructions/ai-review.md
-  ai-fix: ../instructions/ai-fix.md
-  arbitrate: ../instructions/arbitrate.md
-  review-cqrs-es: ../instructions/review-cqrs-es.md
-  review-frontend: ../instructions/review-frontend.md
-  review-security: ../instructions/review-security.md
-  review-qa: ../instructions/review-qa.md
-  fix: ../instructions/fix.md
-  supervise: ../instructions/supervise.md
-  fix-supervisor: ../instructions/fix-supervisor.md
-report_formats:
-  plan: ../output-contracts/plan.md
-  ai-review: ../output-contracts/ai-review.md
-  cqrs-es-review: ../output-contracts/cqrs-es-review.md
-  frontend-review: ../output-contracts/frontend-review.md
-  security-review: ../output-contracts/security-review.md
-  qa-review: ../output-contracts/qa-review.md
-  validation: ../output-contracts/validation.md
-  summary: ../output-contracts/summary.md
-initial_movement: plan
-movements:
-  - name: plan
-    edit: false
-    persona: planner
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Bash
-      - WebSearch
-      - WebFetch
-    instruction: plan
-    rules:
-      - condition: タスク分析と計画が完了した
-        next: implement
-      - condition: 要件が不明確で計画を立てられない
-        next: ABORT
-    output_contracts:
-      report:
-        - name: 00-plan.md
-          format: plan
-  - name: implement
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    session: refresh
-    knowledge:
-      - frontend
-      - backend
-      - cqrs-es
-      - security
-      - architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    instruction: implement
-    rules:
-      - condition: 実装が完了した
-        next: ai_review
-      - condition: 実装未着手（レポートのみ）
-        next: ai_review
-      - condition: 実装を進行できない
-        next: ai_review
-      - condition: ユーザー入力が必要
-        next: implement
-        requires_user_input: true
-        interactive_only: true
-    output_contracts:
-      report:
-        - Scope: 01-coder-scope.md
-        - Decisions: 02-coder-decisions.md
-  - name: ai_review
-    edit: false
-    persona: ai-antipattern-reviewer
-    policy:
-      - review
-      - ai-antipattern
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - WebSearch
-      - WebFetch
-    instruction: ai-review
-    rules:
-      - condition: AI特有の問題が見つからない
-        next: reviewers
-      - condition: AI特有の問題が検出された
-        next: ai_fix
-    output_contracts:
-      report:
-        - name: 03-ai-review.md
-          format: ai-review
-  - name: ai_fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    session: refresh
-    knowledge:
-      - frontend
-      - backend
-      - cqrs-es
-      - security
-      - architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    instruction: ai-fix
-    rules:
-      - condition: AI Reviewerの指摘に対する修正が完了した
-        next: ai_review
-      - condition: 修正不要（指摘対象ファイル/仕様の確認済み）
-        next: ai_no_fix
-      - condition: 修正を進行できない
-        next: ai_no_fix
-  - name: ai_no_fix
-    edit: false
-    persona: architecture-reviewer
-    policy: review
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-    rules:
-      - condition: ai_reviewの指摘が妥当（修正すべき）
-        next: ai_fix
-      - condition: ai_fixの判断が妥当（修正不要）
-        next: reviewers
-    instruction: arbitrate
-  - name: reviewers
-    parallel:
-      - name: cqrs-es-review
-        edit: false
-        persona: cqrs-es-reviewer
-        policy: review
-        knowledge:
-          - cqrs-es
-          - backend
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-cqrs-es
-        output_contracts:
-          report:
-            - name: 04-cqrs-es-review.md
-              format: cqrs-es-review
-      - name: frontend-review
-        edit: false
-        persona: frontend-reviewer
-        policy: review
-        knowledge: frontend
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-frontend
-        output_contracts:
-          report:
-            - name: 05-frontend-review.md
-              format: frontend-review
-      - name: security-review
-        edit: false
-        persona: security-reviewer
-        policy: review
-        knowledge: security
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-security
-        output_contracts:
-          report:
-            - name: 06-security-review.md
-              format: security-review
-      - name: qa-review
-        edit: false
-        persona: qa-reviewer
-        policy:
-          - review
-          - qa
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-qa
-        output_contracts:
-          report:
-            - name: 07-qa-review.md
-              format: qa-review
-    rules:
-      - condition: all("approved")
-        next: supervise
-      - condition: any("needs_fix")
-        next: fix
-  - name: fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    knowledge:
-      - frontend
-      - backend
-      - cqrs-es
-      - security
-      - architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: 修正が完了した
-        next: reviewers
-      - condition: 修正を進行できない
-        next: plan
-    instruction: fix
-  - name: supervise
-    edit: false
-    persona: expert-supervisor
-    policy: review
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - WebSearch
-      - WebFetch
-    instruction: supervise
-    rules:
-      - condition: すべての検証が完了し、マージ可能な状態である
-        next: COMPLETE
-      - condition: 問題が検出された
-        next: fix_supervisor
-    output_contracts:
-      report:
-        - Validation: 08-supervisor-validation.md
-        - Summary: summary.md
-  - name: fix_supervisor
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    knowledge:
-      - frontend
-      - backend
-      - cqrs-es
-      - security
-      - architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    instruction: fix-supervisor
-    rules:
-      - condition: 監督者の指摘に対する修正が完了した
-        next: supervise
-      - condition: 修正を進行できない
-        next: plan
-policies:
-  coding: ../policies/coding.md
-  review: ../policies/review.md
-  testing: ../policies/testing.md
-  ai-antipattern: ../policies/ai-antipattern.md
-  qa: ../policies/qa.md

builtins/ja/pieces/expert-hybrid-codex.yaml

@@ -1,335 +0,0 @@
-# Auto-generated from expert.yaml by tools/generate-hybrid-codex.mjs
-# Do not edit manually. Edit the source piece and re-run the generator.
-
-name: expert-hybrid-codex
-description: アーキテクチャ・フロントエンド・セキュリティ・QA専門家レビュー
-max_iterations: 30
-knowledge:
-  frontend: ../knowledge/frontend.md
-  backend: ../knowledge/backend.md
-  security: ../knowledge/security.md
-  architecture: ../knowledge/architecture.md
-personas:
-  planner: ../personas/planner.md
-  coder: ../personas/coder.md
-  ai-antipattern-reviewer: ../personas/ai-antipattern-reviewer.md
-  architecture-reviewer: ../personas/architecture-reviewer.md
-  frontend-reviewer: ../personas/frontend-reviewer.md
-  security-reviewer: ../personas/security-reviewer.md
-  qa-reviewer: ../personas/qa-reviewer.md
-  expert-supervisor: ../personas/expert-supervisor.md
-instructions:
-  plan: ../instructions/plan.md
-  implement: ../instructions/implement.md
-  ai-review: ../instructions/ai-review.md
-  ai-fix: ../instructions/ai-fix.md
-  arbitrate: ../instructions/arbitrate.md
-  review-arch: ../instructions/review-arch.md
-  review-frontend: ../instructions/review-frontend.md
-  review-security: ../instructions/review-security.md
-  review-qa: ../instructions/review-qa.md
-  fix: ../instructions/fix.md
-  supervise: ../instructions/supervise.md
-  fix-supervisor: ../instructions/fix-supervisor.md
-report_formats:
-  plan: ../output-contracts/plan.md
-  ai-review: ../output-contracts/ai-review.md
-  architecture-review: ../output-contracts/architecture-review.md
-  frontend-review: ../output-contracts/frontend-review.md
-  security-review: ../output-contracts/security-review.md
-  qa-review: ../output-contracts/qa-review.md
-  validation: ../output-contracts/validation.md
-  summary: ../output-contracts/summary.md
-initial_movement: plan
-movements:
-  - name: plan
-    edit: false
-    persona: planner
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Bash
-      - WebSearch
-      - WebFetch
-    instruction: plan
-    rules:
-      - condition: タスク分析と計画が完了した
-        next: implement
-      - condition: 要件が不明確で計画を立てられない
-        next: ABORT
-    output_contracts:
-      report:
-        - name: 00-plan.md
-          format: plan
-  - name: implement
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    session: refresh
-    knowledge:
-      - frontend
-      - backend
-      - security
-      - architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    instruction: implement
-    rules:
-      - condition: 実装が完了した
-        next: ai_review
-      - condition: 実装未着手（レポートのみ）
-        next: ai_review
-      - condition: 実装を進行できない
-        next: ai_review
-      - condition: ユーザー入力が必要
-        next: implement
-        requires_user_input: true
-        interactive_only: true
-    output_contracts:
-      report:
-        - Scope: 01-coder-scope.md
-        - Decisions: 02-coder-decisions.md
-  - name: ai_review
-    edit: false
-    persona: ai-antipattern-reviewer
-    policy:
-      - review
-      - ai-antipattern
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - WebSearch
-      - WebFetch
-    instruction: ai-review
-    rules:
-      - condition: AI特有の問題が見つからない
-        next: reviewers
-      - condition: AI特有の問題が検出された
-        next: ai_fix
-    output_contracts:
-      report:
-        - name: 03-ai-review.md
-          format: ai-review
-  - name: ai_fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    session: refresh
-    knowledge:
-      - frontend
-      - backend
-      - security
-      - architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    instruction: ai-fix
-    rules:
-      - condition: AI Reviewerの指摘に対する修正が完了した
-        next: ai_review
-      - condition: 修正不要（指摘対象ファイル/仕様の確認済み）
-        next: ai_no_fix
-      - condition: 修正を進行できない
-        next: ai_no_fix
-  - name: ai_no_fix
-    edit: false
-    persona: architecture-reviewer
-    policy: review
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-    rules:
-      - condition: ai_reviewの指摘が妥当（修正すべき）
-        next: ai_fix
-      - condition: ai_fixの判断が妥当（修正不要）
-        next: reviewers
-    instruction: arbitrate
-  - name: reviewers
-    parallel:
-      - name: arch-review
-        edit: false
-        persona: architecture-reviewer
-        policy: review
-        knowledge:
-          - architecture
-          - backend
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-arch
-        output_contracts:
-          report:
-            - name: 04-architect-review.md
-              format: architecture-review
-      - name: frontend-review
-        edit: false
-        persona: frontend-reviewer
-        policy: review
-        knowledge: frontend
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-frontend
-        output_contracts:
-          report:
-            - name: 05-frontend-review.md
-              format: frontend-review
-      - name: security-review
-        edit: false
-        persona: security-reviewer
-        policy: review
-        knowledge: security
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-security
-        output_contracts:
-          report:
-            - name: 06-security-review.md
-              format: security-review
-      - name: qa-review
-        edit: false
-        persona: qa-reviewer
-        policy:
-          - review
-          - qa
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-qa
-        output_contracts:
-          report:
-            - name: 07-qa-review.md
-              format: qa-review
-    rules:
-      - condition: all("approved")
-        next: supervise
-      - condition: any("needs_fix")
-        next: fix
-  - name: fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    knowledge:
-      - frontend
-      - backend
-      - security
-      - architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: 修正が完了した
-        next: reviewers
-      - condition: 修正を進行できない
-        next: plan
-    instruction: fix
-  - name: supervise
-    edit: false
-    persona: expert-supervisor
-    policy: review
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - WebSearch
-      - WebFetch
-    instruction: supervise
-    rules:
-      - condition: すべての検証が完了し、マージ可能な状態である
-        next: COMPLETE
-      - condition: 問題が検出された
-        next: fix_supervisor
-    output_contracts:
-      report:
-        - Validation: 08-supervisor-validation.md
-        - Summary: summary.md
-  - name: fix_supervisor
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    knowledge:
-      - frontend
-      - backend
-      - security
-      - architecture
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    instruction: fix-supervisor
-    rules:
-      - condition: 監督者の指摘に対する修正が完了した
-        next: supervise
-      - condition: 修正を進行できない
-        next: plan
-policies:
-  coding: ../policies/coding.md
-  review: ../policies/review.md
-  testing: ../policies/testing.md
-  ai-antipattern: ../policies/ai-antipattern.md
-  qa: ../policies/qa.md

builtins/ja/pieces/minimal-hybrid-codex.yaml

@@ -1,202 +0,0 @@
-# Auto-generated from minimal.yaml by tools/generate-hybrid-codex.mjs
-# Do not edit manually. Edit the source piece and re-run the generator.
-
-name: minimal-hybrid-codex
-description: Minimal development piece (implement -> parallel review -> fix if needed -> complete)
-max_iterations: 20
-personas:
-  coder: ../personas/coder.md
-  ai-antipattern-reviewer: ../personas/ai-antipattern-reviewer.md
-  supervisor: ../personas/supervisor.md
-instructions:
-  implement: ../instructions/implement.md
-  review-ai: ../instructions/review-ai.md
-  ai-fix: ../instructions/ai-fix.md
-  supervise: ../instructions/supervise.md
-  fix-supervisor: ../instructions/fix-supervisor.md
-report_formats:
-  ai-review: ../output-contracts/ai-review.md
-initial_movement: implement
-movements:
-  - name: implement
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    instruction: implement
-    rules:
-      - condition: 実装が完了した
-        next: reviewers
-      - condition: 実装を進行できない
-        next: ABORT
-      - condition: ユーザーへの確認事項があるためユーザー入力が必要
-        next: implement
-        requires_user_input: true
-        interactive_only: true
-    output_contracts:
-      report:
-        - Scope: 01-coder-scope.md
-        - Decisions: 02-coder-decisions.md
-  - name: reviewers
-    parallel:
-      - name: ai_review
-        edit: false
-        persona: ai-antipattern-reviewer
-        policy:
-          - review
-          - ai-antipattern
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        instruction: review-ai
-        rules:
-          - condition: AI特有の問題なし
-          - condition: AI特有の問題あり
-        output_contracts:
-          report:
-            - name: 03-ai-review.md
-              format: ai-review
-      - name: supervise
-        edit: false
-        persona: supervisor
-        policy: review
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - Bash
-          - WebSearch
-          - WebFetch
-        instruction: supervise
-        rules:
-          - condition: すべて問題なし
-          - condition: 要求未達成、テスト失敗、ビルドエラー
-        output_contracts:
-          report:
-            - Validation: 05-supervisor-validation.md
-            - Summary: summary.md
-    rules:
-      - condition: all("AI特有の問題なし", "すべて問題なし")
-        next: COMPLETE
-      - condition: all("AI特有の問題あり", "要求未達成、テスト失敗、ビルドエラー")
-        next: fix_both
-      - condition: any("AI特有の問題あり")
-        next: ai_fix
-      - condition: any("要求未達成、テスト失敗、ビルドエラー")
-        next: supervise_fix
-  - name: fix_both
-    parallel:
-      - name: ai_fix_parallel
-        edit: true
-        persona: coder
-        provider: codex
-        policy:
-          - coding
-          - testing
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - Edit
-          - Bash
-          - WebSearch
-          - WebFetch
-        permission_mode: edit
-        rules:
-          - condition: AI問題の修正完了
-          - condition: 修正不要（指摘対象ファイル/仕様の確認済み）
-          - condition: 判断できない、情報不足
-        instruction: ai-fix
-      - name: supervise_fix_parallel
-        edit: true
-        persona: coder
-        provider: codex
-        policy:
-          - coding
-          - testing
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - Edit
-          - Bash
-          - WebSearch
-          - WebFetch
-        permission_mode: edit
-        rules:
-          - condition: 監督者の指摘に対する修正が完了した
-          - condition: 修正を進行できない
-        instruction: fix-supervisor
-    rules:
-      - condition: all("AI問題の修正完了", "監督者の指摘に対する修正が完了した")
-        next: reviewers
-      - condition: any("修正不要（指摘対象ファイル/仕様の確認済み）", "判断できない、情報不足", "修正を進行できない")
-        next: implement
-  - name: ai_fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: AI問題の修正完了
-        next: reviewers
-      - condition: 修正不要（指摘対象ファイル/仕様の確認済み）
-        next: implement
-      - condition: 判断できない、情報不足
-        next: implement
-    instruction: ai-fix
-  - name: supervise_fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: 監督者の指摘に対する修正が完了した
-        next: reviewers
-      - condition: 修正を進行できない
-        next: implement
-    instruction: fix-supervisor
-policies:
-  coding: ../policies/coding.md
-  review: ../policies/review.md
-  testing: ../policies/testing.md
-  ai-antipattern: ../policies/ai-antipattern.md

builtins/ja/pieces/passthrough-hybrid-codex.yaml

@@ -1,44 +0,0 @@
-# Auto-generated from passthrough.yaml by tools/generate-hybrid-codex.mjs
-# Do not edit manually. Edit the source piece and re-run the generator.
-
-name: passthrough-hybrid-codex
-description: Single-agent thin wrapper. Pass task directly to coder as-is.
-max_iterations: 10
-personas:
-  coder: ../personas/coder.md
-initial_movement: execute
-movements:
-  - name: execute
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: タスク完了
-        next: COMPLETE
-      - condition: 進行できない
-        next: ABORT
-      - condition: ユーザー入力が必要
-        next: execute
-        requires_user_input: true
-        interactive_only: true
-    instruction_template: |
-      タスクをこなしてください。
-    output_contracts:
-      report:
-        - Summary: summary.md
-policies:
-  coding: ../policies/coding.md
-  testing: ../policies/testing.md

builtins/ja/pieces/review-fix-minimal-hybrid-codex.yaml

@@ -1,202 +0,0 @@
-# Auto-generated from review-fix-minimal.yaml by tools/generate-hybrid-codex.mjs
-# Do not edit manually. Edit the source piece and re-run the generator.
-
-name: review-fix-minimal-hybrid-codex
-description: 既存コードのレビューと修正ピース（レビュー開始、実装なし）
-max_iterations: 20
-personas:
-  coder: ../personas/coder.md
-  ai-antipattern-reviewer: ../personas/ai-antipattern-reviewer.md
-  supervisor: ../personas/supervisor.md
-instructions:
-  implement: ../instructions/implement.md
-  review-ai: ../instructions/review-ai.md
-  ai-fix: ../instructions/ai-fix.md
-  supervise: ../instructions/supervise.md
-  fix-supervisor: ../instructions/fix-supervisor.md
-report_formats:
-  ai-review: ../output-contracts/ai-review.md
-initial_movement: reviewers
-movements:
-  - name: implement
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    instruction: implement
-    rules:
-      - condition: 実装が完了した
-        next: reviewers
-      - condition: 実装を進行できない
-        next: ABORT
-      - condition: ユーザーへの確認事項があるためユーザー入力が必要
-        next: implement
-        requires_user_input: true
-        interactive_only: true
-    output_contracts:
-      report:
-        - Scope: 01-coder-scope.md
-        - Decisions: 02-coder-decisions.md
-  - name: reviewers
-    parallel:
-      - name: ai_review
-        edit: false
-        persona: ai-antipattern-reviewer
-        policy:
-          - review
-          - ai-antipattern
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        instruction: review-ai
-        rules:
-          - condition: AI特有の問題なし
-          - condition: AI特有の問題あり
-        output_contracts:
-          report:
-            - name: 03-ai-review.md
-              format: ai-review
-      - name: supervise
-        edit: false
-        persona: supervisor
-        policy: review
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - Bash
-          - WebSearch
-          - WebFetch
-        instruction: supervise
-        rules:
-          - condition: すべて問題なし
-          - condition: 要求未達成、テスト失敗、ビルドエラー
-        output_contracts:
-          report:
-            - Validation: 05-supervisor-validation.md
-            - Summary: summary.md
-    rules:
-      - condition: all("AI特有の問題なし", "すべて問題なし")
-        next: COMPLETE
-      - condition: all("AI特有の問題あり", "要求未達成、テスト失敗、ビルドエラー")
-        next: fix_both
-      - condition: any("AI特有の問題あり")
-        next: ai_fix
-      - condition: any("要求未達成、テスト失敗、ビルドエラー")
-        next: supervise_fix
-  - name: fix_both
-    parallel:
-      - name: ai_fix_parallel
-        edit: true
-        persona: coder
-        provider: codex
-        policy:
-          - coding
-          - testing
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - Edit
-          - Bash
-          - WebSearch
-          - WebFetch
-        permission_mode: edit
-        rules:
-          - condition: AI問題の修正完了
-          - condition: 修正不要（指摘対象ファイル/仕様の確認済み）
-          - condition: 判断できない、情報不足
-        instruction: ai-fix
-      - name: supervise_fix_parallel
-        edit: true
-        persona: coder
-        provider: codex
-        policy:
-          - coding
-          - testing
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - Edit
-          - Bash
-          - WebSearch
-          - WebFetch
-        permission_mode: edit
-        rules:
-          - condition: 監督者の指摘に対する修正が完了した
-          - condition: 修正を進行できない
-        instruction: fix-supervisor
-    rules:
-      - condition: all("AI問題の修正完了", "監督者の指摘に対する修正が完了した")
-        next: reviewers
-      - condition: any("修正不要（指摘対象ファイル/仕様の確認済み）", "判断できない、情報不足", "修正を進行できない")
-        next: implement
-  - name: ai_fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: AI問題の修正完了
-        next: reviewers
-      - condition: 修正不要（指摘対象ファイル/仕様の確認済み）
-        next: implement
-      - condition: 判断できない、情報不足
-        next: implement
-    instruction: ai-fix
-  - name: supervise_fix
-    edit: true
-    persona: coder
-    provider: codex
-    policy:
-      - coding
-      - testing
-    allowed_tools:
-      - Read
-      - Glob
-      - Grep
-      - Edit
-      - Write
-      - Bash
-      - WebSearch
-      - WebFetch
-    permission_mode: edit
-    rules:
-      - condition: 監督者の指摘に対する修正が完了した
-        next: reviewers
-      - condition: 修正を進行できない
-        next: implement
-    instruction: fix-supervisor
-policies:
-  coding: ../policies/coding.md
-  review: ../policies/review.md
-  testing: ../policies/testing.md
-  ai-antipattern: ../policies/ai-antipattern.md

builtins/ja/pieces/structural-reform.yaml

@@ -0,0 +1,455 @@
+name: structural-reform
+description: プロジェクト全体レビューと構造改革 - 段階的なファイル分割による反復的コードベース再構築
+max_iterations: 50
+policies:
+  coding: ../policies/coding.md
+  review: ../policies/review.md
+  testing: ../policies/testing.md
+  qa: ../policies/qa.md
+knowledge:
+  backend: ../knowledge/backend.md
+  architecture: ../knowledge/architecture.md
+personas:
+  planner: ../personas/planner.md
+  coder: ../personas/coder.md
+  architecture-reviewer: ../personas/architecture-reviewer.md
+  qa-reviewer: ../personas/qa-reviewer.md
+  supervisor: ../personas/supervisor.md
+instructions:
+  implement: ../instructions/implement.md
+  review-arch: ../instructions/review-arch.md
+  review-qa: ../instructions/review-qa.md
+  fix: ../instructions/fix.md
+initial_movement: review
+loop_monitors:
+  - cycle:
+      - implement
+      - fix
+    threshold: 3
+    judge:
+      persona: supervisor
+      instruction_template: |
+        implement → reviewers → fix のループが現在の改革ターゲットに対して {cycle_count} 回繰り返されました。
+
+        各サイクルのレポートを確認し、このループが進捗しているか、
+        同じ問題を繰り返しているかを判断してください。
+
+        **参照するレポート:**
+        - アーキテクチャレビュー: {report:04-architect-review.md}
+        - QAレビュー: {report:05-qa-review.md}
+
+        **判断基準:**
+        - 各修正サイクルでレビュー指摘が解消されているか
+        - 同じ問題が解決されずに繰り返されていないか
+        - 実装が承認に向かって収束しているか
+      rules:
+        - condition: 健全（承認に向けて進捗あり）
+          next: implement
+        - condition: 非生産的（同じ問題が繰り返され、収束なし）
+          next: next_target
+movements:
+  - name: review
+    edit: false
+    persona: architecture-reviewer
+    policy: review
+    knowledge:
+      - architecture
+      - backend
+    allowed_tools:
+      - Read
+      - Glob
+      - Grep
+      - WebSearch
+      - WebFetch
+    instruction_template: |
+      ## ピースステータス
+      - イテレーション: {iteration}/{max_iterations}（ピース全体）
+      - ムーブメントイテレーション: {movement_iteration}（このムーブメントの実行回数）
+      - ムーブメント: review（プロジェクト全体レビュー）
+
+      ## ユーザーリクエスト
+      {task}
+
+      ## 指示
+      プロジェクトのコードベース全体に対して、包括的な構造レビューを実施してください。
+
+      **重点領域:**
+      1. **God Class/Function**: 300行を超えるファイル、複数の責務を持つクラス
+      2. **結合度**: 循環依存、モジュール間の密結合
+      3. **凝集度**: 無関係な関心事が混在した低凝集モジュール
+      4. **テスタビリティ**: 密結合や副作用によるテスト困難なコード
+      5. **レイヤー違反**: 依存方向の誤り、アダプター層にドメインロジック
+      6. **DRY違反**: 3箇所以上の重複ロジック
+
+      **発見した問題ごとに報告:**
+      - ファイルパスと行数
+      - 問題カテゴリ（God Class、低凝集など）
+      - 深刻度（Critical / High / Medium）
+      - 分離すべき具体的な責務
+      - 分割により影響を受ける依存先
+
+      **出力フォーマット:**
+
+      ```markdown
+      # プロジェクト全体構造レビュー
+
+      ## サマリー
+      - レビュー対象ファイル数: N
+      - 検出された問題: N（Critical: N, High: N, Medium: N）
+
+      ## Critical Issues
+
+      ### 1. {ファイルパス}（{行数}行）
+      - **問題**: {カテゴリ}
+      - **深刻度**: Critical
+      - **検出された責務**:
+        1. {責務1}
+        2. {責務2}
+      - **分割提案**:
+        - `{新ファイル1}.ts`: {責務}
+        - `{新ファイル2}.ts`: {責務}
+      - **影響を受ける依存先**: {このモジュールをインポートしているファイル}
+
+      ## High Priority Issues
+      ...
+
+      ## Medium Priority Issues
+      ...
+
+      ## 依存グラフの懸念事項
+      - {循環依存、レイヤー違反}
+
+      ## 推奨改革順序
+      1. {ファイル} - {優先理由}
+      2. {ファイル} - {優先理由}
+      ```
+    rules:
+      - condition: 全体レビューが完了し、問題が検出された
+        next: plan_reform
+      - condition: 構造的な問題は見つからなかった
+        next: COMPLETE
+    output_contracts:
+      report:
+        - name: 00-full-review.md
+
+  - name: plan_reform
+    edit: false
+    persona: planner
+    knowledge: architecture
+    allowed_tools:
+      - Read
+      - Glob
+      - Grep
+      - Bash
+      - WebSearch
+      - WebFetch
+    instruction_template: |
+      ## ピースステータス
+      - イテレーション: {iteration}/{max_iterations}（ピース全体）
+      - ムーブメントイテレーション: {movement_iteration}（このムーブメントの実行回数）
+      - ムーブメント: plan_reform（改革計画策定）
+
+      ## ユーザーリクエスト
+      {task}
+
+      ## 全体レビュー結果
+      {previous_response}
+
+      ## ユーザー追加入力
+      {user_inputs}
+
+      ## 指示
+      全体レビュー結果を基に、具体的な改革実行計画を策定してください。
+
+      **計画の原則:**
+      - 1イテレーションにつき1ファイルの分割（変更を管理可能に保つ）
+      - 依存順に実行：まずリーフノードを分割し、内側に向かって進む
+      - 各分割後にテストとビルドが通ること
+      - 後方互換は不要（ユーザー指示通り）
+
+      **各改革ターゲットについて指定:**
+      1. 対象ファイルと現在の行数
+      2. 提案する新ファイルと責務
+      3. 依存先ファイルのインポート変更予定
+      4. テスト戦略（新規テスト、既存テストの更新）
+      5. リスク評価（何が壊れる可能性があるか）
+
+      **出力フォーマット:**
+
+      ```markdown
+      # 構造改革計画
+
+      ## 改革ターゲット（実行優先順）
+
+      ### ターゲット1: {ファイルパス}
+      - **現状**: {行数}行、{N}個の責務
+      - **分割提案**:
+        | 新ファイル | 責務 | 推定行数 |
+        |----------|------|---------|
+        | `{パス}` | {責務} | ~{N} |
+      - **依存先ファイル**: {このモジュールをインポートしているファイル一覧}
+      - **テスト計画**: {追加・更新すべきテスト}
+      - **リスク**: {Low/Medium/High} - {説明}
+
+      ### ターゲット2: {ファイルパス}
+      ...
+
+      ## 実行順序の根拠
+      {この順序がリスクと依存関係の競合を最小化する理由}
+
+      ## 成功基準
+      - 各分割後に全テストが通る
+      - 各分割後にビルドが成功する
+      - 300行を超えるファイルがない
+      - 各ファイルが単一責務を持つ
+      ```
+    rules:
+      - condition: 改革計画が完成し、実行可能な状態
+        next: implement
+      - condition: 実行可能な改革が特定されなかった
+        next: COMPLETE
+      - condition: 要件が不明確、ユーザー入力が必要
+        next: ABORT
+        appendix: |
+          確認事項:
+          - {質問1}
+          - {質問2}
+    output_contracts:
+      report:
+        - name: 01-reform-plan.md
+          format: plan
+
+  - name: implement
+    edit: true
+    persona: coder
+    policy:
+      - coding
+      - testing
+    session: refresh
+    knowledge:
+      - backend
+      - architecture
+    allowed_tools:
+      - Read
+      - Glob
+      - Grep
+      - Edit
+      - Write
+      - Bash
+      - WebSearch
+      - WebFetch
+    permission_mode: edit
+    instruction: implement
+    rules:
+      - condition: 実装完了
+        next: reviewers
+      - condition: 判断できない、情報不足
+        next: reviewers
+      - condition: ユーザー入力が必要
+        next: implement
+        requires_user_input: true
+        interactive_only: true
+    output_contracts:
+      report:
+        - Scope: 02-coder-scope.md
+        - Decisions: 03-coder-decisions.md
+
+  - name: reviewers
+    parallel:
+      - name: arch-review
+        edit: false
+        persona: architecture-reviewer
+        policy: review
+        knowledge:
+          - architecture
+          - backend
+        allowed_tools:
+          - Read
+          - Glob
+          - Grep
+          - WebSearch
+          - WebFetch
+        rules:
+          - condition: approved
+          - condition: needs_fix
+        instruction: review-arch
+        output_contracts:
+          report:
+            - name: 04-architect-review.md
+              format: architecture-review
+      - name: qa-review
+        edit: false
+        persona: qa-reviewer
+        policy:
+          - review
+          - qa
+        allowed_tools:
+          - Read
+          - Glob
+          - Grep
+          - WebSearch
+          - WebFetch
+        rules:
+          - condition: approved
+          - condition: needs_fix
+        instruction: review-qa
+        output_contracts:
+          report:
+            - name: 05-qa-review.md
+              format: qa-review
+    rules:
+      - condition: all("approved")
+        next: verify
+      - condition: any("needs_fix")
+        next: fix
+
+  - name: fix
+    edit: true
+    persona: coder
+    policy:
+      - coding
+      - testing
+    knowledge:
+      - backend
+      - architecture
+    allowed_tools:
+      - Read
+      - Glob
+      - Grep
+      - Edit
+      - Write
+      - Bash
+      - WebSearch
+      - WebFetch
+    permission_mode: edit
+    rules:
+      - condition: 修正完了
+        next: reviewers
+      - condition: 判断できない、情報不足
+        next: plan_reform
+    instruction: fix
+
+  - name: verify
+    edit: false
+    persona: supervisor
+    policy: review
+    allowed_tools:
+      - Read
+      - Glob
+      - Grep
+      - Bash
+      - WebSearch
+      - WebFetch
+    instruction_template: |
+      ## ピースステータス
+      - イテレーション: {iteration}/{max_iterations}（ピース全体）
+      - ムーブメントイテレーション: {movement_iteration}（このムーブメントの実行回数）
+      - ムーブメント: verify（ビルド・テスト検証）
+
+      ## 指示
+      現在の改革ステップが正常に完了したことを検証してください。
+
+      **検証チェックリスト:**
+      1. **ビルド**: ビルドコマンドを実行し、成功することを確認
+      2. **テスト**: テストスイートを実行し、全テストが通ることを確認
+      3. **ファイルサイズ**: 新しいファイルが300行を超えていないことを確認
+      4. **単一責務**: 各新ファイルが明確な単一の目的を持つことを確認
+      5. **インポート整合性**: すべてのインポートが正しく更新されていることを確認
+
+      **レポートフォーマット:**
+
+      ```markdown
+      # 検証結果
+
+      ## 結果: PASS / FAIL
+
+      | チェック項目 | 状態 | 詳細 |
+      |------------|------|------|
+      | ビルド | PASS/FAIL | {出力サマリー} |
+      | テスト | PASS/FAIL | {N passed, N failed} |
+      | ファイルサイズ | PASS/FAIL | {300行超のファイルの有無} |
+      | 単一責務 | PASS/FAIL | {評価} |
+      | インポート整合性 | PASS/FAIL | {壊れたインポートの有無} |
+
+      ## 問題点（FAILの場合）
+      1. {問題の説明}
+      ```
+    rules:
+      - condition: すべての検証に合格
+        next: next_target
+      - condition: 検証に失敗
+        next: fix
+    output_contracts:
+      report:
+        - name: 06-verification.md
+          format: validation
+
+  - name: next_target
+    edit: false
+    persona: planner
+    knowledge: architecture
+    allowed_tools:
+      - Read
+      - Glob
+      - Grep
+      - Bash
+      - WebSearch
+      - WebFetch
+    instruction_template: |
+      ## ピースステータス
+      - イテレーション: {iteration}/{max_iterations}（ピース全体）
+      - ムーブメントイテレーション: {movement_iteration}（このムーブメントの実行回数）
+      - ムーブメント: next_target（進捗確認と次ターゲット選択）
+
+      ## 改革計画
+      {report:01-reform-plan.md}
+
+      ## 最新の検証結果
+      {previous_response}
+
+      ## 指示
+      構造改革の進捗を評価し、次のアクションを決定してください。
+
+      **手順:**
+      1. 改革計画を確認し、完了したターゲットを特定
+      2. 現在のコードベースの状態を計画と照合
+      3. 残りの改革ターゲットがあるか判断
+
+      **出力フォーマット:**
+
+      ```markdown
+      # 改革進捗
+
+      ## 完了ターゲット
+      | # | ターゲット | 状態 |
+      |---|----------|------|
+      | 1 | {ファイル} | 完了 |
+      | 2 | {ファイル} | 完了 |
+
+      ## 残りターゲット
+      | # | ターゲット | 優先度 |
+      |---|----------|-------|
+      | 3 | {ファイル} | 次 |
+      | 4 | {ファイル} | 保留 |
+
+      ## 次のアクション
+      - **ターゲット**: {次に改革するファイル}
+      - **計画**: {分割の概要}
+
+      ## 全体進捗
+      {N}/{合計}ターゲット完了。残りの推定イテレーション数: {N}
+      ```
+    rules:
+      - condition: まだ改革ターゲットが残っている
+        next: implement
+      - condition: すべての改革ターゲットが完了
+        next: COMPLETE
+    output_contracts:
+      report:
+        - name: 07-progress.md
+report_formats:
+  plan: ../output-contracts/plan.md
+  architecture-review: ../output-contracts/architecture-review.md
+  qa-review: ../output-contracts/qa-review.md
+  validation: ../output-contracts/validation.md
+  summary: ../output-contracts/summary.md

builtins/ja/pieces/unit-test.yaml

@@ -1,37 +1,32 @@
-# Auto-generated from default.yaml by tools/generate-hybrid-codex.mjs
-# Do not edit manually. Edit the source piece and re-run the generator.
-
-name: default-hybrid-codex
-description: Standard development piece with planning and specialized reviews
-max_iterations: 30
+name: unit-test
+description: 単体テスト追加に特化したピース（テスト分析→テスト実装→レビュー→修正）
+max_iterations: 20
+policies:
+  coding: ../policies/coding.md
+  review: ../policies/review.md
+  testing: ../policies/testing.md
+  ai-antipattern: ../policies/ai-antipattern.md
+  qa: ../policies/qa.md
 knowledge:
   architecture: ../knowledge/architecture.md
   backend: ../knowledge/backend.md
 personas:
-  planner: ../personas/planner.md
+  test-planner: ../personas/test-planner.md
   coder: ../personas/coder.md
   ai-antipattern-reviewer: ../personas/ai-antipattern-reviewer.md
   architecture-reviewer: ../personas/architecture-reviewer.md
   qa-reviewer: ../personas/qa-reviewer.md
   supervisor: ../personas/supervisor.md
 instructions:
-  plan: ../instructions/plan.md
-  implement: ../instructions/implement.md
+  plan-test: ../instructions/plan-test.md
+  implement-test: ../instructions/implement-test.md
   ai-review: ../instructions/ai-review.md
   ai-fix: ../instructions/ai-fix.md
   arbitrate: ../instructions/arbitrate.md
-  review-arch: ../instructions/review-arch.md
-  review-qa: ../instructions/review-qa.md
+  review-test: ../instructions/review-test.md
   fix: ../instructions/fix.md
   supervise: ../instructions/supervise.md
-report_formats:
-  plan: ../output-contracts/plan.md
-  ai-review: ../output-contracts/ai-review.md
-  architecture-review: ../output-contracts/architecture-review.md
-  qa-review: ../output-contracts/qa-review.md
-  validation: ../output-contracts/validation.md
-  summary: ../output-contracts/summary.md
-initial_movement: plan
+initial_movement: plan_test
 loop_monitors:
   - cycle:
       - ai_review
@@ -56,12 +51,15 @@ loop_monitors:
         - condition: 健全（進捗あり）
           next: ai_review
         - condition: 非生産的（改善なし）
-          next: reviewers
+          next: review_test
 movements:
-  - name: plan
+  - name: plan_test
     edit: false
-    persona: planner
-    knowledge: architecture
+    persona: test-planner
+    policy: testing
+    knowledge:
+      - architecture
+      - backend
     allowed_tools:
       - Read
       - Glob
@@ -70,9 +68,9 @@ movements:
       - WebSearch
       - WebFetch
     rules:
-      - condition: 要件が明確で実装可能
-        next: implement
-      - condition: ユーザーが質問をしている（実装タスクではない）
+      - condition: テスト計画が完了
+        next: implement_test
+      - condition: ユーザーが質問をしている（テスト追加タスクではない）
         next: COMPLETE
       - condition: 要件が不明確、情報不足
         next: ABORT
@@ -80,15 +78,15 @@ movements:
           確認事項:
           - {質問1}
           - {質問2}
-    instruction: plan
+    instruction: plan-test
     output_contracts:
       report:
-        - name: 00-plan.md
-          format: plan
-  - name: implement
+        - name: 00-test-plan.md
+          format: test-plan
+
+  - name: implement_test
     edit: true
     persona: coder
-    provider: codex
     policy:
       - coding
       - testing
@@ -107,21 +105,22 @@ movements:
       - WebFetch
     permission_mode: edit
     rules:
-      - condition: 実装完了
+      - condition: テスト実装完了
         next: ai_review
       - condition: 実装未着手（レポートのみ）
         next: ai_review
       - condition: 判断できない、情報不足
         next: ai_review
       - condition: ユーザー入力が必要
-        next: implement
+        next: implement_test
         requires_user_input: true
         interactive_only: true
-    instruction: implement
+    instruction: implement-test
     output_contracts:
       report:
         - Scope: 02-coder-scope.md
         - Decisions: 03-coder-decisions.md
+
   - name: ai_review
     edit: false
     persona: ai-antipattern-reviewer
@@ -136,7 +135,7 @@ movements:
       - WebFetch
     rules:
       - condition: AI特有の問題なし
-        next: reviewers
+        next: review_test
       - condition: AI特有の問題あり
         next: ai_fix
     instruction: ai-review
@@ -144,10 +143,10 @@ movements:
       report:
         - name: 04-ai-review.md
           format: ai-review
+
   - name: ai_fix
     edit: true
     persona: coder
-    provider: codex
     policy:
       - coding
       - testing
@@ -173,6 +172,7 @@ movements:
       - condition: 判断できない、情報不足
         next: ai_no_fix
     instruction: ai-fix
+
   - name: ai_no_fix
     edit: false
     persona: architecture-reviewer
@@ -185,63 +185,39 @@ movements:
       - condition: ai_reviewの指摘が妥当（修正すべき）
         next: ai_fix
       - condition: ai_fixの判断が妥当（修正不要）
-        next: reviewers
+        next: review_test
     instruction: arbitrate
-  - name: reviewers
-    parallel:
-      - name: arch-review
-        edit: false
-        persona: architecture-reviewer
-        policy: review
-        knowledge:
-          - architecture
-          - backend
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-arch
-        output_contracts:
-          report:
-            - name: 05-architect-review.md
-              format: architecture-review
-      - name: qa-review
-        edit: false
-        persona: qa-reviewer
-        policy:
-          - review
-          - qa
-        allowed_tools:
-          - Read
-          - Glob
-          - Grep
-          - WebSearch
-          - WebFetch
-        rules:
-          - condition: approved
-          - condition: needs_fix
-        instruction: review-qa
-        output_contracts:
-          report:
-            - name: 06-qa-review.md
-              format: qa-review
+
+  - name: review_test
+    edit: false
+    persona: qa-reviewer
+    policy:
+      - review
+      - qa
+    allowed_tools:
+      - Read
+      - Glob
+      - Grep
+      - WebSearch
+      - WebFetch
     rules:
-      - condition: all("approved")
+      - condition: approved
         next: supervise
-      - condition: any("needs_fix")
+      - condition: needs_fix
         next: fix
+    instruction: review-test
+    output_contracts:
+      report:
+        - name: 05-qa-review.md
+          format: qa-review
+
   - name: fix
     edit: true
     persona: coder
-    provider: codex
     policy:
       - coding
       - testing
+    session: refresh
     knowledge:
       - backend
       - architecture
@@ -257,10 +233,11 @@ movements:
     permission_mode: edit
     rules:
       - condition: 修正完了
-        next: reviewers
+        next: review_test
       - condition: 判断できない、情報不足
-        next: plan
+        next: plan_test
     instruction: fix
+
   - name: supervise
     edit: false
     persona: supervisor
@@ -276,15 +253,15 @@ movements:
       - condition: すべて問題なし
         next: COMPLETE
       - condition: 要求未達成、テスト失敗、ビルドエラー
-        next: plan
+        next: plan_test
     instruction: supervise
     output_contracts:
       report:
-        - Validation: 07-supervisor-validation.md
+        - Validation: 06-supervisor-validation.md
         - Summary: summary.md
-policies:
-  coding: ../policies/coding.md
-  review: ../policies/review.md
-  testing: ../policies/testing.md
-  ai-antipattern: ../policies/ai-antipattern.md
-  qa: ../policies/qa.md
+report_formats:
+  test-plan: ../output-contracts/test-plan.md
+  ai-review: ../output-contracts/ai-review.md
+  qa-review: ../output-contracts/qa-review.md
+  validation: ../output-contracts/validation.md
+  summary: ../output-contracts/summary.md

builtins/ja/policies/review.md

@@ -86,6 +86,18 @@
      共通関数に抽出してください」
 ```
 
+## 指摘ID管理（finding_id）
+
+同じ指摘の堂々巡りを防ぐため、指摘をIDで追跡する。
+
+- REJECT時に挙げる各問題には `finding_id` を必須で付ける
+- 同じ問題を再指摘する場合は、同じ `finding_id` を再利用する
+- 再指摘時は状態を `persists` とし、未解決である根拠（ファイル/行）を必ず示す
+- 新規指摘は状態 `new` とする
+- 解消済みは状態 `resolved` として一覧化する
+- `finding_id` のない指摘は無効（判定根拠として扱わない）
+- REJECTは `new` または `persists` の問題が1件以上ある場合のみ許可する
+
 ## ボーイスカウトルール
 
 来たときよりも美しく。

docs/README.ja.md

@@ -76,6 +76,10 @@ takt --pipeline --task "バグを修正して" --auto-pr
 
 ## 使い方
 
+## 実装メモ
+
+- failed タスクの retry とセッション再開: [`docs/implements/retry-and-session.ja.md`](./implements/retry-and-session.ja.md)
+
 ### 対話モード
 
 AI との会話でタスク内容を詰めてから実行するモード。タスクの要件が曖昧な場合や、AI と相談しながら内容を整理したい場合に便利です。
@@ -88,13 +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
 
@@ -447,8 +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。
 
@@ -471,6 +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
@@ -539,8 +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

docs/implements/retry-and-session.ja.md

@@ -0,0 +1,58 @@
+# failed -> retry とセッション再開の動作原理
+
+このドキュメントは、`takt list` の `failed` タスクを `Retry` したときの挙動と、AI セッション再開の実際の仕組みを整理した実装メモです。
+
+## 問題の要約
+
+- `failed -> retry` で「前回のセッションがそのまま復活するか」が分かりづらい。
+- 実際には「タスク固有のセッション再開」ではなく「保存済みの persona セッション利用」で再開される。
+
+## 解決方針（実装の見方）
+
+- `Retry` が何を再投入するかを確認する。
+- 再実行時にどのセッションIDを読み込むかを確認する。
+- `session: refresh` や worktree 条件など、再開しないケースを明記する。
+
+## 実装の流れ
+
+1. `takt list` で `failed` タスクに `Retry` を選ぶ。
+2. `retryFailedTask()` が `TaskRunner.requeueFailedTask()` を呼ぶ。
+3. 失敗タスク配下の元タスクファイルを `.takt/tasks/` にコピーする。
+4. YAML タスクの場合のみ、必要に応じて `start_movement` と `retry_note` を書き換える。
+
+参照:
+- `src/features/tasks/list/taskRetryActions.ts`
+- `src/infra/task/runner.ts`
+
+## 重要: Retry で復元される情報
+
+`requeueFailedTask()` はタスクファイルに `sessionId` を書き戻さない。  
+復元対象は `start_movement` と `retry_note` のみ。
+
+つまり、`failed` タスクを `Retry` しても「その failed ディレクトリに紐づくセッションID」を直接復元する実装にはなっていない。
+
+## セッション再開の実際
+
+再実行時の `pieceExecution` で、プロジェクト保存済みセッションをロードして `initialSessions` に渡す。
+
+- 通常実行: `loadPersonaSessions(projectCwd, provider)`
+- worktree 実行: `loadWorktreeSessions(projectCwd, cwd, provider)`
+
+その後、各 movement の Phase 1 実行で `OptionsBuilder.buildAgentOptions()` が `sessionId` を組み立てる。
+
+参照:
+- `src/features/tasks/execute/pieceExecution.ts`
+- `src/core/piece/engine/OptionsBuilder.ts`
+
+## セッション再開しない条件
+
+- movement 側が `session: refresh` の場合。
+- `cwd !== projectCwd` など、セッション再開を抑止する条件に当たる場合。
+
+このため、`Retry` で常に同じ会話が厳密再開されるわけではない。
+
+## 運用上の注意
+
+- `failed -> retry` は正しい復旧手順。
+- ただしセッションは「タスク単位」ではなく「persona の保存状態」基準で再開される。
+- 前回文脈を確実に残したい場合は、`retry_note` に再試行理由と前提を明記する。

e2e/helpers/test-repo.ts

@@ -11,6 +11,11 @@ export interface TestRepo {
   cleanup: () => void;
 }
 
+export interface CreateTestRepoOptions {
+  /** Skip creating a test branch (stay on default branch). Use for pipeline tests. */
+  skipBranch?: boolean;
+}
+
 function getGitHubUser(): string {
   const user = execFileSync('gh', ['api', 'user', '--jq', '.login'], {
     encoding: 'utf-8',
@@ -33,7 +38,7 @@ function getGitHubUser(): string {
  *   2. Close any PRs created during the test
  *   3. Delete local directory
  */
-export function createTestRepo(): TestRepo {
+export function createTestRepo(options?: CreateTestRepoOptions): TestRepo {
   const user = getGitHubUser();
   const repoName = `${user}/takt-testing`;
 
@@ -56,49 +61,80 @@ export function createTestRepo(): TestRepo {
     stdio: 'pipe',
   });
 
-  // Create test branch
-  const testBranch = `e2e-test-${Date.now()}`;
-  execFileSync('git', ['checkout', '-b', testBranch], {
-    cwd: repoPath,
-    stdio: 'pipe',
-  });
+  // Create test branch (unless skipped for pipeline tests)
+  const testBranch = options?.skipBranch
+    ? undefined
+    : `e2e-test-${Date.now()}`;
+  if (testBranch) {
+    execFileSync('git', ['checkout', '-b', testBranch], {
+      cwd: repoPath,
+      stdio: 'pipe',
+    });
+  }
+
+  const currentBranch = testBranch
+    ?? execFileSync('git', ['branch', '--show-current'], {
+      cwd: repoPath,
+      encoding: 'utf-8',
+    }).trim();
 
   return {
     path: repoPath,
     repoName,
-    branch: testBranch,
+    branch: currentBranch,
     cleanup: () => {
-      // 1. Delete remote branch (best-effort)
-      try {
-        execFileSync(
-          'git',
-          ['push', 'origin', '--delete', testBranch],
-          { cwd: repoPath, stdio: 'pipe' },
-        );
-      } catch {
-        // Branch may not have been pushed; ignore
-      }
-
-      // 2. Close any PRs from this branch (best-effort)
-      try {
-        const prList = execFileSync(
-          'gh',
-          ['pr', 'list', '--head', testBranch, '--repo', repoName, '--json', 'number', '--jq', '.[].number'],
-          { encoding: 'utf-8', stdio: 'pipe' },
-        ).trim();
-
-        for (const prNumber of prList.split('\n').filter(Boolean)) {
+      if (testBranch) {
+        // 1. Delete remote branch (best-effort)
+        try {
           execFileSync(
-            'gh',
-            ['pr', 'close', prNumber, '--repo', repoName, '--delete-branch'],
-            { stdio: 'pipe' },
+            'git',
+            ['push', 'origin', '--delete', testBranch],
+            { cwd: repoPath, stdio: 'pipe' },
           );
+        } catch {
+          // Branch may not have been pushed; ignore
+        }
+
+        // 2. Close any PRs from this branch (best-effort)
+        try {
+          const prList = execFileSync(
+            'gh',
+            ['pr', 'list', '--head', testBranch, '--repo', repoName, '--json', 'number', '--jq', '.[].number'],
+            { encoding: 'utf-8', stdio: 'pipe' },
+          ).trim();
+
+          for (const prNumber of prList.split('\n').filter(Boolean)) {
+            execFileSync(
+              'gh',
+              ['pr', 'close', prNumber, '--repo', repoName, '--delete-branch'],
+              { stdio: 'pipe' },
+            );
+          }
+        } catch {
+          // No PRs or already closed; ignore
+        }
+      } else {
+        // Pipeline mode: clean up takt-created PRs (best-effort)
+        try {
+          const prNumbers = execFileSync(
+            'gh',
+            ['pr', 'list', '--state', 'open', '--repo', repoName, '--json', 'number', '--jq', '.[].number'],
+            { encoding: 'utf-8', stdio: 'pipe' },
+          ).trim();
+
+          for (const prNumber of prNumbers.split('\n').filter(Boolean)) {
+            execFileSync(
+              'gh',
+              ['pr', 'close', prNumber, '--repo', repoName, '--delete-branch'],
+              { stdio: 'pipe' },
+            );
+          }
+        } catch {
+          // ignore
         }
-      } catch {
-        // No PRs or already closed; ignore
       }
 
-      // 3. Delete local directory last
+      // Delete local directory last
       try {
         rmSync(repoPath, { recursive: true, force: true });
       } catch {

e2e/specs/github-issue.e2e.ts

@@ -17,7 +17,7 @@ describe('E2E: GitHub Issue processing', () => {
 
   beforeEach(() => {
     isolatedEnv = createIsolatedEnv();
-    testRepo = createTestRepo();
+    testRepo = createTestRepo({ skipBranch: true });
 
     // Create a test issue
     const createOutput = execFileSync(

e2e/specs/pipeline.e2e.ts

@@ -16,7 +16,7 @@ describe('E2E: Pipeline mode (--pipeline --auto-pr)', () => {
 
   beforeEach(() => {
     isolatedEnv = createIsolatedEnv();
-    testRepo = createTestRepo();
+    testRepo = createTestRepo({ skipBranch: true });
   });
 
   afterEach(() => {

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",

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",

src/__tests__/cli-routing-issue-resolve.test.ts

@@ -45,6 +45,11 @@ vi.mock('../features/pipeline/index.js', () => ({
 
 vi.mock('../features/interactive/index.js', () => ({
   interactiveMode: vi.fn(),
+  selectInteractiveMode: vi.fn(() => 'assistant'),
+  passthroughMode: vi.fn(),
+  quietMode: vi.fn(),
+  personaMode: vi.fn(),
+  resolveLanguage: vi.fn(() => 'en'),
 }));
 
 vi.mock('../infra/config/index.js', () => ({

src/__tests__/engine-parallel.test.ts

@@ -160,4 +160,71 @@ describe('PieceEngine Integration: Parallel Movement Aggregation', () => {
     expect(calledAgents).toContain('../personas/arch-review.md');
     expect(calledAgents).toContain('../personas/security-review.md');
   });
+
+  it('should output rich parallel prefix when taskPrefix/taskColorIndex are provided', async () => {
+    const config = buildDefaultPieceConfig();
+    const stdoutSpy = vi.spyOn(process.stdout, 'write').mockImplementation(() => true);
+    const parentOnStream = vi.fn();
+
+    const responsesByPersona = new Map<string, ReturnType<typeof makeResponse>>([
+      ['../personas/plan.md', makeResponse({ persona: 'plan', content: 'Plan done' })],
+      ['../personas/implement.md', makeResponse({ persona: 'implement', content: 'Impl done' })],
+      ['../personas/ai_review.md', makeResponse({ persona: 'ai_review', content: 'OK' })],
+      ['../personas/arch-review.md', makeResponse({ persona: 'arch-review', content: 'Architecture review content' })],
+      ['../personas/security-review.md', makeResponse({ persona: 'security-review', content: 'Security review content' })],
+      ['../personas/supervise.md', makeResponse({ persona: 'supervise', content: 'All passed' })],
+    ]);
+
+    vi.mocked(runAgent).mockImplementation(async (persona, _task, options) => {
+      const response = responsesByPersona.get(persona ?? '');
+      if (!response) {
+        throw new Error(`Unexpected persona: ${persona}`);
+      }
+
+      if (persona === '../personas/arch-review.md') {
+        options.onStream?.({ type: 'text', data: { text: 'arch stream line\n' } });
+      }
+      if (persona === '../personas/security-review.md') {
+        options.onStream?.({ type: 'text', data: { text: 'security stream line\n' } });
+      }
+
+      return response;
+    });
+
+    mockDetectMatchedRuleSequence([
+      { index: 0, method: 'phase1_tag' },
+      { index: 0, method: 'phase1_tag' },
+      { index: 0, method: 'phase1_tag' },
+      { index: 0, method: 'phase1_tag' },
+      { index: 0, method: 'phase1_tag' },
+      { index: 0, method: 'aggregate' },
+      { index: 0, method: 'phase1_tag' },
+    ]);
+
+    const engine = new PieceEngine(config, tmpDir, 'test task', {
+      projectCwd: tmpDir,
+      onStream: parentOnStream,
+      taskPrefix: 'override-persona-provider',
+      taskColorIndex: 0,
+    });
+
+    try {
+      const state = await engine.run();
+      expect(state.status).toBe('completed');
+
+      const output = stdoutSpy.mock.calls.map((call) => String(call[0])).join('');
+      expect(output).toContain('[over]');
+      expect(output).toContain('[reviewers][arch-review](4/30)(1) arch stream line');
+      expect(output).toContain('[reviewers][security-review](4/30)(1) security stream line');
+    } finally {
+      stdoutSpy.mockRestore();
+    }
+  });
+
+  it('should fail fast when taskPrefix is provided without taskColorIndex', () => {
+    const config = buildDefaultPieceConfig();
+    expect(
+      () => new PieceEngine(config, tmpDir, 'test task', { projectCwd: tmpDir, taskPrefix: 'override-persona-provider' })
+    ).toThrow('taskPrefix and taskColorIndex must be provided together');
+  });
 });

src/__tests__/engine-persona-providers.test.ts

@@ -0,0 +1,204 @@
+/**
+ * Tests for persona_providers config-level provider override.
+ *
+ * Verifies the provider resolution priority:
+ *   1. Movement YAML provider (highest)
+ *   2. persona_providers[personaDisplayName]
+ *   3. CLI/global provider (lowest)
+ */
+
+import { describe, it, expect, beforeEach, vi } from 'vitest';
+
+vi.mock('../agents/runner.js', () => ({
+  runAgent: vi.fn(),
+}));
+
+vi.mock('../core/piece/evaluation/index.js', () => ({
+  detectMatchedRule: vi.fn(),
+}));
+
+vi.mock('../core/piece/phase-runner.js', () => ({
+  needsStatusJudgmentPhase: vi.fn(),
+  runReportPhase: vi.fn(),
+  runStatusJudgmentPhase: vi.fn(),
+}));
+
+vi.mock('../shared/utils/index.js', async (importOriginal) => ({
+  ...(await importOriginal<Record<string, unknown>>()),
+  generateReportDir: vi.fn().mockReturnValue('test-report-dir'),
+}));
+
+import { PieceEngine } from '../core/piece/index.js';
+import { runAgent } from '../agents/runner.js';
+import type { PieceConfig } from '../core/models/index.js';
+import {
+  makeResponse,
+  makeRule,
+  makeMovement,
+  mockRunAgentSequence,
+  mockDetectMatchedRuleSequence,
+  applyDefaultMocks,
+} from './engine-test-helpers.js';
+
+describe('PieceEngine persona_providers override', () => {
+  beforeEach(() => {
+    vi.resetAllMocks();
+    applyDefaultMocks();
+  });
+
+  it('should use persona_providers when movement has no provider and persona matches', async () => {
+    const movement = makeMovement('implement', {
+      personaDisplayName: 'coder',
+      rules: [makeRule('done', 'COMPLETE')],
+    });
+    const config: PieceConfig = {
+      name: 'persona-provider-test',
+      movements: [movement],
+      initialMovement: 'implement',
+      maxIterations: 1,
+    };
+
+    mockRunAgentSequence([
+      makeResponse({ persona: movement.persona, content: 'done' }),
+    ]);
+    mockDetectMatchedRuleSequence([{ index: 0, method: 'phase1_tag' }]);
+
+    const engine = new PieceEngine(config, '/tmp/project', 'test task', {
+      projectCwd: '/tmp/project',
+      provider: 'claude',
+      personaProviders: { coder: 'codex' },
+    });
+
+    await engine.run();
+
+    const options = vi.mocked(runAgent).mock.calls[0][2];
+    expect(options.provider).toBe('codex');
+  });
+
+  it('should use global provider when persona is not in persona_providers', async () => {
+    const movement = makeMovement('plan', {
+      personaDisplayName: 'planner',
+      rules: [makeRule('done', 'COMPLETE')],
+    });
+    const config: PieceConfig = {
+      name: 'persona-provider-nomatch',
+      movements: [movement],
+      initialMovement: 'plan',
+      maxIterations: 1,
+    };
+
+    mockRunAgentSequence([
+      makeResponse({ persona: movement.persona, content: 'done' }),
+    ]);
+    mockDetectMatchedRuleSequence([{ index: 0, method: 'phase1_tag' }]);
+
+    const engine = new PieceEngine(config, '/tmp/project', 'test task', {
+      projectCwd: '/tmp/project',
+      provider: 'claude',
+      personaProviders: { coder: 'codex' },
+    });
+
+    await engine.run();
+
+    const options = vi.mocked(runAgent).mock.calls[0][2];
+    expect(options.provider).toBe('claude');
+  });
+
+  it('should prioritize movement provider over persona_providers', async () => {
+    const movement = makeMovement('implement', {
+      personaDisplayName: 'coder',
+      provider: 'claude',
+      rules: [makeRule('done', 'COMPLETE')],
+    });
+    const config: PieceConfig = {
+      name: 'movement-over-persona',
+      movements: [movement],
+      initialMovement: 'implement',
+      maxIterations: 1,
+    };
+
+    mockRunAgentSequence([
+      makeResponse({ persona: movement.persona, content: 'done' }),
+    ]);
+    mockDetectMatchedRuleSequence([{ index: 0, method: 'phase1_tag' }]);
+
+    const engine = new PieceEngine(config, '/tmp/project', 'test task', {
+      projectCwd: '/tmp/project',
+      provider: 'mock',
+      personaProviders: { coder: 'codex' },
+    });
+
+    await engine.run();
+
+    const options = vi.mocked(runAgent).mock.calls[0][2];
+    expect(options.provider).toBe('claude');
+  });
+
+  it('should work without persona_providers (undefined)', async () => {
+    const movement = makeMovement('plan', {
+      personaDisplayName: 'planner',
+      rules: [makeRule('done', 'COMPLETE')],
+    });
+    const config: PieceConfig = {
+      name: 'no-persona-providers',
+      movements: [movement],
+      initialMovement: 'plan',
+      maxIterations: 1,
+    };
+
+    mockRunAgentSequence([
+      makeResponse({ persona: movement.persona, content: 'done' }),
+    ]);
+    mockDetectMatchedRuleSequence([{ index: 0, method: 'phase1_tag' }]);
+
+    const engine = new PieceEngine(config, '/tmp/project', 'test task', {
+      projectCwd: '/tmp/project',
+      provider: 'claude',
+    });
+
+    await engine.run();
+
+    const options = vi.mocked(runAgent).mock.calls[0][2];
+    expect(options.provider).toBe('claude');
+  });
+
+  it('should apply different providers to different personas in a multi-movement piece', async () => {
+    const planMovement = makeMovement('plan', {
+      personaDisplayName: 'planner',
+      rules: [makeRule('done', 'implement')],
+    });
+    const implementMovement = makeMovement('implement', {
+      personaDisplayName: 'coder',
+      rules: [makeRule('done', 'COMPLETE')],
+    });
+    const config: PieceConfig = {
+      name: 'multi-persona-providers',
+      movements: [planMovement, implementMovement],
+      initialMovement: 'plan',
+      maxIterations: 3,
+    };
+
+    mockRunAgentSequence([
+      makeResponse({ persona: planMovement.persona, content: 'done' }),
+      makeResponse({ persona: implementMovement.persona, content: 'done' }),
+    ]);
+    mockDetectMatchedRuleSequence([
+      { index: 0, method: 'phase1_tag' },
+      { index: 0, method: 'phase1_tag' },
+    ]);
+
+    const engine = new PieceEngine(config, '/tmp/project', 'test task', {
+      projectCwd: '/tmp/project',
+      provider: 'claude',
+      personaProviders: { coder: 'codex' },
+    });
+
+    await engine.run();
+
+    const calls = vi.mocked(runAgent).mock.calls;
+    // Plan movement: planner not in persona_providers → claude
+    expect(calls[0][2].provider).toBe('claude');
+    // Implement movement: coder in persona_providers → codex
+    expect(calls[1][2].provider).toBe('codex');
+  });
+});

src/__tests__/facet-resolution.test.ts

@@ -158,6 +158,21 @@ describe('resolveRefToContent with layer resolution', () => {
     // No context, no file — returns the spec as-is (inline content behavior)
     expect(content).toBe('some-name');
   });
+
+  it('should fall back to resolveResourceContent when facet not found with context', () => {
+    // Given: facetType and context provided, but no matching facet file exists
+    // When: resolveRefToContent is called with a name that has no facet file
+    const content = resolveRefToContent(
+      'nonexistent-facet-xyz',
+      undefined,
+      tempDir,
+      'policies',
+      context,
+    );
+
+    // Then: falls back to resolveResourceContent, which returns the ref as inline content
+    expect(content).toBe('nonexistent-facet-xyz');
+  });
 });
 
 describe('resolveRefList with layer resolution', () => {

src/__tests__/globalConfig-defaults.test.ts

@@ -336,6 +336,63 @@ describe('loadGlobalConfig', () => {
     expect(config.interactivePreviewMovements).toBe(0);
   });
 
+  describe('persona_providers', () => {
+    it('should load persona_providers from config.yaml', () => {
+      const taktDir = join(testHomeDir, '.takt');
+      mkdirSync(taktDir, { recursive: true });
+      writeFileSync(
+        getGlobalConfigPath(),
+        [
+          'language: en',
+          'persona_providers:',
+          '  coder: codex',
+          '  reviewer: claude',
+        ].join('\n'),
+        'utf-8',
+      );
+
+      const config = loadGlobalConfig();
+
+      expect(config.personaProviders).toEqual({
+        coder: 'codex',
+        reviewer: 'claude',
+      });
+    });
+
+    it('should save and reload persona_providers', () => {
+      const taktDir = join(testHomeDir, '.takt');
+      mkdirSync(taktDir, { recursive: true });
+      writeFileSync(getGlobalConfigPath(), 'language: en\n', 'utf-8');
+
+      const config = loadGlobalConfig();
+      config.personaProviders = { coder: 'codex' };
+      saveGlobalConfig(config);
+      invalidateGlobalConfigCache();
+
+      const reloaded = loadGlobalConfig();
+      expect(reloaded.personaProviders).toEqual({ coder: 'codex' });
+    });
+
+    it('should have undefined personaProviders by default', () => {
+      const config = loadGlobalConfig();
+      expect(config.personaProviders).toBeUndefined();
+    });
+
+    it('should not save persona_providers when empty', () => {
+      const taktDir = join(testHomeDir, '.takt');
+      mkdirSync(taktDir, { recursive: true });
+      writeFileSync(getGlobalConfigPath(), 'language: en\n', 'utf-8');
+
+      const config = loadGlobalConfig();
+      config.personaProviders = {};
+      saveGlobalConfig(config);
+      invalidateGlobalConfigCache();
+
+      const reloaded = loadGlobalConfig();
+      expect(reloaded.personaProviders).toBeUndefined();
+    });
+  });
+
   describe('provider/model compatibility validation', () => {
     it('should throw when provider is codex but model is a Claude alias (opus)', () => {
       const taktDir = join(testHomeDir, '.takt');

src/__tests__/interactive-mode.test.ts

@@ -0,0 +1,532 @@
+/**
+ * Tests for interactive mode variants (assistant, persona, quiet, passthrough)
+ */
+
+import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
+
+// ── Mocks ──────────────────────────────────────────────
+
+vi.mock('../infra/config/global/globalConfig.js', () => ({
+  loadGlobalConfig: vi.fn(() => ({ provider: 'mock', language: 'en' })),
+  getBuiltinPiecesEnabled: vi.fn().mockReturnValue(true),
+}));
+
+vi.mock('../infra/providers/index.js', () => ({
+  getProvider: vi.fn(),
+}));
+
+vi.mock('../shared/utils/index.js', async (importOriginal) => ({
+  ...(await importOriginal<Record<string, unknown>>()),
+  createLogger: () => ({
+    info: vi.fn(),
+    debug: vi.fn(),
+    error: vi.fn(),
+  }),
+}));
+
+vi.mock('../shared/context.js', () => ({
+  isQuietMode: vi.fn(() => false),
+}));
+
+vi.mock('../infra/config/paths.js', async (importOriginal) => ({
+  ...(await importOriginal<Record<string, unknown>>()),
+  loadPersonaSessions: vi.fn(() => ({})),
+  updatePersonaSession: vi.fn(),
+  getProjectConfigDir: vi.fn(() => '/tmp'),
+  loadSessionState: vi.fn(() => null),
+  clearSessionState: vi.fn(),
+}));
+
+vi.mock('../shared/ui/index.js', () => ({
+  info: vi.fn(),
+  error: vi.fn(),
+  blankLine: vi.fn(),
+  StreamDisplay: vi.fn().mockImplementation(() => ({
+    createHandler: vi.fn(() => vi.fn()),
+    flush: vi.fn(),
+  })),
+}));
+
+vi.mock('../shared/prompt/index.js', () => ({
+  selectOption: vi.fn(),
+  selectOptionWithDefault: vi.fn(),
+}));
+
+import { getProvider } from '../infra/providers/index.js';
+import { selectOptionWithDefault, selectOption } from '../shared/prompt/index.js';
+
+const mockGetProvider = vi.mocked(getProvider);
+const mockSelectOptionWithDefault = vi.mocked(selectOptionWithDefault);
+const mockSelectOption = vi.mocked(selectOption);
+
+// ── Stdin helpers (same pattern as interactive.test.ts) ──
+
+let savedIsTTY: boolean | undefined;
+let savedIsRaw: boolean | undefined;
+let savedSetRawMode: typeof process.stdin.setRawMode | undefined;
+let savedStdoutWrite: typeof process.stdout.write;
+let savedStdinOn: typeof process.stdin.on;
+let savedStdinRemoveListener: typeof process.stdin.removeListener;
+let savedStdinResume: typeof process.stdin.resume;
+let savedStdinPause: typeof process.stdin.pause;
+
+function setupRawStdin(rawInputs: string[]): void {
+  savedIsTTY = process.stdin.isTTY;
+  savedIsRaw = process.stdin.isRaw;
+  savedSetRawMode = process.stdin.setRawMode;
+  savedStdoutWrite = process.stdout.write;
+  savedStdinOn = process.stdin.on;
+  savedStdinRemoveListener = process.stdin.removeListener;
+  savedStdinResume = process.stdin.resume;
+  savedStdinPause = process.stdin.pause;
+
+  Object.defineProperty(process.stdin, 'isTTY', { value: true, configurable: true });
+  Object.defineProperty(process.stdin, 'isRaw', { value: false, configurable: true, writable: true });
+  process.stdin.setRawMode = vi.fn((mode: boolean) => {
+    (process.stdin as unknown as { isRaw: boolean }).isRaw = mode;
+    return process.stdin;
+  }) as unknown as typeof process.stdin.setRawMode;
+  process.stdout.write = vi.fn(() => true) as unknown as typeof process.stdout.write;
+  process.stdin.resume = vi.fn(() => process.stdin) as unknown as typeof process.stdin.resume;
+  process.stdin.pause = vi.fn(() => process.stdin) as unknown as typeof process.stdin.pause;
+
+  let currentHandler: ((data: Buffer) => void) | null = null;
+  let inputIndex = 0;
+
+  process.stdin.on = vi.fn(((event: string, handler: (...args: unknown[]) => void) => {
+    if (event === 'data') {
+      currentHandler = handler as (data: Buffer) => void;
+      if (inputIndex < rawInputs.length) {
+        const data = rawInputs[inputIndex]!;
+        inputIndex++;
+        queueMicrotask(() => {
+          if (currentHandler) {
+            currentHandler(Buffer.from(data, 'utf-8'));
+          }
+        });
+      }
+    }
+    return process.stdin;
+  }) as typeof process.stdin.on);
+
+  process.stdin.removeListener = vi.fn(((event: string) => {
+    if (event === 'data') {
+      currentHandler = null;
+    }
+    return process.stdin;
+  }) as typeof process.stdin.removeListener);
+}
+
+function restoreStdin(): void {
+  if (savedIsTTY !== undefined) {
+    Object.defineProperty(process.stdin, 'isTTY', { value: savedIsTTY, configurable: true });
+  }
+  if (savedIsRaw !== undefined) {
+    Object.defineProperty(process.stdin, 'isRaw', { value: savedIsRaw, configurable: true, writable: true });
+  }
+  if (savedSetRawMode) process.stdin.setRawMode = savedSetRawMode;
+  if (savedStdoutWrite) process.stdout.write = savedStdoutWrite;
+  if (savedStdinOn) process.stdin.on = savedStdinOn;
+  if (savedStdinRemoveListener) process.stdin.removeListener = savedStdinRemoveListener;
+  if (savedStdinResume) process.stdin.resume = savedStdinResume;
+  if (savedStdinPause) process.stdin.pause = savedStdinPause;
+}
+
+function toRawInputs(inputs: (string | null)[]): string[] {
+  return inputs.map((input) => {
+    if (input === null) return '\x04';
+    return input + '\r';
+  });
+}
+
+function setupMockProvider(responses: string[]): void {
+  let callIndex = 0;
+  const mockCall = vi.fn(async () => {
+    const content = callIndex < responses.length ? responses[callIndex] : 'AI response';
+    callIndex++;
+    return {
+      persona: 'interactive',
+      status: 'done' as const,
+      content: content!,
+      timestamp: new Date(),
+    };
+  });
+  const mockSetup = vi.fn(() => ({ call: mockCall }));
+  const mockProvider = {
+    setup: mockSetup,
+    _call: mockCall,
+    _setup: mockSetup,
+  };
+  mockGetProvider.mockReturnValue(mockProvider);
+}
+
+// ── Imports (after mocks) ──
+
+import { INTERACTIVE_MODES, DEFAULT_INTERACTIVE_MODE } from '../core/models/interactive-mode.js';
+import { selectInteractiveMode } from '../features/interactive/modeSelection.js';
+import { passthroughMode } from '../features/interactive/passthroughMode.js';
+import { quietMode } from '../features/interactive/quietMode.js';
+import { personaMode } from '../features/interactive/personaMode.js';
+import type { PieceContext } from '../features/interactive/interactive.js';
+import type { FirstMovementInfo } from '../infra/config/loaders/pieceResolver.js';
+
+// ── Setup ──
+
+beforeEach(() => {
+  vi.clearAllMocks();
+  mockSelectOptionWithDefault.mockResolvedValue('assistant');
+  mockSelectOption.mockResolvedValue('execute');
+});
+
+afterEach(() => {
+  restoreStdin();
+});
+
+// ── InteractiveMode type & constants tests ──
+
+describe('InteractiveMode type', () => {
+  it('should define all four modes', () => {
+    expect(INTERACTIVE_MODES).toEqual(['assistant', 'persona', 'quiet', 'passthrough']);
+  });
+
+  it('should have assistant as default mode', () => {
+    expect(DEFAULT_INTERACTIVE_MODE).toBe('assistant');
+  });
+});
+
+// ── Mode selection tests ──
+
+describe('selectInteractiveMode', () => {
+  it('should call selectOptionWithDefault with four mode options', async () => {
+    // When
+    await selectInteractiveMode('en');
+
+    // Then
+    expect(mockSelectOptionWithDefault).toHaveBeenCalledWith(
+      expect.any(String),
+      expect.arrayContaining([
+        expect.objectContaining({ value: 'assistant' }),
+        expect.objectContaining({ value: 'persona' }),
+        expect.objectContaining({ value: 'quiet' }),
+        expect.objectContaining({ value: 'passthrough' }),
+      ]),
+      'assistant',
+    );
+  });
+
+  it('should use piece default when provided', async () => {
+    // When
+    await selectInteractiveMode('en', 'quiet');
+
+    // Then
+    expect(mockSelectOptionWithDefault).toHaveBeenCalledWith(
+      expect.any(String),
+      expect.any(Array),
+      'quiet',
+    );
+  });
+
+  it('should return null when user cancels', async () => {
+    // Given
+    mockSelectOptionWithDefault.mockResolvedValue(null);
+
+    // When
+    const result = await selectInteractiveMode('en');
+
+    // Then
+    expect(result).toBeNull();
+  });
+
+  it('should return selected mode value', async () => {
+    // Given
+    mockSelectOptionWithDefault.mockResolvedValue('persona');
+
+    // When
+    const result = await selectInteractiveMode('ja');
+
+    // Then
+    expect(result).toBe('persona');
+  });
+
+  it('should present options in correct order', async () => {
+    // When
+    await selectInteractiveMode('en');
+
+    // Then
+    const options = mockSelectOptionWithDefault.mock.calls[0]?.[1] as Array<{ value: string }>;
+    expect(options?.[0]?.value).toBe('assistant');
+    expect(options?.[1]?.value).toBe('persona');
+    expect(options?.[2]?.value).toBe('quiet');
+    expect(options?.[3]?.value).toBe('passthrough');
+  });
+});
+
+// ── Passthrough mode tests ──
+
+describe('passthroughMode', () => {
+  it('should return initialInput directly when provided', async () => {
+    // When
+    const result = await passthroughMode('en', 'my task text');
+
+    // Then
+    expect(result.action).toBe('execute');
+    expect(result.task).toBe('my task text');
+  });
+
+  it('should return cancel when user sends EOF', async () => {
+    // Given
+    setupRawStdin(toRawInputs([null]));
+
+    // When
+    const result = await passthroughMode('en');
+
+    // Then
+    expect(result.action).toBe('cancel');
+    expect(result.task).toBe('');
+  });
+
+  it('should return cancel when user enters empty input', async () => {
+    // Given
+    setupRawStdin(toRawInputs(['']));
+
+    // When
+    const result = await passthroughMode('en');
+
+    // Then
+    expect(result.action).toBe('cancel');
+  });
+
+  it('should return user input as task when entered', async () => {
+    // Given
+    setupRawStdin(toRawInputs(['implement login feature']));
+
+    // When
+    const result = await passthroughMode('en');
+
+    // Then
+    expect(result.action).toBe('execute');
+    expect(result.task).toBe('implement login feature');
+  });
+
+  it('should trim whitespace from user input', async () => {
+    // Given
+    setupRawStdin(toRawInputs(['  my task  ']));
+
+    // When
+    const result = await passthroughMode('en');
+
+    // Then
+    expect(result.task).toBe('my task');
+  });
+});
+
+// ── Quiet mode tests ──
+
+describe('quietMode', () => {
+  it('should generate instructions from initialInput without questions', async () => {
+    // Given
+    setupMockProvider(['Generated task instruction for login feature.']);
+    mockSelectOption.mockResolvedValue('execute');
+
+    // When
+    const result = await quietMode('/project', 'implement login feature');
+
+    // Then
+    expect(result.action).toBe('execute');
+    expect(result.task).toBe('Generated task instruction for login feature.');
+  });
+
+  it('should return cancel when user sends EOF for input', async () => {
+    // Given
+    setupRawStdin(toRawInputs([null]));
+    setupMockProvider([]);
+
+    // When
+    const result = await quietMode('/project');
+
+    // Then
+    expect(result.action).toBe('cancel');
+  });
+
+  it('should return cancel when user enters empty input', async () => {
+    // Given
+    setupRawStdin(toRawInputs(['']));
+    setupMockProvider([]);
+
+    // When
+    const result = await quietMode('/project');
+
+    // Then
+    expect(result.action).toBe('cancel');
+  });
+
+  it('should prompt for input when no initialInput is provided', async () => {
+    // Given
+    setupRawStdin(toRawInputs(['fix the bug']));
+    setupMockProvider(['Fix the bug instruction.']);
+    mockSelectOption.mockResolvedValue('execute');
+
+    // When
+    const result = await quietMode('/project');
+
+    // Then
+    expect(result.action).toBe('execute');
+    expect(result.task).toBe('Fix the bug instruction.');
+  });
+
+  it('should include piece context in summary generation', async () => {
+    // Given
+    const pieceContext: PieceContext = {
+      name: 'test-piece',
+      description: 'A test piece',
+      pieceStructure: '1. implement\n2. review',
+      movementPreviews: [],
+    };
+    setupMockProvider(['Instruction with piece context.']);
+    mockSelectOption.mockResolvedValue('execute');
+
+    // When
+    const result = await quietMode('/project', 'some task', pieceContext);
+
+    // Then
+    expect(result.action).toBe('execute');
+    expect(result.task).toBe('Instruction with piece context.');
+  });
+});
+
+// ── Persona mode tests ──
+
+describe('personaMode', () => {
+  const mockFirstMovement: FirstMovementInfo = {
+    personaContent: 'You are a senior coder. Write clean, maintainable code.',
+    personaDisplayName: 'Coder',
+    allowedTools: ['Read', 'Glob', 'Grep', 'Edit', 'Write', 'Bash'],
+  };
+
+  it('should return cancel when user types /cancel', async () => {
+    // Given
+    setupRawStdin(toRawInputs(['/cancel']));
+    setupMockProvider([]);
+
+    // When
+    const result = await personaMode('/project', mockFirstMovement);
+
+    // Then
+    expect(result.action).toBe('cancel');
+    expect(result.task).toBe('');
+  });
+
+  it('should return cancel on EOF', async () => {
+    // Given
+    setupRawStdin(toRawInputs([null]));
+    setupMockProvider([]);
+
+    // When
+    const result = await personaMode('/project', mockFirstMovement);
+
+    // Then
+    expect(result.action).toBe('cancel');
+  });
+
+  it('should use first movement persona as system prompt', async () => {
+    // Given
+    setupRawStdin(toRawInputs(['fix bug', '/cancel']));
+    setupMockProvider(['I see the issue.']);
+
+    // When
+    await personaMode('/project', mockFirstMovement);
+
+    // Then: the provider should be set up with persona content as system prompt
+    const mockProvider = mockGetProvider.mock.results[0]!.value as { _setup: ReturnType<typeof vi.fn> };
+    expect(mockProvider._setup).toHaveBeenCalledWith(
+      expect.objectContaining({
+        systemPrompt: 'You are a senior coder. Write clean, maintainable code.',
+      }),
+    );
+  });
+
+  it('should use first movement allowed tools', async () => {
+    // Given
+    setupRawStdin(toRawInputs(['check the code', '/cancel']));
+    setupMockProvider(['Looking at the code.']);
+
+    // When
+    await personaMode('/project', mockFirstMovement);
+
+    // Then
+    const mockProvider = mockGetProvider.mock.results[0]!.value as { _call: ReturnType<typeof vi.fn> };
+    expect(mockProvider._call).toHaveBeenCalledWith(
+      expect.any(String),
+      expect.objectContaining({
+        allowedTools: ['Read', 'Glob', 'Grep', 'Edit', 'Write', 'Bash'],
+      }),
+    );
+  });
+
+  it('should process initialInput as first message', async () => {
+    // Given
+    setupRawStdin(toRawInputs(['/go']));
+    setupMockProvider(['I analyzed the issue.', 'Task summary.']);
+    mockSelectOption.mockResolvedValue('execute');
+
+    // When
+    const result = await personaMode('/project', mockFirstMovement, 'fix the login');
+
+    // Then
+    expect(result.action).toBe('execute');
+    const mockProvider = mockGetProvider.mock.results[0]!.value as { _call: ReturnType<typeof vi.fn> };
+    expect(mockProvider._call).toHaveBeenCalledTimes(2);
+    const firstPrompt = mockProvider._call.mock.calls[0]?.[0] as string;
+    expect(firstPrompt).toBe('fix the login');
+  });
+
+  it('should handle /play command', async () => {
+    // Given
+    setupRawStdin(toRawInputs(['/play direct task text']));
+    setupMockProvider([]);
+
+    // When
+    const result = await personaMode('/project', mockFirstMovement);
+
+    // Then
+    expect(result.action).toBe('execute');
+    expect(result.task).toBe('direct task text');
+  });
+
+  it('should fall back to default tools when first movement has none', async () => {
+    // Given
+    const noToolsMovement: FirstMovementInfo = {
+      personaContent: 'Persona prompt',
+      personaDisplayName: 'Agent',
+      allowedTools: [],
+    };
+    setupRawStdin(toRawInputs(['test', '/cancel']));
+    setupMockProvider(['response']);
+
+    // When
+    await personaMode('/project', noToolsMovement);
+
+    // Then
+    const mockProvider = mockGetProvider.mock.results[0]!.value as { _call: ReturnType<typeof vi.fn> };
+    expect(mockProvider._call).toHaveBeenCalledWith(
+      expect.any(String),
+      expect.objectContaining({
+        allowedTools: ['Read', 'Glob', 'Grep', 'Bash', 'WebSearch', 'WebFetch'],
+      }),
+    );
+  });
+
+  it('should handle multi-turn conversation before /go', async () => {
+    // Given
+    setupRawStdin(toRawInputs(['first message', 'second message', '/go']));
+    setupMockProvider(['reply 1', 'reply 2', 'Final summary.']);
+    mockSelectOption.mockResolvedValue('execute');
+
+    // When
+    const result = await personaMode('/project', mockFirstMovement);
+
+    // Then
+    expect(result.action).toBe('execute');
+    expect(result.task).toBe('Final summary.');
+  });
+});

src/__tests__/it-piece-loader.test.ts

@@ -25,6 +25,7 @@ vi.mock('../infra/config/global/globalConfig.js', () => ({
 // --- Imports (after mocks) ---
 
 import { loadPiece } from '../infra/config/index.js';
+import { listBuiltinPieceNames } from '../infra/config/loaders/pieceResolver.js';
 
 // --- Test helpers ---
 
@@ -45,7 +46,7 @@ describe('Piece Loader IT: builtin piece loading', () => {
     rmSync(testDir, { recursive: true, force: true });
   });
 
-  const builtinNames = ['default', 'minimal', 'expert', 'expert-cqrs', 'research', 'magi', 'review-only', 'review-fix-minimal'];
+  const builtinNames = listBuiltinPieceNames({ includeDisabled: true });
 
   for (const name of builtinNames) {
     it(`should load builtin piece: ${name}`, () => {
@@ -572,6 +573,139 @@ movements:
   });
 });
 
+describe('Piece Loader IT: structural-reform piece', () => {
+  let testDir: string;
+
+  beforeEach(() => {
+    testDir = createTestDir();
+  });
+
+  afterEach(() => {
+    rmSync(testDir, { recursive: true, force: true });
+  });
+
+  it('should load structural-reform with 7 movements', () => {
+    const config = loadPiece('structural-reform', testDir);
+
+    expect(config).not.toBeNull();
+    expect(config!.name).toBe('structural-reform');
+    expect(config!.movements.length).toBe(7);
+    expect(config!.maxIterations).toBe(50);
+    expect(config!.initialMovement).toBe('review');
+  });
+
+  it('should have expected movement names in order', () => {
+    const config = loadPiece('structural-reform', testDir);
+    expect(config).not.toBeNull();
+
+    const movementNames = config!.movements.map((m) => m.name);
+    expect(movementNames).toEqual([
+      'review',
+      'plan_reform',
+      'implement',
+      'reviewers',
+      'fix',
+      'verify',
+      'next_target',
+    ]);
+  });
+
+  it('should have review as read-only with instruction_template', () => {
+    const config = loadPiece('structural-reform', testDir);
+    expect(config).not.toBeNull();
+
+    const review = config!.movements.find((m) => m.name === 'review');
+    expect(review).toBeDefined();
+    expect(review!.edit).not.toBe(true);
+    expect(review!.instructionTemplate).toBeDefined();
+    expect(review!.instructionTemplate).toContain('{task}');
+  });
+
+  it('should have implement with edit: true and session: refresh', () => {
+    const config = loadPiece('structural-reform', testDir);
+    expect(config).not.toBeNull();
+
+    const implement = config!.movements.find((m) => m.name === 'implement');
+    expect(implement).toBeDefined();
+    expect(implement!.edit).toBe(true);
+    expect(implement!.session).toBe('refresh');
+  });
+
+  it('should have 2 parallel reviewers (arch-review and qa-review)', () => {
+    const config = loadPiece('structural-reform', testDir);
+    expect(config).not.toBeNull();
+
+    const reviewers = config!.movements.find(
+      (m) => m.parallel && m.parallel.length > 0,
+    );
+    expect(reviewers).toBeDefined();
+    expect(reviewers!.parallel!.length).toBe(2);
+
+    const subNames = reviewers!.parallel!.map((s) => s.name);
+    expect(subNames).toContain('arch-review');
+    expect(subNames).toContain('qa-review');
+  });
+
+  it('should have aggregate rules on reviewers movement', () => {
+    const config = loadPiece('structural-reform', testDir);
+    expect(config).not.toBeNull();
+
+    const reviewers = config!.movements.find(
+      (m) => m.parallel && m.parallel.length > 0,
+    );
+    expect(reviewers).toBeDefined();
+
+    const allRule = reviewers!.rules?.find(
+      (r) => r.isAggregateCondition && r.aggregateType === 'all',
+    );
+    expect(allRule).toBeDefined();
+    expect(allRule!.aggregateConditionText).toBe('approved');
+    expect(allRule!.next).toBe('verify');
+
+    const anyRule = reviewers!.rules?.find(
+      (r) => r.isAggregateCondition && r.aggregateType === 'any',
+    );
+    expect(anyRule).toBeDefined();
+    expect(anyRule!.aggregateConditionText).toBe('needs_fix');
+    expect(anyRule!.next).toBe('fix');
+  });
+
+  it('should have verify movement with instruction_template', () => {
+    const config = loadPiece('structural-reform', testDir);
+    expect(config).not.toBeNull();
+
+    const verify = config!.movements.find((m) => m.name === 'verify');
+    expect(verify).toBeDefined();
+    expect(verify!.edit).not.toBe(true);
+    expect(verify!.instructionTemplate).toBeDefined();
+  });
+
+  it('should have next_target movement routing to implement or COMPLETE', () => {
+    const config = loadPiece('structural-reform', testDir);
+    expect(config).not.toBeNull();
+
+    const nextTarget = config!.movements.find((m) => m.name === 'next_target');
+    expect(nextTarget).toBeDefined();
+    expect(nextTarget!.edit).not.toBe(true);
+
+    const nextValues = nextTarget!.rules?.map((r) => r.next);
+    expect(nextValues).toContain('implement');
+    expect(nextValues).toContain('COMPLETE');
+  });
+
+  it('should have loop_monitors for implement-fix cycle', () => {
+    const config = loadPiece('structural-reform', testDir);
+    expect(config).not.toBeNull();
+    expect(config!.loopMonitors).toBeDefined();
+    expect(config!.loopMonitors!.length).toBe(1);
+
+    const monitor = config!.loopMonitors![0]!;
+    expect(monitor.cycle).toEqual(['implement', 'fix']);
+    expect(monitor.threshold).toBe(3);
+    expect(monitor.judge).toBeDefined();
+  });
+});
+
 describe('Piece Loader IT: invalid YAML handling', () => {
   let testDir: string;
 

src/__tests__/lineEditor.test.ts

@@ -131,9 +131,11 @@ describe('readMultilineInput cursor navigation', () => {
   let savedStdinRemoveListener: typeof process.stdin.removeListener;
   let savedStdinResume: typeof process.stdin.resume;
   let savedStdinPause: typeof process.stdin.pause;
+  let savedColumns: number | undefined;
+  let columnsOverridden = false;
   let stdoutCalls: string[];
 
-  function setupRawStdin(rawInputs: string[]): void {
+  function setupRawStdin(rawInputs: string[], termColumns?: number): void {
     savedIsTTY = process.stdin.isTTY;
     savedIsRaw = process.stdin.isRaw;
     savedSetRawMode = process.stdin.setRawMode;
@@ -142,6 +144,13 @@ describe('readMultilineInput cursor navigation', () => {
     savedStdinRemoveListener = process.stdin.removeListener;
     savedStdinResume = process.stdin.resume;
     savedStdinPause = process.stdin.pause;
+    savedColumns = process.stdout.columns;
+    columnsOverridden = false;
+
+    if (termColumns !== undefined) {
+      Object.defineProperty(process.stdout, 'columns', { value: termColumns, configurable: true, writable: true });
+      columnsOverridden = true;
+    }
 
     Object.defineProperty(process.stdin, 'isTTY', { value: true, configurable: true });
     Object.defineProperty(process.stdin, 'isRaw', { value: false, configurable: true, writable: true });
@@ -197,6 +206,10 @@ describe('readMultilineInput cursor navigation', () => {
     if (savedStdinRemoveListener) process.stdin.removeListener = savedStdinRemoveListener;
     if (savedStdinResume) process.stdin.resume = savedStdinResume;
     if (savedStdinPause) process.stdin.pause = savedStdinPause;
+    if (columnsOverridden) {
+      Object.defineProperty(process.stdout, 'columns', { value: savedColumns, configurable: true, writable: true });
+      columnsOverridden = false;
+    }
   }
 
   beforeEach(() => {
@@ -611,4 +624,338 @@ describe('readMultilineInput cursor navigation', () => {
       expect(result).toBe('abc\ndef\nghiX');
     });
   });
+
+  describe('soft-wrap: arrow up within wrapped line', () => {
+    it('should move to previous display row within same logical line', async () => {
+      // Given: termWidth=20, prompt "> " (2 cols), first display row = 18 chars, second = 20 chars
+      // Type 30 chars "abcdefghijklmnopqrstuvwxyz1234" → wraps at pos 18
+      // Display row 1: "abcdefghijklmnopqr" (18 chars, cols 3-20 with prompt)
+      // Display row 2: "stuvwxyz1234" (12 chars, cols 1-12)
+      // Cursor at end (pos 30, display col 12), press ↑ → display col 12 in row 1 → pos 12
+      // Insert "X" → "abcdefghijklXmnopqrstuvwxyz1234"
+      setupRawStdin([
+        'abcdefghijklmnopqrstuvwxyz1234\x1B[AX\r',
+      ], 20);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('abcdefghijklXmnopqrstuvwxyz1234');
+    });
+
+    it('should do nothing when on first display row of first logical line', async () => {
+      // Given: termWidth=20, prompt "> " (2 cols), type "abcdefghij" (10 chars, fits in first row of 18 cols)
+      // Cursor at end (pos 10, first display row), press ↑ → no previous row, nothing happens
+      // Insert "X" → "abcdefghijX"
+      setupRawStdin([
+        'abcdefghij\x1B[AX\r',
+      ], 20);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('abcdefghijX');
+    });
+  });
+
+  describe('soft-wrap: arrow down within wrapped line', () => {
+    it('should move to next display row within same logical line', async () => {
+      // Given: termWidth=20, prompt "> " (2 cols), first row = 18 chars
+      // Type 30 chars, Home → pos 0, then ↓ → display col 0 in row 2 → pos 18
+      // Insert "X" → "abcdefghijklmnopqrXstuvwxyz1234"
+      setupRawStdin([
+        'abcdefghijklmnopqrstuvwxyz1234\x1B[H\x1B[BX\r',
+      ], 20);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('abcdefghijklmnopqrXstuvwxyz1234');
+    });
+
+    it('should do nothing when on last display row of last logical line', async () => {
+      // Given: termWidth=20, prompt "> " (2 cols), type 30 chars (wraps into 2 display rows)
+      // Cursor at end (last display row), press ↓ → nothing happens
+      // Insert "X" → "abcdefghijklmnopqrstuvwxyz1234X"
+      setupRawStdin([
+        'abcdefghijklmnopqrstuvwxyz1234\x1B[BX\r',
+      ], 20);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('abcdefghijklmnopqrstuvwxyz1234X');
+    });
+  });
+
+  describe('soft-wrap: Ctrl+A moves to display row start', () => {
+    it('should move to display row start on wrapped second row', async () => {
+      // Given: termWidth=20, prompt "> " (2 cols), type 30 chars
+      // Cursor at end (pos 30), Ctrl+A → display row start (pos 18), insert "X"
+      // Result: "abcdefghijklmnopqrXstuvwxyz1234"
+      setupRawStdin([
+        'abcdefghijklmnopqrstuvwxyz1234\x01X\r',
+      ], 20);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('abcdefghijklmnopqrXstuvwxyz1234');
+    });
+
+    it('should move to display row start on first row', async () => {
+      // Given: termWidth=20, prompt "> " (2 cols), type 30 chars
+      // Move cursor to middle of first display row (Home, Right*5 → pos 5)
+      // Ctrl+A → pos 0, insert "X"
+      // Result: "Xabcdefghijklmnopqrstuvwxyz1234"
+      setupRawStdin([
+        'abcdefghijklmnopqrstuvwxyz1234\x1B[H\x1B[C\x1B[C\x1B[C\x1B[C\x1B[C\x01X\r',
+      ], 20);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('Xabcdefghijklmnopqrstuvwxyz1234');
+    });
+  });
+
+  describe('soft-wrap: Ctrl+E moves to display row end', () => {
+    it('should move to display row end on first row', async () => {
+      // Given: termWidth=20, prompt "> " (2 cols), type 30 chars
+      // Home → pos 0, Ctrl+E → end of first display row (pos 18), insert "X"
+      // Result: "abcdefghijklmnopqrXstuvwxyz1234"
+      setupRawStdin([
+        'abcdefghijklmnopqrstuvwxyz1234\x1B[H\x05X\r',
+      ], 20);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('abcdefghijklmnopqrXstuvwxyz1234');
+    });
+  });
+
+  describe('soft-wrap: Home moves to logical line start', () => {
+    it('should move from wrapped second row to logical line start', async () => {
+      // Given: termWidth=20, prompt "> " (2 cols), first row = 18 chars
+      // Type 30 chars, cursor at end (pos 30, second display row)
+      // Home → logical line start (pos 0), insert "X"
+      // Result: "Xabcdefghijklmnopqrstuvwxyz1234"
+      setupRawStdin([
+        'abcdefghijklmnopqrstuvwxyz1234\x1B[HX\r',
+      ], 20);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('Xabcdefghijklmnopqrstuvwxyz1234');
+    });
+
+    it('should emit cursor up sequence when crossing display rows', async () => {
+      // Given: termWidth=20, prompt "> " (2 cols), type 30 chars (wraps into 2 rows)
+      // Cursor at end (second display row), Home → pos 0 (first display row)
+      setupRawStdin([
+        'abcdefghijklmnopqrstuvwxyz1234\x1B[H\r',
+      ], 20);
+
+      // When
+      await callReadMultilineInput('> ');
+
+      // Then: should contain \x1B[{n}A for moving up display rows
+      const hasUpMove = stdoutCalls.some(c => /^\x1B\[\d+A$/.test(c));
+      expect(hasUpMove).toBe(true);
+    });
+  });
+
+  describe('soft-wrap: End moves to logical line end', () => {
+    it('should move from first display row to logical line end', async () => {
+      // Given: termWidth=20, prompt "> " (2 cols), first row = 18 chars
+      // Type 30 chars, Home → pos 0, End → logical line end (pos 30), insert "X"
+      // Result: "abcdefghijklmnopqrstuvwxyz1234X"
+      setupRawStdin([
+        'abcdefghijklmnopqrstuvwxyz1234\x1B[H\x1B[FX\r',
+      ], 20);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('abcdefghijklmnopqrstuvwxyz1234X');
+    });
+
+    it('should emit cursor down sequence when crossing display rows', async () => {
+      // Given: termWidth=20, prompt "> " (2 cols), type 30 chars (wraps into 2 rows)
+      // Home → pos 0 (first display row), End → pos 30 (second display row)
+      setupRawStdin([
+        'abcdefghijklmnopqrstuvwxyz1234\x1B[H\x1B[F\r',
+      ], 20);
+
+      // When
+      await callReadMultilineInput('> ');
+
+      // Then: should contain \x1B[{n}B for moving down display rows
+      const hasDownMove = stdoutCalls.some(c => /^\x1B\[\d+B$/.test(c));
+      expect(hasDownMove).toBe(true);
+    });
+
+    it('should stay at end when already at logical line end on last display row', async () => {
+      // Given: termWidth=20, prompt "> " (2 cols), type 30 chars
+      // Cursor at end (pos 30, already at logical line end), End → nothing changes, insert "X"
+      // Result: "abcdefghijklmnopqrstuvwxyz1234X"
+      setupRawStdin([
+        'abcdefghijklmnopqrstuvwxyz1234\x1B[FX\r',
+      ], 20);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('abcdefghijklmnopqrstuvwxyz1234X');
+    });
+  });
+
+  describe('soft-wrap: non-wrapped text retains original behavior', () => {
+    it('should not affect arrow up on short single-line text', async () => {
+      // Given: termWidth=80, short text "abc" (no wrap), ↑ does nothing
+      setupRawStdin([
+        'abc\x1B[AX\r',
+      ], 80);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('abcX');
+    });
+
+    it('should not affect arrow down on short single-line text', async () => {
+      // Given: termWidth=80, short text "abc" (no wrap), ↓ does nothing
+      setupRawStdin([
+        'abc\x1B[BX\r',
+      ], 80);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('abcX');
+    });
+
+    it('should still navigate between logical lines with arrow up', async () => {
+      // Given: termWidth=80, "abcde\nfgh" (no wrap), cursor at end of "fgh", ↑ → "abcde" at col 3
+      setupRawStdin([
+        'abcde\x1B[13;2ufgh\x1B[AX\r',
+      ], 80);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('abcXde\nfgh');
+    });
+  });
+
+  describe('soft-wrap: full-width characters', () => {
+    it('should calculate display row boundaries with full-width chars', async () => {
+      // Given: termWidth=10, prompt "> " (2 cols), first row available = 8 cols
+      // Type "あいうえ" (4 full-width chars = 8 display cols = fills first row exactly)
+      // Then type "お" (2 cols, starts second row)
+      // Cursor at end (after "お"), Ctrl+A → display row start (pos 4, start of "お")
+      // Insert "X"
+      // Result: "あいうえXお"
+      setupRawStdin([
+        'あいうえお\x01X\r',
+      ], 10);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('あいうえXお');
+    });
+
+    it('should push full-width char to next row when only 1 column remains', async () => {
+      // Given: termWidth=10, prompt "> " (2 cols), first row available = 8 cols
+      // Type "abcdefg" (7 cols) then "あ" (2 cols) → 7+2=9 > 8, "あ" goes to row 2
+      // Cursor at end (after "あ"), Ctrl+A → display row start at "あ" (pos 7)
+      // Insert "X"
+      // Result: "abcdefgXあ"
+      setupRawStdin([
+        'abcdefgあ\x01X\r',
+      ], 10);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('abcdefgXあ');
+    });
+  });
+
+  describe('soft-wrap: prompt width consideration', () => {
+    it('should account for prompt width in first display row', async () => {
+      // Given: termWidth=10, prompt "> " (2 cols), first row = 8 chars
+      // Type "12345678" (8 chars = fills first row) then "9" (starts row 2)
+      // Cursor at "9" (pos 9), ↑ → row 1 at display col 1, but only 8 chars available
+      // Display col 1 → pos 1
+      // Insert "X" → "1X234567890" ... wait, let me recalculate.
+      // Actually: cursor at end of "123456789" (pos 9, display col 1 in row 2)
+      // ↑ → display col 1 in row 1 → pos 1
+      // Insert "X" → "1X23456789"
+      setupRawStdin([
+        '123456789\x1B[AX\r',
+      ], 10);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('1X23456789');
+    });
+
+    it('should not add prompt offset for second logical line', async () => {
+      // Given: termWidth=10, prompt "> " (2 cols)
+      // Type "ab\n123456789" → second logical line "123456789" (9 chars, fits in 10 col row)
+      // Cursor at end (pos 12), ↑ → "ab" at display col 9 → clamped to col 2 → pos 2 (end of "ab")
+      // Insert "X" → "abX\n123456789"
+      setupRawStdin([
+        'ab\x1B[13;2u123456789\x1B[AX\r',
+      ], 10);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('abX\n123456789');
+    });
+  });
+
+  describe('soft-wrap: cross logical line with display rows', () => {
+    it('should move from wrapped logical line to previous logical line last display row', async () => {
+      // Given: termWidth=20, prompt "> " (2 cols)
+      // Line 1: "abcdefghijklmnopqrstuvwx" (24 chars) → wraps: row 1 (18 chars) + row 2 (6 chars)
+      // Line 2: "123"
+      // Cursor at end of "123" (display col 3), ↑ → last display row of line 1 (row 2: "uvwx", 6 chars)
+      // Display col 3 → pos 21 ("v" position... let me calculate)
+      // Row 2 of line 1 starts at pos 18 ("stuvwx"), display col 3 → pos 21
+      // Insert "X" → "abcdefghijklmnopqrstuXvwx\n123"
+      setupRawStdin([
+        'abcdefghijklmnopqrstuvwx\x1B[13;2u123\x1B[AX\r',
+      ], 20);
+
+      // When
+      const result = await callReadMultilineInput('> ');
+
+      // Then
+      expect(result).toBe('abcdefghijklmnopqrstuXvwx\n123');
+    });
+  });
 });

src/__tests__/parallel-logger.test.ts

@@ -67,6 +67,29 @@ describe('ParallelLogger', () => {
       // No padding needed (0 spaces)
       expect(prefix).toMatch(/\x1b\[0m $/);
     });
+
+    it('should build rich prefix with task and parent movement for parallel task mode', () => {
+      const logger = new ParallelLogger({
+        subMovementNames: ['arch-review'],
+        writeFn,
+        progressInfo: {
+          iteration: 4,
+          maxIterations: 30,
+        },
+        taskLabel: 'override-persona-provider',
+        taskColorIndex: 0,
+        parentMovementName: 'reviewers',
+        movementIteration: 1,
+      });
+
+      const prefix = logger.buildPrefix('arch-review', 0);
+      expect(prefix).toContain('\x1b[36m');
+      expect(prefix).toContain('[over]');
+      expect(prefix).toContain('[reviewers]');
+      expect(prefix).toContain('[arch-review]');
+      expect(prefix).toContain('(4/30)(1)');
+      expect(prefix).not.toContain('step 1/1');
+    });
   });
 
   describe('text event line buffering', () => {

src/__tests__/pieceExecution-debug-prompts.test.ts

@@ -14,10 +14,12 @@ const { mockIsDebugEnabled, mockWritePromptLog, MockPieceEngine } = vi.hoisted((
 
   class MockPieceEngine extends EE {
     private config: PieceConfig;
+    private task: string;
 
-    constructor(config: PieceConfig, _cwd: string, _task: string, _options: unknown) {
+    constructor(config: PieceConfig, _cwd: string, task: string, _options: unknown) {
       super();
       this.config = config;
+      this.task = task;
     }
 
     abort(): void {}
@@ -26,6 +28,7 @@ const { mockIsDebugEnabled, mockWritePromptLog, MockPieceEngine } = vi.hoisted((
       const step = this.config.movements[0]!;
       const timestamp = new Date('2026-02-07T00:00:00.000Z');
 
+      const shouldRepeatMovement = this.task === 'repeat-movement-task';
       this.emit('movement:start', step, 1, 'movement instruction');
       this.emit('phase:start', step, 1, 'execute', 'phase prompt');
       this.emit('phase:complete', step, 1, 'execute', 'phase response', 'done');
@@ -40,9 +43,23 @@ const { mockIsDebugEnabled, mockWritePromptLog, MockPieceEngine } = vi.hoisted((
         },
         'movement instruction'
       );
+      if (shouldRepeatMovement) {
+        this.emit('movement:start', step, 2, 'movement instruction repeat');
+        this.emit(
+          'movement:complete',
+          step,
+          {
+            persona: step.personaDisplayName,
+            status: 'done',
+            content: 'movement response repeat',
+            timestamp,
+          },
+          'movement instruction repeat'
+        );
+      }
       this.emit('piece:complete', { status: 'completed', iteration: 1 });
 
-      return { status: 'completed', iteration: 1 };
+      return { status: 'completed', iteration: shouldRepeatMovement ? 2 : 1 };
     }
   }
 
@@ -187,4 +204,32 @@ describe('executePiece debug prompts logging', () => {
 
     expect(mockWritePromptLog).not.toHaveBeenCalled();
   });
+
+  it('should update movement prefix context on each movement:start event', async () => {
+    const stdoutSpy = vi.spyOn(process.stdout, 'write').mockImplementation(() => true);
+
+    try {
+      await executePiece(makeConfig(), 'repeat-movement-task', '/tmp/project', {
+        projectCwd: '/tmp/project',
+        taskPrefix: 'override-persona-provider',
+        taskColorIndex: 0,
+      });
+
+      const output = stdoutSpy.mock.calls.map((call) => String(call[0])).join('');
+      const normalizedOutput = output.replace(/\x1b\[[0-9;]*m/g, '');
+      expect(normalizedOutput).toContain('[over][implement](1/5)(1) [INFO] [1/5] implement (coder)');
+      expect(normalizedOutput).toContain('[over][implement](2/5)(2) [INFO] [2/5] implement (coder)');
+    } finally {
+      stdoutSpy.mockRestore();
+    }
+  });
+
+  it('should fail fast when taskPrefix is provided without taskColorIndex', async () => {
+    await expect(
+      executePiece(makeConfig(), 'task', '/tmp/project', {
+        projectCwd: '/tmp/project',
+        taskPrefix: 'override-persona-provider',
+      })
+    ).rejects.toThrow('taskPrefix and taskColorIndex must be provided together');
+  });
 });

src/__tests__/pieceResolver.test.ts

@@ -563,3 +563,215 @@ movements:
     expect(result.movementPreviews[0].personaDisplayName).toBe('Custom Agent Name');
   });
 });
+
+describe('getPieceDescription interactiveMode field', () => {
+  let tempDir: string;
+
+  beforeEach(() => {
+    tempDir = mkdtempSync(join(tmpdir(), 'takt-test-interactive-mode-'));
+  });
+
+  afterEach(() => {
+    rmSync(tempDir, { recursive: true, force: true });
+  });
+
+  it('should return interactiveMode when piece defines interactive_mode', () => {
+    const pieceYaml = `name: test-mode
+initial_movement: step1
+max_iterations: 1
+interactive_mode: quiet
+
+movements:
+  - name: step1
+    persona: agent
+    instruction: "Do something"
+`;
+
+    const piecePath = join(tempDir, 'test-mode.yaml');
+    writeFileSync(piecePath, pieceYaml);
+
+    const result = getPieceDescription(piecePath, tempDir);
+
+    expect(result.interactiveMode).toBe('quiet');
+  });
+
+  it('should return undefined interactiveMode when piece omits interactive_mode', () => {
+    const pieceYaml = `name: test-no-mode
+initial_movement: step1
+max_iterations: 1
+
+movements:
+  - name: step1
+    persona: agent
+    instruction: "Do something"
+`;
+
+    const piecePath = join(tempDir, 'test-no-mode.yaml');
+    writeFileSync(piecePath, pieceYaml);
+
+    const result = getPieceDescription(piecePath, tempDir);
+
+    expect(result.interactiveMode).toBeUndefined();
+  });
+
+  it('should return interactiveMode for each valid mode value', () => {
+    for (const mode of ['assistant', 'persona', 'quiet', 'passthrough'] as const) {
+      const pieceYaml = `name: test-${mode}
+initial_movement: step1
+max_iterations: 1
+interactive_mode: ${mode}
+
+movements:
+  - name: step1
+    persona: agent
+    instruction: "Do something"
+`;
+
+      const piecePath = join(tempDir, `test-${mode}.yaml`);
+      writeFileSync(piecePath, pieceYaml);
+
+      const result = getPieceDescription(piecePath, tempDir);
+
+      expect(result.interactiveMode).toBe(mode);
+    }
+  });
+});
+
+describe('getPieceDescription firstMovement field', () => {
+  let tempDir: string;
+
+  beforeEach(() => {
+    tempDir = mkdtempSync(join(tmpdir(), 'takt-test-first-movement-'));
+  });
+
+  afterEach(() => {
+    rmSync(tempDir, { recursive: true, force: true });
+  });
+
+  it('should return firstMovement with inline persona content', () => {
+    const pieceYaml = `name: test-first
+initial_movement: plan
+max_iterations: 1
+
+movements:
+  - name: plan
+    persona: You are a planner.
+    persona_name: Planner
+    instruction: "Plan the task"
+    allowed_tools:
+      - Read
+      - Glob
+`;
+
+    const piecePath = join(tempDir, 'test-first.yaml');
+    writeFileSync(piecePath, pieceYaml);
+
+    const result = getPieceDescription(piecePath, tempDir);
+
+    expect(result.firstMovement).toBeDefined();
+    expect(result.firstMovement!.personaContent).toBe('You are a planner.');
+    expect(result.firstMovement!.personaDisplayName).toBe('Planner');
+    expect(result.firstMovement!.allowedTools).toEqual(['Read', 'Glob']);
+  });
+
+  it('should return firstMovement with persona file content', () => {
+    const personaContent = '# Expert Planner\nYou plan tasks with precision.';
+    const personaPath = join(tempDir, 'planner-persona.md');
+    writeFileSync(personaPath, personaContent);
+
+    const pieceYaml = `name: test-persona-file
+initial_movement: plan
+max_iterations: 1
+
+personas:
+  planner: ./planner-persona.md
+
+movements:
+  - name: plan
+    persona: planner
+    persona_name: Planner
+    instruction: "Plan the task"
+`;
+
+    const piecePath = join(tempDir, 'test-persona-file.yaml');
+    writeFileSync(piecePath, pieceYaml);
+
+    const result = getPieceDescription(piecePath, tempDir);
+
+    expect(result.firstMovement).toBeDefined();
+    expect(result.firstMovement!.personaContent).toBe(personaContent);
+  });
+
+  it('should return undefined firstMovement when initialMovement not found', () => {
+    const pieceYaml = `name: test-missing
+initial_movement: nonexistent
+max_iterations: 1
+
+movements:
+  - name: step1
+    persona: agent
+    instruction: "Do something"
+`;
+
+    const piecePath = join(tempDir, 'test-missing.yaml');
+    writeFileSync(piecePath, pieceYaml);
+
+    const result = getPieceDescription(piecePath, tempDir);
+
+    expect(result.firstMovement).toBeUndefined();
+  });
+
+  it('should return empty allowedTools array when movement has no tools', () => {
+    const pieceYaml = `name: test-no-tools
+initial_movement: step1
+max_iterations: 1
+
+movements:
+  - name: step1
+    persona: agent
+    persona_name: Agent
+    instruction: "Do something"
+`;
+
+    const piecePath = join(tempDir, 'test-no-tools.yaml');
+    writeFileSync(piecePath, pieceYaml);
+
+    const result = getPieceDescription(piecePath, tempDir);
+
+    expect(result.firstMovement).toBeDefined();
+    expect(result.firstMovement!.allowedTools).toEqual([]);
+  });
+
+  it('should fallback to inline persona when personaPath is unreadable', () => {
+    const personaPath = join(tempDir, 'unreadable.md');
+    writeFileSync(personaPath, '# Persona');
+    chmodSync(personaPath, 0o000);
+
+    const pieceYaml = `name: test-fallback
+initial_movement: step1
+max_iterations: 1
+
+personas:
+  myagent: ./unreadable.md
+
+movements:
+  - name: step1
+    persona: myagent
+    persona_name: Agent
+    instruction: "Do something"
+`;
+
+    const piecePath = join(tempDir, 'test-fallback.yaml');
+    writeFileSync(piecePath, pieceYaml);
+
+    try {
+      const result = getPieceDescription(piecePath, tempDir);
+
+      expect(result.firstMovement).toBeDefined();
+      // personaPath is unreadable, so fallback to empty (persona was resolved to a path)
+      expect(result.firstMovement!.personaContent).toBe('');
+    } finally {
+      chmodSync(personaPath, 0o644);
+    }
+  });
+});

src/__tests__/runAllTasks-concurrency.test.ts

@@ -14,6 +14,7 @@ vi.mock('../infra/config/index.js', () => ({
     defaultPiece: 'default',
     logLevel: 'info',
     concurrency: 1,
+    taskPollIntervalMs: 500,
   })),
 }));
 
@@ -142,6 +143,7 @@ describe('runAllTasks concurrency', () => {
         defaultPiece: 'default',
         logLevel: 'info',
         concurrency: 1,
+        taskPollIntervalMs: 500,
       });
     });
 
@@ -182,6 +184,7 @@ describe('runAllTasks concurrency', () => {
         defaultPiece: 'default',
         logLevel: 'info',
         concurrency: 3,
+        taskPollIntervalMs: 500,
       });
     });
 
@@ -209,13 +212,25 @@ describe('runAllTasks concurrency', () => {
         .mockReturnValueOnce([task1, task2, task3])
         .mockReturnValueOnce([]);
 
+      // In parallel mode, task start messages go through TaskPrefixWriter → process.stdout.write
+      const stdoutChunks: string[] = [];
+      const writeSpy = vi.spyOn(process.stdout, 'write').mockImplementation((chunk: unknown) => {
+        stdoutChunks.push(String(chunk));
+        return true;
+      });
+
       // When
       await runAllTasks('/project');
+      writeSpy.mockRestore();
 
-      // Then: Task names displayed
-      expect(mockInfo).toHaveBeenCalledWith('=== Task: task-1 ===');
-      expect(mockInfo).toHaveBeenCalledWith('=== Task: task-2 ===');
-      expect(mockInfo).toHaveBeenCalledWith('=== Task: task-3 ===');
+      // Then: Task names displayed with prefix in stdout
+      const allOutput = stdoutChunks.join('');
+      expect(allOutput).toContain('[task]');
+      expect(allOutput).toContain('=== Task: task-1 ===');
+      expect(allOutput).toContain('[task]');
+      expect(allOutput).toContain('=== Task: task-2 ===');
+      expect(allOutput).toContain('[task]');
+      expect(allOutput).toContain('=== Task: task-3 ===');
       expect(mockStatus).toHaveBeenCalledWith('Total', '3');
     });
 
@@ -245,6 +260,7 @@ describe('runAllTasks concurrency', () => {
         defaultPiece: 'default',
         logLevel: 'info',
         concurrency: 1,
+        taskPollIntervalMs: 500,
       });
 
       const task1 = createTask('task-1');
@@ -277,6 +293,7 @@ describe('runAllTasks concurrency', () => {
         defaultPiece: 'default',
         logLevel: 'info',
         concurrency: 3,
+        taskPollIntervalMs: 500,
       });
       // Return a valid piece config so executeTask reaches executePiece
       mockLoadPieceByIdentifier.mockReturnValue(fakePieceConfig as never);
@@ -323,6 +340,7 @@ describe('runAllTasks concurrency', () => {
         defaultPiece: 'default',
         logLevel: 'info',
         concurrency: 2,
+        taskPollIntervalMs: 500,
       });
 
       const task1 = createTask('fast');
@@ -412,6 +430,7 @@ describe('runAllTasks concurrency', () => {
         defaultPiece: 'default',
         logLevel: 'info',
         concurrency: 1,
+        taskPollIntervalMs: 500,
       });
 
       const task1 = createTask('sequential-task');

src/__tests__/task-prefix-writer.test.ts

@@ -0,0 +1,203 @@
+/**
+ * Tests for TaskPrefixWriter
+ */
+
+import { describe, it, expect, beforeEach } from 'vitest';
+import { TaskPrefixWriter } from '../shared/ui/TaskPrefixWriter.js';
+
+describe('TaskPrefixWriter', () => {
+  let output: string[];
+  let writeFn: (text: string) => void;
+
+  beforeEach(() => {
+    output = [];
+    writeFn = (text: string) => output.push(text);
+  });
+
+  describe('constructor', () => {
+    it('should cycle colors for different colorIndex values', () => {
+      const writer0 = new TaskPrefixWriter({ taskName: 'task-a', colorIndex: 0, writeFn });
+      const writer4 = new TaskPrefixWriter({ taskName: 'task-a', colorIndex: 4, writeFn });
+
+      writer0.writeLine('hello');
+      writer4.writeLine('hello');
+
+      // Both index 0 and 4 should use cyan (\x1b[36m)
+      expect(output[0]).toContain('\x1b[36m');
+      expect(output[1]).toContain('\x1b[36m');
+    });
+
+    it('should assign correct colors in order', () => {
+      const writers = [0, 1, 2, 3].map(
+        (i) => new TaskPrefixWriter({ taskName: `t${i}`, colorIndex: i, writeFn }),
+      );
+
+      writers.forEach((w) => w.writeLine('x'));
+
+      expect(output[0]).toContain('\x1b[36m'); // cyan
+      expect(output[1]).toContain('\x1b[33m'); // yellow
+      expect(output[2]).toContain('\x1b[35m'); // magenta
+      expect(output[3]).toContain('\x1b[32m'); // green
+    });
+  });
+
+  describe('writeLine', () => {
+    it('should output single line with truncated task prefix', () => {
+      const writer = new TaskPrefixWriter({ taskName: 'my-task', colorIndex: 0, writeFn });
+
+      writer.writeLine('Hello World');
+
+      expect(output).toHaveLength(1);
+      expect(output[0]).toContain('[my-t]');
+      expect(output[0]).toContain('Hello World');
+      expect(output[0]).toMatch(/\n$/);
+    });
+
+    it('should output empty line as bare newline', () => {
+      const writer = new TaskPrefixWriter({ taskName: 'my-task', colorIndex: 0, writeFn });
+
+      writer.writeLine('');
+
+      expect(output).toHaveLength(1);
+      expect(output[0]).toBe('\n');
+    });
+
+    it('should split multi-line text and prefix each non-empty line', () => {
+      const writer = new TaskPrefixWriter({ taskName: 'my-task', colorIndex: 0, writeFn });
+
+      writer.writeLine('Line 1\nLine 2\n\nLine 4');
+
+      expect(output).toHaveLength(4);
+      expect(output[0]).toContain('Line 1');
+      expect(output[1]).toContain('Line 2');
+      expect(output[2]).toBe('\n'); // empty line
+      expect(output[3]).toContain('Line 4');
+    });
+
+    it('should strip ANSI codes from input text', () => {
+      const writer = new TaskPrefixWriter({ taskName: 'my-task', colorIndex: 0, writeFn });
+
+      writer.writeLine('\x1b[31mRed Text\x1b[0m');
+
+      expect(output).toHaveLength(1);
+      expect(output[0]).toContain('Red Text');
+      expect(output[0]).not.toContain('\x1b[31m');
+    });
+  });
+
+  describe('writeChunk (line buffering)', () => {
+    it('should buffer partial line and output on newline', () => {
+      const writer = new TaskPrefixWriter({ taskName: 'task-a', colorIndex: 0, writeFn });
+
+      writer.writeChunk('Hello');
+      expect(output).toHaveLength(0);
+
+      writer.writeChunk(' World\n');
+      expect(output).toHaveLength(1);
+      expect(output[0]).toContain('[task]');
+      expect(output[0]).toContain('Hello World');
+    });
+
+    it('should handle multiple lines in single chunk', () => {
+      const writer = new TaskPrefixWriter({ taskName: 'task-a', colorIndex: 0, writeFn });
+
+      writer.writeChunk('Line 1\nLine 2\n');
+
+      expect(output).toHaveLength(2);
+      expect(output[0]).toContain('Line 1');
+      expect(output[1]).toContain('Line 2');
+    });
+
+    it('should output empty line without prefix', () => {
+      const writer = new TaskPrefixWriter({ taskName: 'task-a', colorIndex: 0, writeFn });
+
+      writer.writeChunk('Hello\n\nWorld\n');
+
+      expect(output).toHaveLength(3);
+      expect(output[0]).toContain('Hello');
+      expect(output[1]).toBe('\n');
+      expect(output[2]).toContain('World');
+    });
+
+    it('should keep trailing partial in buffer', () => {
+      const writer = new TaskPrefixWriter({ taskName: 'task-a', colorIndex: 0, writeFn });
+
+      writer.writeChunk('Complete\nPartial');
+
+      expect(output).toHaveLength(1);
+      expect(output[0]).toContain('Complete');
+
+      writer.flush();
+      expect(output).toHaveLength(2);
+      expect(output[1]).toContain('Partial');
+    });
+
+    it('should strip ANSI codes from streamed chunks', () => {
+      const writer = new TaskPrefixWriter({ taskName: 'task-a', colorIndex: 0, writeFn });
+
+      writer.writeChunk('\x1b[31mHello');
+      writer.writeChunk(' World\x1b[0m\n');
+
+      expect(output).toHaveLength(1);
+      expect(output[0]).toContain('Hello World');
+      expect(output[0]).not.toContain('\x1b[31m');
+    });
+  });
+
+  describe('flush', () => {
+    it('should output remaining buffered content with prefix', () => {
+      const writer = new TaskPrefixWriter({ taskName: 'task-a', colorIndex: 0, writeFn });
+
+      writer.writeChunk('partial content');
+      expect(output).toHaveLength(0);
+
+      writer.flush();
+
+      expect(output).toHaveLength(1);
+      expect(output[0]).toContain('[task]');
+      expect(output[0]).toContain('partial content');
+      expect(output[0]).toMatch(/\n$/);
+    });
+
+    it('should not output anything when buffer is empty', () => {
+      const writer = new TaskPrefixWriter({ taskName: 'task-a', colorIndex: 0, writeFn });
+
+      writer.writeChunk('complete line\n');
+      output.length = 0;
+
+      writer.flush();
+      expect(output).toHaveLength(0);
+    });
+
+    it('should clear buffer after flush', () => {
+      const writer = new TaskPrefixWriter({ taskName: 'task-a', colorIndex: 0, writeFn });
+
+      writer.writeChunk('content');
+      writer.flush();
+      output.length = 0;
+
+      writer.flush();
+      expect(output).toHaveLength(0);
+    });
+  });
+
+  describe('setMovementContext', () => {
+    it('should include movement context in prefix after context update', () => {
+      const writer = new TaskPrefixWriter({ taskName: 'override-persona-provider', colorIndex: 0, writeFn });
+
+      writer.setMovementContext({
+        movementName: 'implement',
+        iteration: 4,
+        maxIterations: 30,
+        movementIteration: 2,
+      });
+      writer.writeLine('content');
+
+      expect(output).toHaveLength(1);
+      expect(output[0]).toContain('[over]');
+      expect(output[0]).toContain('[implement](4/30)(2)');
+      expect(output[0]).toContain('content');
+    });
+  });
+
+});

src/__tests__/workerPool.test.ts

@@ -23,6 +23,15 @@ vi.mock('../shared/i18n/index.js', () => ({
   getLabel: vi.fn((key: string) => key),
 }));
 
+vi.mock('../shared/utils/index.js', async (importOriginal) => ({
+  ...(await importOriginal<Record<string, unknown>>()),
+  createLogger: () => ({
+    info: vi.fn(),
+    debug: vi.fn(),
+    error: vi.fn(),
+  }),
+}));
+
 const mockExecuteAndCompleteTask = vi.fn();
 
 vi.mock('../features/tasks/execute/taskExecution.js', () => ({
@@ -34,6 +43,8 @@ import { info } from '../shared/ui/index.js';
 
 const mockInfo = vi.mocked(info);
 
+const TEST_POLL_INTERVAL_MS = 50;
+
 function createTask(name: string): TaskInfo {
   return {
     name,
@@ -68,7 +79,7 @@ describe('runWithWorkerPool', () => {
     const runner = createMockTaskRunner([]);
 
     // When
-    const result = await runWithWorkerPool(runner as never, tasks, 2, '/cwd', 'default');
+    const result = await runWithWorkerPool(runner as never, tasks, 2, '/cwd', 'default', undefined, TEST_POLL_INTERVAL_MS);
 
     // Then
     expect(result).toEqual({ success: 2, fail: 0 });
@@ -85,23 +96,32 @@ describe('runWithWorkerPool', () => {
     const runner = createMockTaskRunner([]);
 
     // When
-    const result = await runWithWorkerPool(runner as never, tasks, 3, '/cwd', 'default');
+    const result = await runWithWorkerPool(runner as never, tasks, 3, '/cwd', 'default', undefined, TEST_POLL_INTERVAL_MS);
 
     // Then
     expect(result).toEqual({ success: 2, fail: 1 });
   });
 
-  it('should display task name for each task', async () => {
+  it('should display task name for each task via prefix writer in parallel mode', async () => {
     // Given
     const tasks = [createTask('alpha'), createTask('beta')];
     const runner = createMockTaskRunner([]);
+    const stdoutChunks: string[] = [];
+    const writeSpy = vi.spyOn(process.stdout, 'write').mockImplementation((chunk: unknown) => {
+      stdoutChunks.push(String(chunk));
+      return true;
+    });
 
     // When
-    await runWithWorkerPool(runner as never, tasks, 2, '/cwd', 'default');
+    await runWithWorkerPool(runner as never, tasks, 2, '/cwd', 'default', undefined, TEST_POLL_INTERVAL_MS);
 
-    // Then
-    expect(mockInfo).toHaveBeenCalledWith('=== Task: alpha ===');
-    expect(mockInfo).toHaveBeenCalledWith('=== Task: beta ===');
+    // Then: Task names appear in prefixed stdout output
+    writeSpy.mockRestore();
+    const allOutput = stdoutChunks.join('');
+    expect(allOutput).toContain('[alph]');
+    expect(allOutput).toContain('=== Task: alpha ===');
+    expect(allOutput).toContain('[beta]');
+    expect(allOutput).toContain('=== Task: beta ===');
   });
 
   it('should pass taskPrefix for parallel execution (concurrency > 1)', async () => {
@@ -110,7 +130,7 @@ describe('runWithWorkerPool', () => {
     const runner = createMockTaskRunner([]);
 
     // When
-    await runWithWorkerPool(runner as never, tasks, 2, '/cwd', 'default');
+    await runWithWorkerPool(runner as never, tasks, 2, '/cwd', 'default', undefined, TEST_POLL_INTERVAL_MS);
 
     // Then
     expect(mockExecuteAndCompleteTask).toHaveBeenCalledTimes(1);
@@ -118,6 +138,7 @@ describe('runWithWorkerPool', () => {
     expect(parallelOpts).toEqual({
       abortSignal: expect.any(AbortSignal),
       taskPrefix: 'my-task',
+      taskColorIndex: 0,
     });
   });
 
@@ -127,7 +148,7 @@ describe('runWithWorkerPool', () => {
     const runner = createMockTaskRunner([]);
 
     // When
-    await runWithWorkerPool(runner as never, tasks, 1, '/cwd', 'default');
+    await runWithWorkerPool(runner as never, tasks, 1, '/cwd', 'default', undefined, TEST_POLL_INTERVAL_MS);
 
     // Then
     expect(mockExecuteAndCompleteTask).toHaveBeenCalledTimes(1);
@@ -135,6 +156,7 @@ describe('runWithWorkerPool', () => {
     expect(parallelOpts).toEqual({
       abortSignal: undefined,
       taskPrefix: undefined,
+      taskColorIndex: undefined,
     });
   });
 
@@ -145,7 +167,7 @@ describe('runWithWorkerPool', () => {
     const runner = createMockTaskRunner([[task2]]);
 
     // When
-    await runWithWorkerPool(runner as never, [task1], 2, '/cwd', 'default');
+    await runWithWorkerPool(runner as never, [task1], 2, '/cwd', 'default', undefined, TEST_POLL_INTERVAL_MS);
 
     // Then
     expect(mockExecuteAndCompleteTask).toHaveBeenCalledTimes(2);
@@ -173,7 +195,7 @@ describe('runWithWorkerPool', () => {
     const runner = createMockTaskRunner([]);
 
     // When
-    await runWithWorkerPool(runner as never, tasks, 2, '/cwd', 'default');
+    await runWithWorkerPool(runner as never, tasks, 2, '/cwd', 'default', undefined, TEST_POLL_INTERVAL_MS);
 
     // Then: Never exceeded concurrency of 2
     expect(maxActive).toBeLessThanOrEqual(2);
@@ -192,7 +214,7 @@ describe('runWithWorkerPool', () => {
     });
 
     // When
-    await runWithWorkerPool(runner as never, tasks, 3, '/cwd', 'default');
+    await runWithWorkerPool(runner as never, tasks, 3, '/cwd', 'default', undefined, TEST_POLL_INTERVAL_MS);
 
     // Then: All tasks received the same AbortSignal
     expect(receivedSignals).toHaveLength(3);
@@ -208,7 +230,7 @@ describe('runWithWorkerPool', () => {
     const runner = createMockTaskRunner([]);
 
     // When
-    const result = await runWithWorkerPool(runner as never, [], 2, '/cwd', 'default');
+    const result = await runWithWorkerPool(runner as never, [], 2, '/cwd', 'default', undefined, TEST_POLL_INTERVAL_MS);
 
     // Then
     expect(result).toEqual({ success: 0, fail: 0 });
@@ -222,9 +244,107 @@ describe('runWithWorkerPool', () => {
     const runner = createMockTaskRunner([]);
 
     // When
-    const result = await runWithWorkerPool(runner as never, tasks, 1, '/cwd', 'default');
+    const result = await runWithWorkerPool(runner as never, tasks, 1, '/cwd', 'default', undefined, TEST_POLL_INTERVAL_MS);
 
     // Then: Treated as failure
     expect(result).toEqual({ success: 0, fail: 1 });
   });
+
+  describe('polling', () => {
+    it('should pick up tasks added during execution via polling', async () => {
+      // Given: 1 initial task running with concurrency=2, a second task appears via poll
+      const task1 = createTask('initial');
+      const task2 = createTask('added-later');
+
+      const executionOrder: string[] = [];
+
+      mockExecuteAndCompleteTask.mockImplementation((task: TaskInfo) => {
+        executionOrder.push(`start:${task.name}`);
+        return new Promise((resolve) => {
+          setTimeout(() => {
+            executionOrder.push(`end:${task.name}`);
+            resolve(true);
+          }, 80);
+        });
+      });
+
+      let claimCallCount = 0;
+      const runner = {
+        getNextTask: vi.fn(() => null),
+        claimNextTasks: vi.fn(() => {
+          claimCallCount++;
+          // Return the new task on the second call (triggered by polling)
+          if (claimCallCount === 2) return [task2];
+          return [];
+        }),
+        completeTask: vi.fn(),
+        failTask: vi.fn(),
+      };
+
+      // When: pollIntervalMs=30 so polling fires before task1 completes (80ms)
+      const result = await runWithWorkerPool(
+        runner as never, [task1], 2, '/cwd', 'default', undefined, 30,
+      );
+
+      // Then: Both tasks were executed
+      expect(result).toEqual({ success: 2, fail: 0 });
+      expect(executionOrder).toContain('start:initial');
+      expect(executionOrder).toContain('start:added-later');
+      // task2 started before task1 ended (picked up by polling, not by task completion)
+      const task2Start = executionOrder.indexOf('start:added-later');
+      const task1End = executionOrder.indexOf('end:initial');
+      expect(task2Start).toBeLessThan(task1End);
+    });
+
+    it('should work correctly with concurrency=1 (sequential behavior preserved)', async () => {
+      // Given: concurrency=1, tasks claimed sequentially
+      const task1 = createTask('seq-1');
+      const task2 = createTask('seq-2');
+
+      const executionOrder: string[] = [];
+      mockExecuteAndCompleteTask.mockImplementation((task: TaskInfo) => {
+        executionOrder.push(`start:${task.name}`);
+        return new Promise((resolve) => {
+          setTimeout(() => {
+            executionOrder.push(`end:${task.name}`);
+            resolve(true);
+          }, 20);
+        });
+      });
+
+      const runner = createMockTaskRunner([[task2]]);
+
+      // When
+      const result = await runWithWorkerPool(
+        runner as never, [task1], 1, '/cwd', 'default', undefined, TEST_POLL_INTERVAL_MS,
+      );
+
+      // Then: Tasks executed sequentially — task2 starts after task1 ends
+      expect(result).toEqual({ success: 2, fail: 0 });
+      const task2Start = executionOrder.indexOf('start:seq-2');
+      const task1End = executionOrder.indexOf('end:seq-1');
+      expect(task2Start).toBeGreaterThan(task1End);
+    });
+
+    it('should not leak poll timer when task completes before poll fires', async () => {
+      // Given: A task that completes in 200ms, poll interval is 5000ms
+      const task1 = createTask('fast-task');
+
+      mockExecuteAndCompleteTask.mockImplementation(() => {
+        return new Promise((resolve) => {
+          setTimeout(() => resolve(true), 200);
+        });
+      });
+
+      const runner = createMockTaskRunner([]);
+
+      // When: Task completes before poll timer fires; cancel() cleans up timer
+      const result = await runWithWorkerPool(
+        runner as never, [task1], 1, '/cwd', 'default', undefined, 5000,
+      );
+
+      // Then: Result is returned without hanging (timer was cleaned up by cancel())
+      expect(result).toEqual({ success: 1, fail: 0 });
+    });
+  });
 });

src/agents/runner.ts

@@ -100,6 +100,7 @@ export class AgentRunner {
   ): ProviderCallOptions {
     return {
       cwd: options.cwd,
+      abortSignal: options.abortSignal,
       sessionId: options.sessionId,
       allowedTools: options.allowedTools ?? agentConfig?.allowedTools,
       mcpServers: options.mcpServers,

src/agents/types.ts

@@ -10,6 +10,7 @@ export type { StreamCallback };
 /** Common options for running agents */
 export interface RunAgentOptions {
   cwd: string;
+  abortSignal?: AbortSignal;
   sessionId?: string;
   model?: string;
   provider?: 'claude' | 'codex' | 'mock';

src/app/cli/routing.ts

@@ -7,10 +7,19 @@
 
 import { info, error } from '../../shared/ui/index.js';
 import { getErrorMessage } from '../../shared/utils/index.js';
+import { getLabel } from '../../shared/i18n/index.js';
 import { fetchIssue, formatIssueAsTask, checkGhCli, parseIssueNumbers, type GitHubIssue } from '../../infra/github/index.js';
 import { selectAndExecuteTask, determinePiece, saveTaskFromInteractive, createIssueFromTask, type SelectAndExecuteOptions } from '../../features/tasks/index.js';
 import { executePipeline } from '../../features/pipeline/index.js';
-import { interactiveMode } from '../../features/interactive/index.js';
+import {
+  interactiveMode,
+  selectInteractiveMode,
+  passthroughMode,
+  quietMode,
+  personaMode,
+  resolveLanguage,
+  type InteractiveModeResult,
+} from '../../features/interactive/index.js';
 import { getPieceDescription, loadGlobalConfig } from '../../infra/config/index.js';
 import { DEFAULT_PIECE_NAME } from '../../shared/constants.js';
 import { program, resolvedCwd, pipelineMode } from './program.js';
@@ -118,16 +127,57 @@ export async function executeDefaultAction(task?: string): Promise<void> {
   }
 
   // All paths below go through interactive mode
+  const globalConfig = loadGlobalConfig();
+  const lang = resolveLanguage(globalConfig.language);
+
   const pieceId = await determinePiece(resolvedCwd, selectOptions.piece);
   if (pieceId === null) {
-    info('Cancelled');
+    info(getLabel('interactive.ui.cancelled', lang));
     return;
   }
 
-  const globalConfig = loadGlobalConfig();
   const previewCount = globalConfig.interactivePreviewMovements;
-  const pieceContext = getPieceDescription(pieceId, resolvedCwd, previewCount);
-  const result = await interactiveMode(resolvedCwd, initialInput, pieceContext);
+  const pieceDesc = getPieceDescription(pieceId, resolvedCwd, previewCount);
+
+  // Mode selection after piece selection
+  const selectedMode = await selectInteractiveMode(lang, pieceDesc.interactiveMode);
+  if (selectedMode === null) {
+    info(getLabel('interactive.ui.cancelled', lang));
+    return;
+  }
+
+  const pieceContext = {
+    name: pieceDesc.name,
+    description: pieceDesc.description,
+    pieceStructure: pieceDesc.pieceStructure,
+    movementPreviews: pieceDesc.movementPreviews,
+  };
+
+  let result: InteractiveModeResult;
+
+  switch (selectedMode) {
+    case 'assistant':
+      result = await interactiveMode(resolvedCwd, initialInput, pieceContext);
+      break;
+
+    case 'passthrough':
+      result = await passthroughMode(lang, initialInput);
+      break;
+
+    case 'quiet':
+      result = await quietMode(resolvedCwd, initialInput, pieceContext);
+      break;
+
+    case 'persona': {
+      if (!pieceDesc.firstMovement) {
+        info(getLabel('interactive.ui.personaFallback', lang));
+        result = await interactiveMode(resolvedCwd, initialInput, pieceContext);
+      } else {
+        result = await personaMode(resolvedCwd, pieceDesc.firstMovement, initialInput, pieceContext);
+      }
+      break;
+    }
+  }
 
   switch (result.action) {
     case 'execute':

src/core/models/global-config.ts

@@ -61,6 +61,8 @@ export interface GlobalConfig {
   bookmarksFile?: string;
   /** Path to piece categories file (default: ~/.takt/preferences/piece-categories.yaml) */
   pieceCategoriesFile?: string;
+  /** Per-persona provider overrides (e.g., { coder: 'codex' }) */
+  personaProviders?: Record<string, 'claude' | 'codex' | 'mock'>;
   /** Branch name generation strategy: 'romaji' (fast, default) or 'ai' (slow) */
   branchNameStrategy?: 'romaji' | 'ai';
   /** Prevent macOS idle sleep during takt execution using caffeinate (default: false) */
@@ -71,6 +73,8 @@ export interface GlobalConfig {
   interactivePreviewMovements?: number;
   /** Number of tasks to run concurrently in takt run (default: 1 = sequential) */
   concurrency: number;
+  /** Polling interval in ms for picking up new tasks during takt run (default: 500, range: 100-5000) */
+  taskPollIntervalMs: number;
 }
 
 /** Project-level configuration */

src/core/models/index.ts

@@ -35,6 +35,9 @@ export * from './config.js';
 // Re-export from schemas.ts
 export * from './schemas.js';
 
+// Re-export from interactive-mode.ts
+export { INTERACTIVE_MODES, DEFAULT_INTERACTIVE_MODE, type InteractiveMode } from './interactive-mode.js';
+
 // Re-export from session.ts (functions only, not types)
 export {
   createSessionState,

src/core/models/interactive-mode.ts

@@ -0,0 +1,18 @@
+/**
+ * Interactive mode variants for conversational task input.
+ *
+ * Defines the four modes available when using interactive mode:
+ * - assistant: Asks clarifying questions before generating instructions (default)
+ * - persona: Uses the first movement's persona for conversation
+ * - quiet: Generates instructions without asking questions (best-effort)
+ * - passthrough: Passes user input directly as task text
+ */
+
+/** Available interactive mode variants */
+export const INTERACTIVE_MODES = ['assistant', 'persona', 'quiet', 'passthrough'] as const;
+
+/** Interactive mode type */
+export type InteractiveMode = typeof INTERACTIVE_MODES[number];
+
+/** Default interactive mode */
+export const DEFAULT_INTERACTIVE_MODE: InteractiveMode = 'assistant';

src/core/models/piece-types.ts

@@ -4,6 +4,7 @@
 
 import type { PermissionMode } from './status.js';
 import type { AgentResponse } from './response.js';
+import type { InteractiveMode } from './interactive-mode.js';
 
 /** Rule-based transition configuration (unified format) */
 export interface PieceRule {
@@ -184,6 +185,8 @@ export interface PieceConfig {
    * instead of prompting the user interactively.
    */
   answerAgent?: string;
+  /** Default interactive mode for this piece (overrides user default) */
+  interactiveMode?: InteractiveMode;
 }
 
 /** Runtime state of a piece execution */

src/core/models/schemas.ts

@@ -7,6 +7,7 @@
 import { z } from 'zod/v4';
 import { DEFAULT_LANGUAGE } from '../../shared/constants.js';
 import { McpServersSchema } from './mcp-schemas.js';
+import { INTERACTIVE_MODES } from './interactive-mode.js';
 
 export { McpServerConfigSchema, McpServersSchema } from './mcp-schemas.js';
 
@@ -218,6 +219,9 @@ export const LoopMonitorSchema = z.object({
   judge: LoopMonitorJudgeSchema,
 });
 
+/** Interactive mode schema for piece-level default */
+export const InteractiveModeSchema = z.enum(INTERACTIVE_MODES);
+
 /** Piece configuration schema - raw YAML format */
 export const PieceConfigRawSchema = z.object({
   name: z.string().min(1),
@@ -237,6 +241,8 @@ export const PieceConfigRawSchema = z.object({
   max_iterations: z.number().int().positive().optional().default(10),
   loop_monitors: z.array(LoopMonitorSchema).optional(),
   answer_agent: z.string().optional(),
+  /** Default interactive mode for this piece (overrides user default) */
+  interactive_mode: InteractiveModeSchema.optional(),
 });
 
 /** Custom agent configuration schema */
@@ -312,6 +318,8 @@ export const GlobalConfigSchema = z.object({
   bookmarks_file: z.string().optional(),
   /** Path to piece categories file (default: ~/.takt/preferences/piece-categories.yaml) */
   piece_categories_file: z.string().optional(),
+  /** Per-persona provider overrides (e.g., { coder: 'codex' }) */
+  persona_providers: z.record(z.string(), z.enum(['claude', 'codex', 'mock'])).optional(),
   /** Branch name generation strategy: 'romaji' (fast, default) or 'ai' (slow) */
   branch_name_strategy: z.enum(['romaji', 'ai']).optional(),
   /** Prevent macOS idle sleep during takt execution using caffeinate (default: false) */
@@ -322,6 +330,8 @@ export const GlobalConfigSchema = z.object({
   interactive_preview_movements: z.number().int().min(0).max(10).optional().default(3),
   /** Number of tasks to run concurrently in takt run (default: 1 = sequential, max: 10) */
   concurrency: z.number().int().min(1).max(10).optional().default(1),
+  /** Polling interval in ms for picking up new tasks during takt run (default: 500, range: 100-5000) */
+  task_poll_interval_ms: z.number().int().min(100).max(5000).optional().default(500),
 });
 
 /** Project config schema */

src/core/piece/engine/OptionsBuilder.ts

@@ -33,8 +33,9 @@ export class OptionsBuilder {
 
     return {
       cwd: this.getCwd(),
+      abortSignal: this.engineOptions.abortSignal,
       personaPath: step.personaPath,
-      provider: step.provider ?? this.engineOptions.provider,
+      provider: step.provider ?? this.engineOptions.personaProviders?.[step.personaDisplayName] ?? this.engineOptions.provider,
       model: step.model ?? this.engineOptions.model,
       permissionMode: step.permissionMode,
       language: this.getLanguage(),

src/core/piece/engine/ParallelRunner.ts

@@ -20,6 +20,7 @@ import { buildSessionKey } from '../session-key.js';
 import type { OptionsBuilder } from './OptionsBuilder.js';
 import type { MovementExecutor } from './MovementExecutor.js';
 import type { PieceEngineOptions, PhaseName } from '../types.js';
+import type { ParallelLoggerOptions } from './parallel-logger.js';
 
 const log = createLogger('parallel-runner');
 
@@ -69,14 +70,7 @@ export class ParallelRunner {
 
     // Create parallel logger for prefixed output (only when streaming is enabled)
     const parallelLogger = this.deps.engineOptions.onStream
-      ? new ParallelLogger({
-          subMovementNames: subMovements.map((s) => s.name),
-          parentOnStream: this.deps.engineOptions.onStream,
-          progressInfo: {
-            iteration: state.iteration,
-            maxIterations,
-          },
-        })
+      ? new ParallelLogger(this.buildParallelLoggerOptions(step.name, movementIteration, subMovements.map((s) => s.name), state.iteration, maxIterations))
       : undefined;
 
     const ruleCtx = {
@@ -202,4 +196,33 @@ export class ParallelRunner {
     return { response: aggregatedResponse, instruction: aggregatedInstruction };
   }
 
+  private buildParallelLoggerOptions(
+    movementName: string,
+    movementIteration: number,
+    subMovementNames: string[],
+    iteration: number,
+    maxIterations: number,
+  ): ParallelLoggerOptions {
+    const options: ParallelLoggerOptions = {
+      subMovementNames,
+      parentOnStream: this.deps.engineOptions.onStream,
+      progressInfo: {
+        iteration,
+        maxIterations,
+      },
+    };
+
+    if (this.deps.engineOptions.taskPrefix != null && this.deps.engineOptions.taskColorIndex != null) {
+      return {
+        ...options,
+        taskLabel: this.deps.engineOptions.taskPrefix,
+        taskColorIndex: this.deps.engineOptions.taskColorIndex,
+        parentMovementName: movementName,
+        movementIteration,
+      };
+    }
+
+    return options;
+  }
+
 }

src/core/piece/engine/PieceEngine.ts

@@ -69,6 +69,7 @@ export class PieceEngine extends EventEmitter {
 
   constructor(config: PieceConfig, cwd: string, task: string, options: PieceEngineOptions) {
     super();
+    this.assertTaskPrefixPair(options.taskPrefix, options.taskColorIndex);
     this.config = config;
     this.projectCwd = options.projectCwd;
     this.cwd = cwd;
@@ -146,6 +147,14 @@ export class PieceEngine extends EventEmitter {
     });
   }
 
+  private assertTaskPrefixPair(taskPrefix: string | undefined, taskColorIndex: number | undefined): void {
+    const hasTaskPrefix = taskPrefix != null;
+    const hasTaskColorIndex = taskColorIndex != null;
+    if (hasTaskPrefix !== hasTaskColorIndex) {
+      throw new Error('taskPrefix and taskColorIndex must be provided together');
+    }
+  }
+
   /** Ensure report directory exists (in cwd, which is clone dir in worktree mode) */
   private ensureReportDirExists(): void {
     const reportDirPath = join(this.cwd, this.reportDir);

src/core/piece/engine/parallel-logger.ts

@@ -30,6 +30,14 @@ export interface ParallelLoggerOptions {
   writeFn?: (text: string) => void;
   /** Progress information for display */
   progressInfo?: ParallelProgressInfo;
+  /** Task label for rich parallel prefix display */
+  taskLabel?: string;
+  /** Task color index for rich parallel prefix display */
+  taskColorIndex?: number;
+  /** Parent movement name for rich parallel prefix display */
+  parentMovementName?: string;
+  /** Parent movement iteration count for rich parallel prefix display */
+  movementIteration?: number;
 }
 
 /**
@@ -47,6 +55,10 @@ export class ParallelLogger {
   private readonly writeFn: (text: string) => void;
   private readonly progressInfo?: ParallelProgressInfo;
   private readonly totalSubMovements: number;
+  private readonly taskLabel?: string;
+  private readonly taskColorIndex?: number;
+  private readonly parentMovementName?: string;
+  private readonly movementIteration?: number;
 
   constructor(options: ParallelLoggerOptions) {
     this.maxNameLength = Math.max(...options.subMovementNames.map((n) => n.length));
@@ -54,6 +66,10 @@ export class ParallelLogger {
     this.writeFn = options.writeFn ?? ((text: string) => process.stdout.write(text));
     this.progressInfo = options.progressInfo;
     this.totalSubMovements = options.subMovementNames.length;
+    this.taskLabel = options.taskLabel ? options.taskLabel.slice(0, 4) : undefined;
+    this.taskColorIndex = options.taskColorIndex;
+    this.parentMovementName = options.parentMovementName;
+    this.movementIteration = options.movementIteration;
 
     for (const name of options.subMovementNames) {
       this.lineBuffers.set(name, '');
@@ -65,6 +81,12 @@ export class ParallelLogger {
    * Format: `\x1b[COLORm[name](iteration/max) step index/total\x1b[0m` + padding spaces
    */
   buildPrefix(name: string, index: number): string {
+    if (this.taskLabel && this.parentMovementName && this.progressInfo && this.movementIteration != null && this.taskColorIndex != null) {
+      const taskColor = COLORS[this.taskColorIndex % COLORS.length];
+      const { iteration, maxIterations } = this.progressInfo;
+      return `${taskColor}[${this.taskLabel}]${RESET}[${this.parentMovementName}][${name}](${iteration}/${maxIterations})(${this.movementIteration}) `;
+    }
+
     const color = COLORS[index % COLORS.length];
     const padding = ' '.repeat(this.maxNameLength - name.length);
 

src/core/piece/types.ts

@@ -153,6 +153,7 @@ export type IterationLimitCallback = (request: IterationLimitRequest) => Promise
 
 /** Options for piece engine */
 export interface PieceEngineOptions {
+  abortSignal?: AbortSignal;
   /** Callback for streaming real-time output */
   onStream?: StreamCallback;
   /** Callback for requesting user input when an agent is blocked */
@@ -177,6 +178,8 @@ export interface PieceEngineOptions {
   language?: Language;
   provider?: ProviderType;
   model?: string;
+  /** Per-persona provider overrides (e.g., { coder: 'codex' }) */
+  personaProviders?: Record<string, ProviderType>;
   /** Enable interactive-only rules and user-input transitions */
   interactive?: boolean;
   /** Rule tag index detector (required for rules evaluation) */
@@ -187,6 +190,10 @@ export interface PieceEngineOptions {
   startMovement?: string;
   /** Retry note explaining why task is being retried */
   retryNote?: string;
+  /** Task name prefix for parallel task execution output */
+  taskPrefix?: string;
+  /** Color index for task prefix (cycled across tasks) */
+  taskColorIndex?: number;
 }
 
 /** Loop detection result */

src/features/interactive/conversationLoop.ts

@@ -0,0 +1,300 @@
+/**
+ * Shared conversation loop for interactive modes (assistant & persona).
+ *
+ * Extracts the common patterns:
+ * - Provider/session initialization
+ * - AI call with retry on stale session
+ * - Session state display/clear
+ * - Conversation loop (slash commands, AI messaging, /go summary)
+ */
+
+import chalk from 'chalk';
+import {
+  loadGlobalConfig,
+  loadPersonaSessions,
+  updatePersonaSession,
+  loadSessionState,
+  clearSessionState,
+} from '../../infra/config/index.js';
+import { isQuietMode } from '../../shared/context.js';
+import { getProvider, type ProviderType } from '../../infra/providers/index.js';
+import { createLogger, getErrorMessage } from '../../shared/utils/index.js';
+import { info, error, blankLine, StreamDisplay } from '../../shared/ui/index.js';
+import { getLabel, getLabelObject } from '../../shared/i18n/index.js';
+import { readMultilineInput } from './lineEditor.js';
+import {
+  type PieceContext,
+  type InteractiveModeResult,
+  type InteractiveUIText,
+  type ConversationMessage,
+  resolveLanguage,
+  buildSummaryPrompt,
+  selectPostSummaryAction,
+  formatSessionStatus,
+} from './interactive.js';
+
+const log = createLogger('conversation-loop');
+
+/** Result from a single AI call */
+export interface CallAIResult {
+  content: string;
+  sessionId?: string;
+  success: boolean;
+}
+
+/** Initialized session context for conversation loops */
+export interface SessionContext {
+  provider: ReturnType<typeof getProvider>;
+  providerType: ProviderType;
+  model: string | undefined;
+  lang: 'en' | 'ja';
+  personaName: string;
+  sessionId: string | undefined;
+}
+
+/**
+ * Initialize provider, session, and language for interactive conversation.
+ */
+export function initializeSession(cwd: string, personaName: string): SessionContext {
+  const globalConfig = loadGlobalConfig();
+  const lang = resolveLanguage(globalConfig.language);
+  if (!globalConfig.provider) {
+    throw new Error('Provider is not configured.');
+  }
+  const providerType = globalConfig.provider as ProviderType;
+  const provider = getProvider(providerType);
+  const model = globalConfig.model as string | undefined;
+  const savedSessions = loadPersonaSessions(cwd, providerType);
+  const sessionId: string | undefined = savedSessions[personaName];
+
+  return { provider, providerType, model, lang, personaName, sessionId };
+}
+
+/**
+ * Display and clear previous session state if present.
+ */
+export function displayAndClearSessionState(cwd: string, lang: 'en' | 'ja'): void {
+  const sessionState = loadSessionState(cwd);
+  if (sessionState) {
+    const statusLabel = formatSessionStatus(sessionState, lang);
+    info(statusLabel);
+    blankLine();
+    clearSessionState(cwd);
+  }
+}
+
+/**
+ * Call AI with automatic retry on stale/invalid session.
+ *
+ * On session failure, clears sessionId and retries once without session.
+ * Updates sessionId and persists it on success.
+ */
+export async function callAIWithRetry(
+  prompt: string,
+  systemPrompt: string,
+  allowedTools: string[],
+  cwd: string,
+  ctx: SessionContext,
+): Promise<{ result: CallAIResult | null; sessionId: string | undefined }> {
+  const display = new StreamDisplay('assistant', isQuietMode());
+  let { sessionId } = ctx;
+
+  try {
+    const agent = ctx.provider.setup({ name: ctx.personaName, systemPrompt });
+    const response = await agent.call(prompt, {
+      cwd,
+      model: ctx.model,
+      sessionId,
+      allowedTools,
+      onStream: display.createHandler(),
+    });
+    display.flush();
+    const success = response.status !== 'blocked';
+
+    if (!success && sessionId) {
+      log.info('Session invalid, retrying without session');
+      sessionId = undefined;
+      const retryDisplay = new StreamDisplay('assistant', isQuietMode());
+      const retryAgent = ctx.provider.setup({ name: ctx.personaName, systemPrompt });
+      const retry = await retryAgent.call(prompt, {
+        cwd,
+        model: ctx.model,
+        sessionId: undefined,
+        allowedTools,
+        onStream: retryDisplay.createHandler(),
+      });
+      retryDisplay.flush();
+      if (retry.sessionId) {
+        sessionId = retry.sessionId;
+        updatePersonaSession(cwd, ctx.personaName, sessionId, ctx.providerType);
+      }
+      return {
+        result: { content: retry.content, sessionId: retry.sessionId, success: retry.status !== 'blocked' },
+        sessionId,
+      };
+    }
+
+    if (response.sessionId) {
+      sessionId = response.sessionId;
+      updatePersonaSession(cwd, ctx.personaName, sessionId, ctx.providerType);
+    }
+    return {
+      result: { content: response.content, sessionId: response.sessionId, success },
+      sessionId,
+    };
+  } catch (e) {
+    const msg = getErrorMessage(e);
+    log.error('AI call failed', { error: msg });
+    error(msg);
+    blankLine();
+    return { result: null, sessionId };
+  }
+}
+
+/** Strategy for customizing conversation loop behavior */
+export interface ConversationStrategy {
+  /** System prompt for AI calls */
+  systemPrompt: string;
+  /** Allowed tools for AI calls */
+  allowedTools: string[];
+  /** Transform user message before sending to AI (e.g., policy injection) */
+  transformPrompt: (userMessage: string) => string;
+  /** Intro message displayed at start */
+  introMessage: string;
+}
+
+/**
+ * Run the shared conversation loop.
+ *
+ * Handles: EOF, /play, /go (summary), /cancel, regular AI messaging.
+ * The Strategy object controls system prompt, tool access, and prompt transformation.
+ */
+export async function runConversationLoop(
+  cwd: string,
+  ctx: SessionContext,
+  strategy: ConversationStrategy,
+  pieceContext: PieceContext | undefined,
+  initialInput: string | undefined,
+): Promise<InteractiveModeResult> {
+  const history: ConversationMessage[] = [];
+  let sessionId = ctx.sessionId;
+  const ui = getLabelObject<InteractiveUIText>('interactive.ui', ctx.lang);
+  const conversationLabel = getLabel('interactive.conversationLabel', ctx.lang);
+  const noTranscript = getLabel('interactive.noTranscript', ctx.lang);
+
+  info(strategy.introMessage);
+  if (sessionId) {
+    info(ui.resume);
+  }
+  blankLine();
+
+  /** Helper: call AI with current session and update session state */
+  async function doCallAI(prompt: string, sysPrompt: string, tools: string[]): Promise<CallAIResult | null> {
+    const { result, sessionId: newSessionId } = await callAIWithRetry(
+      prompt, sysPrompt, tools, cwd, { ...ctx, sessionId },
+    );
+    sessionId = newSessionId;
+    return result;
+  }
+
+  if (initialInput) {
+    history.push({ role: 'user', content: initialInput });
+    log.debug('Processing initial input', { initialInput, sessionId });
+
+    const promptWithTransform = strategy.transformPrompt(initialInput);
+    const result = await doCallAI(promptWithTransform, strategy.systemPrompt, strategy.allowedTools);
+    if (result) {
+      if (!result.success) {
+        error(result.content);
+        blankLine();
+        return { action: 'cancel', task: '' };
+      }
+      history.push({ role: 'assistant', content: result.content });
+      blankLine();
+    } else {
+      history.pop();
+    }
+  }
+
+  while (true) {
+    const input = await readMultilineInput(chalk.green('> '));
+
+    if (input === null) {
+      blankLine();
+      info(ui.cancelled);
+      return { action: 'cancel', task: '' };
+    }
+
+    const trimmed = input.trim();
+
+    if (!trimmed) {
+      continue;
+    }
+
+    if (trimmed.startsWith('/play')) {
+      const task = trimmed.slice(5).trim();
+      if (!task) {
+        info(ui.playNoTask);
+        continue;
+      }
+      log.info('Play command', { task });
+      return { action: 'execute', task };
+    }
+
+    if (trimmed.startsWith('/go')) {
+      const userNote = trimmed.slice(3).trim();
+      let summaryPrompt = buildSummaryPrompt(
+        history, !!sessionId, ctx.lang, noTranscript, conversationLabel, pieceContext,
+      );
+      if (!summaryPrompt) {
+        info(ui.noConversation);
+        continue;
+      }
+      if (userNote) {
+        summaryPrompt = `${summaryPrompt}\n\nUser Note:\n${userNote}`;
+      }
+      const summaryResult = await doCallAI(summaryPrompt, summaryPrompt, strategy.allowedTools);
+      if (!summaryResult) {
+        info(ui.summarizeFailed);
+        continue;
+      }
+      if (!summaryResult.success) {
+        error(summaryResult.content);
+        blankLine();
+        return { action: 'cancel', task: '' };
+      }
+      const task = summaryResult.content.trim();
+      const selectedAction = await selectPostSummaryAction(task, ui.proposed, ui);
+      if (selectedAction === 'continue' || selectedAction === null) {
+        info(ui.continuePrompt);
+        continue;
+      }
+      log.info('Conversation action selected', { action: selectedAction, messageCount: history.length });
+      return { action: selectedAction, task };
+    }
+
+    if (trimmed === '/cancel') {
+      info(ui.cancelled);
+      return { action: 'cancel', task: '' };
+    }
+
+    history.push({ role: 'user', content: trimmed });
+    log.debug('Sending to AI', { messageCount: history.length, sessionId });
+    process.stdin.pause();
+
+    const promptWithTransform = strategy.transformPrompt(trimmed);
+    const result = await doCallAI(promptWithTransform, strategy.systemPrompt, strategy.allowedTools);
+    if (result) {
+      if (!result.success) {
+        error(result.content);
+        blankLine();
+        history.pop();
+        return { action: 'cancel', task: '' };
+      }
+      history.push({ role: 'assistant', content: result.content });
+      blankLine();
+    } else {
+      history.pop();
+    }
+  }
+}

src/features/interactive/index.ts

@@ -4,7 +4,17 @@
 
 export {
   interactiveMode,
+  resolveLanguage,
+  buildSummaryPrompt,
+  selectPostSummaryAction,
+  formatMovementPreviews,
+  formatSessionStatus,
   type PieceContext,
   type InteractiveModeResult,
   type InteractiveModeAction,
 } from './interactive.js';
+
+export { selectInteractiveMode } from './modeSelection.js';
+export { passthroughMode } from './passthroughMode.js';
+export { quietMode } from './quietMode.js';
+export { personaMode } from './personaMode.js';

src/features/interactive/interactive.ts

@@ -10,29 +10,23 @@
  *   /cancel - Cancel and exit
  */
 
-import chalk from 'chalk';
 import type { Language } from '../../core/models/index.js';
 import {
-  loadGlobalConfig,
-  loadPersonaSessions,
-  updatePersonaSession,
-  loadSessionState,
-  clearSessionState,
   type SessionState,
   type MovementPreview,
 } from '../../infra/config/index.js';
-import { isQuietMode } from '../../shared/context.js';
-import { getProvider, type ProviderType } from '../../infra/providers/index.js';
 import { selectOption } from '../../shared/prompt/index.js';
-import { createLogger, getErrorMessage } from '../../shared/utils/index.js';
-import { info, error, blankLine, StreamDisplay } from '../../shared/ui/index.js';
+import { info, blankLine } from '../../shared/ui/index.js';
 import { loadTemplate } from '../../shared/prompts/index.js';
 import { getLabel, getLabelObject } from '../../shared/i18n/index.js';
-import { readMultilineInput } from './lineEditor.js';
-const log = createLogger('interactive');
+import {
+  initializeSession,
+  displayAndClearSessionState,
+  runConversationLoop,
+} from './conversationLoop.js';
 
 /** Shape of interactive UI text */
-interface InteractiveUIText {
+export interface InteractiveUIText {
   intro: string;
   resume: string;
   noConversation: string;
@@ -53,7 +47,7 @@ interface InteractiveUIText {
 /**
  * Format session state for display
  */
-function formatSessionStatus(state: SessionState, lang: 'en' | 'ja'): string {
+export function formatSessionStatus(state: SessionState, lang: 'en' | 'ja'): string {
   const lines: string[] = [];
 
   // Status line
@@ -87,7 +81,7 @@ function formatSessionStatus(state: SessionState, lang: 'en' | 'ja'): string {
   return lines.join('\n');
 }
 
-function resolveLanguage(lang?: Language): 'en' | 'ja' {
+export function resolveLanguage(lang?: Language): 'en' | 'ja' {
   return lang === 'ja' ? 'ja' : 'en';
 }
 
@@ -122,37 +116,11 @@ export function formatMovementPreviews(previews: MovementPreview[], lang: 'en' |
   }).join('\n\n');
 }
 
-function getInteractivePrompts(lang: 'en' | 'ja', pieceContext?: PieceContext) {
-  const hasPreview = !!pieceContext?.movementPreviews?.length;
-  const systemPrompt = loadTemplate('score_interactive_system_prompt', lang, {
-    hasPiecePreview: hasPreview,
-    pieceStructure: pieceContext?.pieceStructure ?? '',
-    movementDetails: hasPreview ? formatMovementPreviews(pieceContext!.movementPreviews!, lang) : '',
-  });
-  const policyContent = loadTemplate('score_interactive_policy', lang, {});
-
-  return {
-    systemPrompt,
-    policyContent,
-    lang,
-    pieceContext,
-    conversationLabel: getLabel('interactive.conversationLabel', lang),
-    noTranscript: getLabel('interactive.noTranscript', lang),
-    ui: getLabelObject<InteractiveUIText>('interactive.ui', lang),
-  };
-}
-
-interface ConversationMessage {
+export interface ConversationMessage {
   role: 'user' | 'assistant';
   content: string;
 }
 
-interface CallAIResult {
-  content: string;
-  sessionId?: string;
-  success: boolean;
-}
-
 /**
  * Build the final task description from conversation history for executeTask.
  */
@@ -167,7 +135,7 @@ function buildTaskFromHistory(history: ConversationMessage[]): string {
  * Renders the complete score_summary_system_prompt template with conversation data.
  * Returns empty string if there is no conversation to summarize.
  */
-function buildSummaryPrompt(
+export function buildSummaryPrompt(
   history: ConversationMessage[],
   hasSession: boolean,
   lang: 'en' | 'ja',
@@ -199,9 +167,9 @@ function buildSummaryPrompt(
   });
 }
 
-type PostSummaryAction = InteractiveModeAction | 'continue';
+export type PostSummaryAction = InteractiveModeAction | 'continue';
 
-async function selectPostSummaryAction(
+export async function selectPostSummaryAction(
   task: string,
   proposedLabel: string,
   ui: InteractiveUIText,
@@ -218,34 +186,6 @@ async function selectPostSummaryAction(
   ]);
 }
 
-/**
- * Call AI with the same pattern as piece execution.
- * The key requirement is passing onStream — the Agent SDK requires
- * includePartialMessages to be true for the async iterator to yield.
- */
-async function callAI(
-  provider: ReturnType<typeof getProvider>,
-  prompt: string,
-  cwd: string,
-  model: string | undefined,
-  sessionId: string | undefined,
-  display: StreamDisplay,
-  systemPrompt: string,
-): Promise<CallAIResult> {
-  const agent = provider.setup({ name: 'interactive', systemPrompt });
-  const response = await agent.call(prompt, {
-    cwd,
-    model,
-    sessionId,
-    allowedTools: ['Read', 'Glob', 'Grep', 'Bash', 'WebSearch', 'WebFetch'],
-    onStream: display.createHandler(),
-  });
-
-  display.flush();
-  const success = response.status !== 'blocked';
-  return { content: response.content, sessionId: response.sessionId, success };
-}
-
 export type InteractiveModeAction = 'execute' | 'save_task' | 'create_issue' | 'cancel';
 
 export interface InteractiveModeResult {
@@ -266,6 +206,8 @@ export interface PieceContext {
   movementPreviews?: MovementPreview[];
 }
 
+export const DEFAULT_INTERACTIVE_TOOLS = ['Read', 'Glob', 'Grep', 'Bash', 'WebSearch', 'WebFetch'];
+
 /**
  * Run the interactive task input mode.
  *
@@ -280,206 +222,37 @@ export async function interactiveMode(
   initialInput?: string,
   pieceContext?: PieceContext,
 ): Promise<InteractiveModeResult> {
-  const globalConfig = loadGlobalConfig();
-  const lang = resolveLanguage(globalConfig.language);
-  const prompts = getInteractivePrompts(lang, pieceContext);
-  if (!globalConfig.provider) {
-    throw new Error('Provider is not configured.');
-  }
-  const providerType = globalConfig.provider as ProviderType;
-  const provider = getProvider(providerType);
-  const model = (globalConfig.model as string | undefined);
+  const ctx = initializeSession(cwd, 'interactive');
 
-  const history: ConversationMessage[] = [];
-  const personaName = 'interactive';
-  const savedSessions = loadPersonaSessions(cwd, providerType);
-  let sessionId: string | undefined = savedSessions[personaName];
+  displayAndClearSessionState(cwd, ctx.lang);
 
-  // Load and display previous task state
-  const sessionState = loadSessionState(cwd);
-  if (sessionState) {
-    const statusLabel = formatSessionStatus(sessionState, lang);
-    info(statusLabel);
-    blankLine();
-    clearSessionState(cwd);
-  }
-
-  info(prompts.ui.intro);
-  if (sessionId) {
-    info(prompts.ui.resume);
-  }
-  blankLine();
-
-  /** Call AI with automatic retry on session error (stale/invalid session ID). */
-  async function callAIWithRetry(prompt: string, systemPrompt: string): Promise<CallAIResult | null> {
-    const display = new StreamDisplay('assistant', isQuietMode());
-    try {
-      const result = await callAI(
-        provider,
-        prompt,
-        cwd,
-        model,
-        sessionId,
-        display,
-        systemPrompt,
-      );
-      // If session failed, clear it and retry without session
-      if (!result.success && sessionId) {
-        log.info('Session invalid, retrying without session');
-        sessionId = undefined;
-        const retryDisplay = new StreamDisplay('assistant', isQuietMode());
-        const retry = await callAI(
-          provider,
-          prompt,
-          cwd,
-          model,
-          undefined,
-          retryDisplay,
-          systemPrompt,
-        );
-        if (retry.sessionId) {
-          sessionId = retry.sessionId;
-          updatePersonaSession(cwd, personaName, sessionId, providerType);
-        }
-        return retry;
-      }
-      if (result.sessionId) {
-        sessionId = result.sessionId;
-        updatePersonaSession(cwd, personaName, sessionId, providerType);
-      }
-      return result;
-    } catch (e) {
-      const msg = getErrorMessage(e);
-      log.error('AI call failed', { error: msg });
-      error(msg);
-      blankLine();
-      return null;
-    }
-  }
+  const hasPreview = !!pieceContext?.movementPreviews?.length;
+  const systemPrompt = loadTemplate('score_interactive_system_prompt', ctx.lang, {
+    hasPiecePreview: hasPreview,
+    pieceStructure: pieceContext?.pieceStructure ?? '',
+    movementDetails: hasPreview ? formatMovementPreviews(pieceContext!.movementPreviews!, ctx.lang) : '',
+  });
+  const policyContent = loadTemplate('score_interactive_policy', ctx.lang, {});
+  const ui = getLabelObject<InteractiveUIText>('interactive.ui', ctx.lang);
 
   /**
    * Inject policy into user message for AI call.
    * Follows the same pattern as piece execution (perform_phase1_message.md).
    */
   function injectPolicy(userMessage: string): string {
-    const policyIntro = lang === 'ja'
+    const policyIntro = ctx.lang === 'ja'
       ? '以下のポリシーは行動規範です。必ず遵守してください。'
       : 'The following policy defines behavioral guidelines. Please follow them.';
-    const reminderLabel = lang === 'ja'
+    const reminderLabel = ctx.lang === 'ja'
       ? '上記の Policy セクションで定義されたポリシー規範を遵守してください。'
       : 'Please follow the policy guidelines defined in the Policy section above.';
-    return `## Policy\n${policyIntro}\n\n${prompts.policyContent}\n\n---\n\n${userMessage}\n\n---\n**Policy Reminder:** ${reminderLabel}`;
+    return `## Policy\n${policyIntro}\n\n${policyContent}\n\n---\n\n${userMessage}\n\n---\n**Policy Reminder:** ${reminderLabel}`;
   }
 
-  // Process initial input if provided (e.g. from `takt a`)
-  if (initialInput) {
-    history.push({ role: 'user', content: initialInput });
-    log.debug('Processing initial input', { initialInput, sessionId });
-
-    const promptWithPolicy = injectPolicy(initialInput);
-    const result = await callAIWithRetry(promptWithPolicy, prompts.systemPrompt);
-    if (result) {
-      if (!result.success) {
-        error(result.content);
-        blankLine();
-        return { action: 'cancel', task: '' };
-      }
-      history.push({ role: 'assistant', content: result.content });
-      blankLine();
-    } else {
-      history.pop();
-    }
-  }
-
-  while (true) {
-    const input = await readMultilineInput(chalk.green('> '));
-
-    // EOF (Ctrl+D)
-    if (input === null) {
-      blankLine();
-      info('Cancelled');
-      return { action: 'cancel', task: '' };
-    }
-
-    const trimmed = input.trim();
-
-    // Empty input — skip
-    if (!trimmed) {
-      continue;
-    }
-
-    // Handle slash commands
-    if (trimmed.startsWith('/play')) {
-      const task = trimmed.slice(5).trim();
-      if (!task) {
-        info(prompts.ui.playNoTask);
-        continue;
-      }
-      log.info('Play command', { task });
-      return { action: 'execute', task };
-    }
-
-    if (trimmed.startsWith('/go')) {
-      const userNote = trimmed.slice(3).trim();
-      let summaryPrompt = buildSummaryPrompt(
-        history,
-        !!sessionId,
-        prompts.lang,
-        prompts.noTranscript,
-        prompts.conversationLabel,
-        prompts.pieceContext,
-      );
-      if (!summaryPrompt) {
-        info(prompts.ui.noConversation);
-        continue;
-      }
-      if (userNote) {
-        summaryPrompt = `${summaryPrompt}\n\nUser Note:\n${userNote}`;
-      }
-      const summaryResult = await callAIWithRetry(summaryPrompt, summaryPrompt);
-      if (!summaryResult) {
-        info(prompts.ui.summarizeFailed);
-        continue;
-      }
-      if (!summaryResult.success) {
-        error(summaryResult.content);
-        blankLine();
-        return { action: 'cancel', task: '' };
-      }
-      const task = summaryResult.content.trim();
-      const selectedAction = await selectPostSummaryAction(task, prompts.ui.proposed, prompts.ui);
-      if (selectedAction === 'continue' || selectedAction === null) {
-        info(prompts.ui.continuePrompt);
-        continue;
-      }
-      log.info('Interactive mode action selected', { action: selectedAction, messageCount: history.length });
-      return { action: selectedAction, task };
-    }
-
-    if (trimmed === '/cancel') {
-      info(prompts.ui.cancelled);
-      return { action: 'cancel', task: '' };
-    }
-
-    // Regular input — send to AI
-    history.push({ role: 'user', content: trimmed });
-
-    log.debug('Sending to AI', { messageCount: history.length, sessionId });
-    process.stdin.pause();
-
-    const promptWithPolicy = injectPolicy(trimmed);
-    const result = await callAIWithRetry(promptWithPolicy, prompts.systemPrompt);
-    if (result) {
-      if (!result.success) {
-        error(result.content);
-        blankLine();
-        history.pop();
-        return { action: 'cancel', task: '' };
-      }
-      history.push({ role: 'assistant', content: result.content });
-      blankLine();
-    } else {
-      history.pop();
-    }
-  }
+  return runConversationLoop(cwd, ctx, {
+    systemPrompt,
+    allowedTools: DEFAULT_INTERACTIVE_TOOLS,
+    transformPrompt: injectPolicy,
+    introMessage: ui.intro,
+  }, pieceContext, initialInput);
 }

src/features/interactive/lineEditor.ts

@@ -250,46 +250,103 @@ export function readMultilineInput(prompt: string): Promise<string | null> {
 
     // --- Buffer position helpers ---
 
-    function getLineStart(): number {
-      const lastNl = buffer.lastIndexOf('\n', cursorPos - 1);
-      return lastNl + 1;
-    }
-
-    function getLineEnd(): number {
-      const nextNl = buffer.indexOf('\n', cursorPos);
-      return nextNl >= 0 ? nextNl : buffer.length;
-    }
-
     function getLineStartAt(pos: number): number {
       const lastNl = buffer.lastIndexOf('\n', pos - 1);
       return lastNl + 1;
     }
 
+    function getLineStart(): number {
+      return getLineStartAt(cursorPos);
+    }
+
     function getLineEndAt(pos: number): number {
       const nextNl = buffer.indexOf('\n', pos);
       return nextNl >= 0 ? nextNl : buffer.length;
     }
 
-    /** Display width from line start to cursor */
-    function getDisplayColumn(): number {
-      return getDisplayWidth(buffer.slice(getLineStart(), cursorPos));
+    function getLineEnd(): number {
+      return getLineEndAt(cursorPos);
     }
 
     const promptWidth = getDisplayWidth(stripAnsi(prompt));
 
-    /** Terminal column (1-based) for a given buffer position */
-    function getTerminalColumn(pos: number): number {
-      const lineStart = getLineStartAt(pos);
-      const col = getDisplayWidth(buffer.slice(lineStart, pos));
-      const isFirstLine = lineStart === 0;
-      return isFirstLine ? promptWidth + col + 1 : col + 1;
+    // --- Display row helpers (soft-wrap awareness) ---
+
+    function getTermWidth(): number {
+      return process.stdout.columns || 80;
     }
 
-    /** Find the buffer position in a line that matches a target display column */
-    function findPositionByDisplayColumn(lineStart: number, lineEnd: number, targetDisplayCol: number): number {
+    /** Buffer position of the display row start that contains `pos` */
+    function getDisplayRowStart(pos: number): number {
+      const logicalStart = getLineStartAt(pos);
+      const termWidth = getTermWidth();
+      const isFirstLogicalLine = logicalStart === 0;
+      let firstRowWidth = isFirstLogicalLine ? termWidth - promptWidth : termWidth;
+      if (firstRowWidth <= 0) firstRowWidth = 1;
+
+      let rowStart = logicalStart;
+      let accumulated = 0;
+      let available = firstRowWidth;
+      let i = logicalStart;
+      for (const ch of buffer.slice(logicalStart, pos)) {
+        const w = getDisplayWidth(ch);
+        if (accumulated + w > available) {
+          rowStart = i;
+          accumulated = w;
+          available = termWidth;
+        } else {
+          accumulated += w;
+          // Row exactly filled — next position starts a new display row
+          if (accumulated === available) {
+            rowStart = i + ch.length;
+            accumulated = 0;
+            available = termWidth;
+          }
+        }
+        i += ch.length;
+      }
+      return rowStart;
+    }
+
+    /** Buffer position of the display row end that contains `pos` */
+    function getDisplayRowEnd(pos: number): number {
+      const logicalEnd = getLineEndAt(pos);
+      const rowStart = getDisplayRowStart(pos);
+      const termWidth = getTermWidth();
+      // The first display row of the first logical line has reduced width
+      const isFirstDisplayRow = rowStart === 0;
+      const available = isFirstDisplayRow ? termWidth - promptWidth : termWidth;
+
+      let accumulated = 0;
+      let i = rowStart;
+      for (const ch of buffer.slice(rowStart, logicalEnd)) {
+        const w = getDisplayWidth(ch);
+        if (accumulated + w > available) return i;
+        accumulated += w;
+        i += ch.length;
+      }
+      return logicalEnd;
+    }
+
+    /** Display column (0-based) within the display row that contains `pos` */
+    function getDisplayRowColumn(pos: number): number {
+      return getDisplayWidth(buffer.slice(getDisplayRowStart(pos), pos));
+    }
+
+    /** Terminal column (1-based) for a given buffer position */
+    function getTerminalColumn(pos: number): number {
+      const displayRowStart = getDisplayRowStart(pos);
+      const col = getDisplayWidth(buffer.slice(displayRowStart, pos));
+      // Only the first display row of the first logical line has the prompt offset
+      const isFirstDisplayRow = displayRowStart === 0;
+      return isFirstDisplayRow ? promptWidth + col + 1 : col + 1;
+    }
+
+    /** Find the buffer position in a range that matches a target display column */
+    function findPositionByDisplayColumn(rangeStart: number, rangeEnd: number, targetDisplayCol: number): number {
       let displayCol = 0;
-      let pos = lineStart;
-      for (const ch of buffer.slice(lineStart, lineEnd)) {
+      let pos = rangeStart;
+      for (const ch of buffer.slice(rangeStart, rangeEnd)) {
         const w = getDisplayWidth(ch);
         if (displayCol + w > targetDisplayCol) break;
         displayCol += w;
@@ -322,23 +379,77 @@ export function readMultilineInput(prompt: string): Promise<string | null> {
 
     // --- Cursor movement ---
 
-    function moveCursorToLineStart(): void {
-      const displayOffset = getDisplayColumn();
+    function moveCursorToDisplayRowStart(): void {
+      const displayRowStart = getDisplayRowStart(cursorPos);
+      const displayOffset = getDisplayRowColumn(cursorPos);
       if (displayOffset > 0) {
-        cursorPos = getLineStart();
+        cursorPos = displayRowStart;
         process.stdout.write(`\x1B[${displayOffset}D`);
       }
     }
 
-    function moveCursorToLineEnd(): void {
-      const lineEnd = getLineEnd();
-      const displayOffset = getDisplayWidth(buffer.slice(cursorPos, lineEnd));
+    function moveCursorToDisplayRowEnd(): void {
+      const displayRowEnd = getDisplayRowEnd(cursorPos);
+      const displayOffset = getDisplayWidth(buffer.slice(cursorPos, displayRowEnd));
       if (displayOffset > 0) {
-        cursorPos = lineEnd;
+        cursorPos = displayRowEnd;
         process.stdout.write(`\x1B[${displayOffset}C`);
       }
     }
 
+    /** Move cursor to a target display row, positioning at the given display column */
+    function moveCursorToDisplayRow(
+      targetRowStart: number,
+      targetRowEnd: number,
+      displayCol: number,
+      direction: 'A' | 'B',
+    ): void {
+      cursorPos = findPositionByDisplayColumn(targetRowStart, targetRowEnd, displayCol);
+      const termCol = getTerminalColumn(cursorPos);
+      process.stdout.write(`\x1B[${direction}`);
+      process.stdout.write(`\x1B[${termCol}G`);
+    }
+
+    /** Count how many display rows lie between two buffer positions in the same logical line */
+    function countDisplayRowsBetween(from: number, to: number): number {
+      if (from === to) return 0;
+      const start = Math.min(from, to);
+      const end = Math.max(from, to);
+      let count = 0;
+      let pos = start;
+      while (pos < end) {
+        const nextRowStart = getDisplayRowEnd(pos);
+        if (nextRowStart >= end) break;
+        pos = nextRowStart;
+        count++;
+      }
+      return count;
+    }
+
+    function moveCursorToLogicalLineStart(): void {
+      const lineStart = getLineStart();
+      if (cursorPos === lineStart) return;
+      const rowDiff = countDisplayRowsBetween(lineStart, cursorPos);
+      cursorPos = lineStart;
+      if (rowDiff > 0) {
+        process.stdout.write(`\x1B[${rowDiff}A`);
+      }
+      const termCol = getTerminalColumn(cursorPos);
+      process.stdout.write(`\x1B[${termCol}G`);
+    }
+
+    function moveCursorToLogicalLineEnd(): void {
+      const lineEnd = getLineEnd();
+      if (cursorPos === lineEnd) return;
+      const rowDiff = countDisplayRowsBetween(cursorPos, lineEnd);
+      cursorPos = lineEnd;
+      if (rowDiff > 0) {
+        process.stdout.write(`\x1B[${rowDiff}B`);
+      }
+      const termCol = getTerminalColumn(cursorPos);
+      process.stdout.write(`\x1B[${termCol}G`);
+    }
+
     // --- Buffer editing ---
 
     function insertAt(pos: number, text: string): void {
@@ -461,27 +572,40 @@ export function readMultilineInput(prompt: string): Promise<string | null> {
           },
           onArrowUp() {
             if (state !== 'normal') return;
-            const lineStart = getLineStart();
-            if (lineStart === 0) return;
-            const displayCol = getDisplayColumn();
-            const prevLineStart = getLineStartAt(lineStart - 1);
-            const prevLineEnd = lineStart - 1;
-            cursorPos = findPositionByDisplayColumn(prevLineStart, prevLineEnd, displayCol);
-            const termCol = getTerminalColumn(cursorPos);
-            process.stdout.write('\x1B[A');
-            process.stdout.write(`\x1B[${termCol}G`);
+            const logicalLineStart = getLineStart();
+            const displayRowStart = getDisplayRowStart(cursorPos);
+            const displayCol = getDisplayRowColumn(cursorPos);
+
+            if (displayRowStart > logicalLineStart) {
+              // Move to previous display row within the same logical line
+              const prevRowStart = getDisplayRowStart(displayRowStart - 1);
+              const prevRowEnd = getDisplayRowEnd(displayRowStart - 1);
+              moveCursorToDisplayRow(prevRowStart, prevRowEnd, displayCol, 'A');
+            } else if (logicalLineStart > 0) {
+              // Move to the last display row of the previous logical line
+              const prevLogicalLineEnd = logicalLineStart - 1;
+              const prevRowStart = getDisplayRowStart(prevLogicalLineEnd);
+              const prevRowEnd = getDisplayRowEnd(prevLogicalLineEnd);
+              moveCursorToDisplayRow(prevRowStart, prevRowEnd, displayCol, 'A');
+            }
           },
           onArrowDown() {
             if (state !== 'normal') return;
-            const lineEnd = getLineEnd();
-            if (lineEnd >= buffer.length) return;
-            const displayCol = getDisplayColumn();
-            const nextLineStart = lineEnd + 1;
-            const nextLineEnd = getLineEndAt(nextLineStart);
-            cursorPos = findPositionByDisplayColumn(nextLineStart, nextLineEnd, displayCol);
-            const termCol = getTerminalColumn(cursorPos);
-            process.stdout.write('\x1B[B');
-            process.stdout.write(`\x1B[${termCol}G`);
+            const logicalLineEnd = getLineEnd();
+            const displayRowEnd = getDisplayRowEnd(cursorPos);
+            const displayCol = getDisplayRowColumn(cursorPos);
+
+            if (displayRowEnd < logicalLineEnd) {
+              // Move to next display row within the same logical line
+              const nextRowStart = displayRowEnd;
+              const nextRowEnd = getDisplayRowEnd(displayRowEnd);
+              moveCursorToDisplayRow(nextRowStart, nextRowEnd, displayCol, 'B');
+            } else if (logicalLineEnd < buffer.length) {
+              // Move to the first display row of the next logical line
+              const nextLineStart = logicalLineEnd + 1;
+              const nextRowEnd = getDisplayRowEnd(nextLineStart);
+              moveCursorToDisplayRow(nextLineStart, nextRowEnd, displayCol, 'B');
+            }
           },
           onWordLeft() {
             if (state !== 'normal') return;
@@ -507,11 +631,11 @@ export function readMultilineInput(prompt: string): Promise<string | null> {
           },
           onHome() {
             if (state !== 'normal') return;
-            moveCursorToLineStart();
+            moveCursorToLogicalLineStart();
           },
           onEnd() {
             if (state !== 'normal') return;
-            moveCursorToLineEnd();
+            moveCursorToLogicalLineEnd();
           },
           onChar(ch: string) {
             if (state === 'paste') {
@@ -543,8 +667,8 @@ export function readMultilineInput(prompt: string): Promise<string | null> {
             }
             // Editing
             if (ch === '\x7F' || ch === '\x08') { deleteCharBefore(); return; }
-            if (ch === '\x01') { moveCursorToLineStart(); return; }
-            if (ch === '\x05') { moveCursorToLineEnd(); return; }
+            if (ch === '\x01') { moveCursorToDisplayRowStart(); return; }
+            if (ch === '\x05') { moveCursorToDisplayRowEnd(); return; }
             if (ch === '\x0B') { deleteToLineEnd(); return; }
             if (ch === '\x15') { deleteToLineStart(); return; }
             if (ch === '\x17') { deleteWord(); return; }

src/features/interactive/modeSelection.ts

@@ -0,0 +1,35 @@
+/**
+ * Interactive mode selection UI.
+ *
+ * Presents the four interactive mode options after piece selection
+ * and returns the user's choice.
+ */
+
+import type { InteractiveMode } from '../../core/models/index.js';
+import { DEFAULT_INTERACTIVE_MODE, INTERACTIVE_MODES } from '../../core/models/index.js';
+import { selectOptionWithDefault } from '../../shared/prompt/index.js';
+import { getLabel } from '../../shared/i18n/index.js';
+
+/**
+ * Prompt the user to select an interactive mode.
+ *
+ * @param lang - Display language
+ * @param pieceDefault - Piece-level default mode (overrides user default)
+ * @returns Selected mode, or null if cancelled
+ */
+export async function selectInteractiveMode(
+  lang: 'en' | 'ja',
+  pieceDefault?: InteractiveMode,
+): Promise<InteractiveMode | null> {
+  const defaultMode = pieceDefault ?? DEFAULT_INTERACTIVE_MODE;
+
+  const options: { label: string; value: InteractiveMode; description: string }[] = INTERACTIVE_MODES.map((mode) => ({
+    label: getLabel(`interactive.modeSelection.${mode}`, lang),
+    value: mode,
+    description: getLabel(`interactive.modeSelection.${mode}Description`, lang),
+  }));
+
+  const prompt = getLabel('interactive.modeSelection.prompt', lang);
+
+  return selectOptionWithDefault<InteractiveMode>(prompt, options, defaultMode);
+}

src/features/interactive/passthroughMode.ts

@@ -0,0 +1,50 @@
+/**
+ * Passthrough interactive mode.
+ *
+ * Passes user input directly as the task string without any
+ * AI-assisted instruction generation or system prompt injection.
+ */
+
+import chalk from 'chalk';
+import { info, blankLine } from '../../shared/ui/index.js';
+import { getLabel } from '../../shared/i18n/index.js';
+import { readMultilineInput } from './lineEditor.js';
+import type { InteractiveModeResult } from './interactive.js';
+
+/**
+ * Run passthrough mode: collect user input and return it as-is.
+ *
+ * If initialInput is provided, it is used directly as the task.
+ * Otherwise, prompts the user for input.
+ *
+ * @param lang - Display language
+ * @param initialInput - Pre-filled input (e.g., from issue reference)
+ * @returns Result with the raw user input as task
+ */
+export async function passthroughMode(
+  lang: 'en' | 'ja',
+  initialInput?: string,
+): Promise<InteractiveModeResult> {
+  if (initialInput) {
+    return { action: 'execute', task: initialInput };
+  }
+
+  info(getLabel('interactive.ui.intro', lang));
+  blankLine();
+
+  const input = await readMultilineInput(chalk.green('> '));
+
+  if (input === null) {
+    blankLine();
+    info(getLabel('interactive.ui.cancelled', lang));
+    return { action: 'cancel', task: '' };
+  }
+
+  const trimmed = input.trim();
+  if (!trimmed) {
+    info(getLabel('interactive.ui.cancelled', lang));
+    return { action: 'cancel', task: '' };
+  }
+
+  return { action: 'execute', task: trimmed };
+}

src/features/interactive/personaMode.ts

@@ -0,0 +1,58 @@
+/**
+ * Persona interactive mode.
+ *
+ * Uses the first movement's persona and tools for the interactive
+ * conversation. The persona acts as the conversational agent,
+ * performing code exploration and analysis while discussing the task.
+ * The conversation result is passed as the task to the piece.
+ */
+
+import type { FirstMovementInfo } from '../../infra/config/index.js';
+import { getLabel } from '../../shared/i18n/index.js';
+import {
+  type PieceContext,
+  type InteractiveModeResult,
+  DEFAULT_INTERACTIVE_TOOLS,
+} from './interactive.js';
+import {
+  initializeSession,
+  displayAndClearSessionState,
+  runConversationLoop,
+} from './conversationLoop.js';
+
+/**
+ * Run persona mode: converse as the first movement's persona.
+ *
+ * The persona's system prompt is used for all AI calls.
+ * The first movement's allowed tools are made available.
+ * After the conversation, the result is summarized as a task.
+ *
+ * @param cwd - Working directory
+ * @param firstMovement - First movement's persona and tool info
+ * @param initialInput - Pre-filled input
+ * @param pieceContext - Piece context for summary generation
+ * @returns Result with conversation-derived task
+ */
+export async function personaMode(
+  cwd: string,
+  firstMovement: FirstMovementInfo,
+  initialInput?: string,
+  pieceContext?: PieceContext,
+): Promise<InteractiveModeResult> {
+  const ctx = initializeSession(cwd, 'persona-interactive');
+
+  displayAndClearSessionState(cwd, ctx.lang);
+
+  const allowedTools = firstMovement.allowedTools.length > 0
+    ? firstMovement.allowedTools
+    : DEFAULT_INTERACTIVE_TOOLS;
+
+  const introMessage = `${getLabel('interactive.ui.intro', ctx.lang)} [${firstMovement.personaDisplayName}]`;
+
+  return runConversationLoop(cwd, ctx, {
+    systemPrompt: firstMovement.personaContent,
+    allowedTools,
+    transformPrompt: (msg) => msg,
+    introMessage,
+  }, pieceContext, initialInput);
+}

src/features/interactive/quietMode.ts

@@ -0,0 +1,111 @@
+/**
+ * Quiet interactive mode.
+ *
+ * Generates task instructions without asking clarifying questions.
+ * Uses the same summarization logic as assistant mode but skips
+ * the conversational loop — goes directly to summary generation.
+ */
+
+import chalk from 'chalk';
+import { createLogger } from '../../shared/utils/index.js';
+import { info, error, blankLine } from '../../shared/ui/index.js';
+import { getLabel, getLabelObject } from '../../shared/i18n/index.js';
+import { readMultilineInput } from './lineEditor.js';
+import {
+  type PieceContext,
+  type InteractiveModeResult,
+  type InteractiveUIText,
+  type ConversationMessage,
+  DEFAULT_INTERACTIVE_TOOLS,
+  buildSummaryPrompt,
+  selectPostSummaryAction,
+} from './interactive.js';
+import {
+  initializeSession,
+  callAIWithRetry,
+} from './conversationLoop.js';
+
+const log = createLogger('quiet-mode');
+
+/**
+ * Run quiet mode: collect user input and generate instructions without questions.
+ *
+ * Flow:
+ * 1. If initialInput is provided, use it; otherwise prompt for input
+ * 2. Build summary prompt from the user input
+ * 3. Call AI to generate task instructions (best-effort, no questions)
+ * 4. Present the result and let user choose action
+ *
+ * @param cwd - Working directory
+ * @param initialInput - Pre-filled input (e.g., from issue reference)
+ * @param pieceContext - Piece context for template rendering
+ * @returns Result with generated task instructions
+ */
+export async function quietMode(
+  cwd: string,
+  initialInput?: string,
+  pieceContext?: PieceContext,
+): Promise<InteractiveModeResult> {
+  const ctx = initializeSession(cwd, 'interactive');
+
+  let userInput = initialInput;
+
+  if (!userInput) {
+    info(getLabel('interactive.ui.intro', ctx.lang));
+    blankLine();
+
+    const input = await readMultilineInput(chalk.green('> '));
+    if (input === null) {
+      blankLine();
+      info(getLabel('interactive.ui.cancelled', ctx.lang));
+      return { action: 'cancel', task: '' };
+    }
+    const trimmed = input.trim();
+    if (!trimmed) {
+      info(getLabel('interactive.ui.cancelled', ctx.lang));
+      return { action: 'cancel', task: '' };
+    }
+    userInput = trimmed;
+  }
+
+  const history: ConversationMessage[] = [
+    { role: 'user', content: userInput },
+  ];
+
+  const conversationLabel = getLabel('interactive.conversationLabel', ctx.lang);
+  const noTranscript = getLabel('interactive.noTranscript', ctx.lang);
+
+  const summaryPrompt = buildSummaryPrompt(
+    history, !!ctx.sessionId, ctx.lang, noTranscript, conversationLabel, pieceContext,
+  );
+
+  if (!summaryPrompt) {
+    info(getLabel('interactive.ui.noConversation', ctx.lang));
+    return { action: 'cancel', task: '' };
+  }
+
+  const { result } = await callAIWithRetry(
+    summaryPrompt, summaryPrompt, DEFAULT_INTERACTIVE_TOOLS, cwd, ctx,
+  );
+
+  if (!result) {
+    return { action: 'cancel', task: '' };
+  }
+
+  if (!result.success) {
+    error(result.content);
+    blankLine();
+    return { action: 'cancel', task: '' };
+  }
+
+  const task = result.content.trim();
+  const ui = getLabelObject<InteractiveUIText>('interactive.ui', ctx.lang);
+
+  const selectedAction = await selectPostSummaryAction(task, ui.proposed, ui);
+  if (selectedAction === 'continue' || selectedAction === null) {
+    return { action: 'cancel', task: '' };
+  }
+
+  log.info('Quiet mode action selected', { action: selectedAction });
+  return { action: selectedAction, task };
+}

src/features/tasks/execute/parallelExecution.ts

@@ -5,19 +5,75 @@
  * available task as soon as it finishes the current one, maximizing slot
  * utilization. Works for both sequential (concurrency=1) and parallel
  * (concurrency>1) execution through the same code path.
+ *
+ * Polls for newly added tasks at a configurable interval so that tasks
+ * added to .takt/tasks/ during execution are picked up without waiting
+ * for an active task to complete.
  */
 
 import type { TaskRunner, TaskInfo } from '../../../infra/task/index.js';
 import { info, blankLine } from '../../../shared/ui/index.js';
+import { TaskPrefixWriter } from '../../../shared/ui/TaskPrefixWriter.js';
+import { createLogger } from '../../../shared/utils/index.js';
 import { executeAndCompleteTask } from './taskExecution.js';
 import { installSigIntHandler } from './sigintHandler.js';
 import type { TaskExecutionOptions } from './types.js';
 
+const log = createLogger('worker-pool');
+
 export interface WorkerPoolResult {
   success: number;
   fail: number;
 }
 
+type RaceResult =
+  | { type: 'completion'; promise: Promise<boolean>; result: boolean }
+  | { type: 'poll' };
+
+interface PollTimer {
+  promise: Promise<RaceResult>;
+  cancel: () => void;
+}
+
+function createPollTimer(intervalMs: number, signal: AbortSignal): PollTimer {
+  let timeoutId: ReturnType<typeof setTimeout> | undefined;
+  let onAbort: (() => void) | undefined;
+
+  const promise = new Promise<RaceResult>((resolve) => {
+    if (signal.aborted) {
+      resolve({ type: 'poll' });
+      return;
+    }
+
+    onAbort = () => {
+      if (timeoutId !== undefined) clearTimeout(timeoutId);
+      resolve({ type: 'poll' });
+    };
+
+    timeoutId = setTimeout(() => {
+      signal.removeEventListener('abort', onAbort!);
+      onAbort = undefined;
+      resolve({ type: 'poll' });
+    }, intervalMs);
+
+    signal.addEventListener('abort', onAbort, { once: true });
+  });
+
+  return {
+    promise,
+    cancel: () => {
+      if (timeoutId !== undefined) {
+        clearTimeout(timeoutId);
+        timeoutId = undefined;
+      }
+      if (onAbort) {
+        signal.removeEventListener('abort', onAbort);
+        onAbort = undefined;
+      }
+    },
+  };
+}
+
 /**
  * Run tasks using a worker pool with the given concurrency.
  *
@@ -25,9 +81,10 @@ export interface WorkerPoolResult {
  * 1. Create a shared AbortController
  * 2. Maintain a queue of pending tasks and a set of active promises
  * 3. Fill available slots from the queue
- * 4. Wait for any active task to complete (Promise.race)
- * 5. Record result, fill freed slot from queue
- * 6. Repeat until queue is empty and all active tasks complete
+ * 4. Wait for any active task to complete OR a poll timer to fire (Promise.race)
+ * 5. On task completion: record result
+ * 6. On poll tick or completion: claim new tasks and fill freed slots
+ * 7. Repeat until queue is empty and all active tasks complete
  */
 export async function runWithWorkerPool(
   taskRunner: TaskRunner,
@@ -35,7 +92,8 @@ export async function runWithWorkerPool(
   concurrency: number,
   cwd: string,
   pieceName: string,
-  options?: TaskExecutionOptions,
+  options: TaskExecutionOptions | undefined,
+  pollIntervalMs: number,
 ): Promise<WorkerPoolResult> {
   const abortController = new AbortController();
   const { cleanup } = installSigIntHandler(() => abortController.abort());
@@ -45,6 +103,7 @@ export async function runWithWorkerPool(
 
   const queue = [...initialTasks];
   const active = new Map<Promise<boolean>, TaskInfo>();
+  const colorCounter = { value: 0 };
 
   try {
     while (queue.length > 0 || active.size > 0) {
@@ -52,33 +111,50 @@ export async function runWithWorkerPool(
         break;
       }
 
-      fillSlots(queue, active, concurrency, taskRunner, cwd, pieceName, options, abortController);
+      fillSlots(queue, active, concurrency, taskRunner, cwd, pieceName, options, abortController, colorCounter);
 
       if (active.size === 0) {
         break;
       }
 
-      const settled = await Promise.race(
-        [...active.keys()].map((p) => p.then(
-          (result) => ({ promise: p, result }),
-          () => ({ promise: p, result: false }),
-        )),
+      const pollTimer = createPollTimer(pollIntervalMs, abortController.signal);
+
+      const completionPromises: Promise<RaceResult>[] = [...active.keys()].map((p) =>
+        p.then(
+          (result): RaceResult => ({ type: 'completion', promise: p, result }),
+          (): RaceResult => ({ type: 'completion', promise: p, result: false }),
+        ),
       );
 
-      const task = active.get(settled.promise);
-      active.delete(settled.promise);
+      const settled = await Promise.race([...completionPromises, pollTimer.promise]);
 
-      if (task) {
-        if (settled.result) {
-          successCount++;
-        } else {
-          failCount++;
+      pollTimer.cancel();
+
+      if (settled.type === 'completion') {
+        const task = active.get(settled.promise);
+        active.delete(settled.promise);
+
+        if (task) {
+          if (settled.result) {
+            successCount++;
+          } else {
+            failCount++;
+          }
         }
       }
 
-      if (!abortController.signal.aborted && queue.length === 0) {
-        const nextTasks = taskRunner.claimNextTasks(concurrency - active.size);
-        queue.push(...nextTasks);
+      if (!abortController.signal.aborted) {
+        const freeSlots = concurrency - active.size;
+        if (freeSlots > 0) {
+          const newTasks = taskRunner.claimNextTasks(freeSlots);
+          log.debug('poll_tick', { active: active.size, queued: queue.length, freeSlots });
+          if (newTasks.length > 0) {
+            log.debug('poll_new_tasks', { count: newTasks.length });
+            queue.push(...newTasks);
+          } else {
+            log.debug('no_new_tasks');
+          }
+        }
       }
     }
   } finally {
@@ -97,17 +173,25 @@ function fillSlots(
   pieceName: string,
   options: TaskExecutionOptions | undefined,
   abortController: AbortController,
+  colorCounter: { value: number },
 ): void {
   while (active.size < concurrency && queue.length > 0) {
     const task = queue.shift()!;
     const isParallel = concurrency > 1;
+    const colorIndex = colorCounter.value++;
 
-    blankLine();
-    info(`=== Task: ${task.name} ===`);
+    if (isParallel) {
+      const writer = new TaskPrefixWriter({ taskName: task.name, colorIndex });
+      writer.writeLine(`=== Task: ${task.name} ===`);
+    } else {
+      blankLine();
+      info(`=== Task: ${task.name} ===`);
+    }
 
     const promise = executeAndCompleteTask(task, taskRunner, cwd, pieceName, options, {
       abortSignal: isParallel ? abortController.signal : undefined,
       taskPrefix: isParallel ? task.name : undefined,
+      taskColorIndex: isParallel ? colorIndex : undefined,
     });
     active.set(promise, task);
   }

src/features/tasks/execute/pieceExecution.ts

@@ -21,15 +21,16 @@ import {
 } from '../../../infra/config/index.js';
 import { isQuietMode } from '../../../shared/context.js';
 import {
-  header,
-  info,
-  warn,
-  error,
-  success,
-  status,
-  blankLine,
+  header as rawHeader,
+  info as rawInfo,
+  warn as rawWarn,
+  error as rawError,
+  success as rawSuccess,
+  status as rawStatus,
+  blankLine as rawBlankLine,
   StreamDisplay,
 } from '../../../shared/ui/index.js';
+import { TaskPrefixWriter } from '../../../shared/ui/TaskPrefixWriter.js';
 import {
   generateSessionId,
   createSessionLog,
@@ -62,6 +63,91 @@ import { installSigIntHandler } from './sigintHandler.js';
 
 const log = createLogger('piece');
 
+/**
+ * Output facade — routes through TaskPrefixWriter when task prefix is active,
+ * or falls through to the raw module functions for single-task execution.
+ */
+interface OutputFns {
+  header: (title: string) => void;
+  info: (message: string) => void;
+  warn: (message: string) => void;
+  error: (message: string) => void;
+  success: (message: string) => void;
+  status: (label: string, value: string, color?: 'green' | 'yellow' | 'red') => void;
+  blankLine: () => void;
+  logLine: (text: string) => void;
+}
+
+function assertTaskPrefixPair(
+  taskPrefix: string | undefined,
+  taskColorIndex: number | undefined
+): void {
+  const hasTaskPrefix = taskPrefix != null;
+  const hasTaskColorIndex = taskColorIndex != null;
+  if (hasTaskPrefix !== hasTaskColorIndex) {
+    throw new Error('taskPrefix and taskColorIndex must be provided together');
+  }
+}
+
+function createOutputFns(prefixWriter: TaskPrefixWriter | undefined): OutputFns {
+  if (!prefixWriter) {
+    return {
+      header: rawHeader,
+      info: rawInfo,
+      warn: rawWarn,
+      error: rawError,
+      success: rawSuccess,
+      status: rawStatus,
+      blankLine: rawBlankLine,
+      logLine: (text: string) => console.log(text),
+    };
+  }
+  return {
+    header: (title: string) => prefixWriter.writeLine(`=== ${title} ===`),
+    info: (message: string) => prefixWriter.writeLine(`[INFO] ${message}`),
+    warn: (message: string) => prefixWriter.writeLine(`[WARN] ${message}`),
+    error: (message: string) => prefixWriter.writeLine(`[ERROR] ${message}`),
+    success: (message: string) => prefixWriter.writeLine(message),
+    status: (label: string, value: string) => prefixWriter.writeLine(`${label}: ${value}`),
+    blankLine: () => prefixWriter.writeLine(''),
+    logLine: (text: string) => prefixWriter.writeLine(text),
+  };
+}
+
+/**
+ * Create a stream handler that routes all stream events through TaskPrefixWriter.
+ * Text and tool_output are line-buffered; block events are output per-line with prefix.
+ */
+function createPrefixedStreamHandler(
+  writer: TaskPrefixWriter,
+): (event: Parameters<ReturnType<StreamDisplay['createHandler']>>[0]) => void {
+  return (event) => {
+    switch (event.type) {
+      case 'text':
+        writer.writeChunk(event.data.text);
+        break;
+      case 'tool_use':
+        writer.writeLine(`[tool] ${event.data.tool}`);
+        break;
+      case 'tool_result': {
+        const label = event.data.isError ? '✗' : '✓';
+        writer.writeLine(`  ${label} ${event.data.content}`);
+        break;
+      }
+      case 'tool_output':
+        writer.writeChunk(event.data.output);
+        break;
+      case 'thinking':
+        writer.writeChunk(event.data.thinking);
+        break;
+      case 'init':
+      case 'result':
+      case 'error':
+        break;
+    }
+  };
+}
+
 /**
  * Truncate string to maximum length
  */
@@ -106,11 +192,18 @@ export async function executePiece(
 
   // projectCwd is where .takt/ lives (project root, not the clone)
   const projectCwd = options.projectCwd;
+  assertTaskPrefixPair(options.taskPrefix, options.taskColorIndex);
+
+  // When taskPrefix is set (parallel execution), route all output through TaskPrefixWriter
+  const prefixWriter = options.taskPrefix != null
+    ? new TaskPrefixWriter({ taskName: options.taskPrefix, colorIndex: options.taskColorIndex! })
+    : undefined;
+  const out = createOutputFns(prefixWriter);
 
   // Always continue from previous sessions (use /clear to reset)
   log.debug('Continuing session (use /clear to reset)');
 
-  header(`${headerPrefix} ${pieceConfig.name}`);
+  out.header(`${headerPrefix} ${pieceConfig.name}`);
 
   const pieceSessionId = generateSessionId();
   let sessionLog = createSessionLog(task, projectCwd, pieceConfig.name);
@@ -139,14 +232,16 @@ export async function executePiece(
   // Track current display for streaming
   const displayRef: { current: StreamDisplay | null } = { current: null };
 
-  // Create stream handler that delegates to UI display
-  const streamHandler = (
-    event: Parameters<ReturnType<StreamDisplay['createHandler']>>[0]
-  ): void => {
-    if (!displayRef.current) return;
-    if (event.type === 'result') return;
-    displayRef.current.createHandler()(event);
-  };
+  // Create stream handler — when prefixWriter is active, use it for line-buffered
+  // output to prevent mid-line interleaving between concurrent tasks.
+  // When not in parallel mode, delegate to StreamDisplay as before.
+  const streamHandler = prefixWriter
+    ? createPrefixedStreamHandler(prefixWriter)
+    : (event: Parameters<ReturnType<StreamDisplay['createHandler']>>[0]): void => {
+        if (!displayRef.current) return;
+        if (event.type === 'result') return;
+        displayRef.current.createHandler()(event);
+      };
 
   // Load saved agent sessions for continuity (from project root or clone-specific storage)
   const isWorktree = cwd !== projectCwd;
@@ -180,14 +275,14 @@ export async function executePiece(
       displayRef.current = null;
     }
 
-    blankLine();
-    warn(
+    out.blankLine();
+    out.warn(
       getLabel('piece.iterationLimit.maxReached', undefined, {
         currentIteration: String(request.currentIteration),
         maxIterations: String(request.maxIterations),
       })
     );
-    info(getLabel('piece.iterationLimit.currentMovement', undefined, { currentMovement: request.currentMovement }));
+    out.info(getLabel('piece.iterationLimit.currentMovement', undefined, { currentMovement: request.currentMovement }));
 
     if (shouldNotify) {
       playWarningSound();
@@ -214,11 +309,11 @@ export async function executePiece(
 
       const additionalIterations = Number.parseInt(input, 10);
       if (Number.isInteger(additionalIterations) && additionalIterations > 0) {
-        pieceConfig.maxIterations += additionalIterations;
+        pieceConfig.maxIterations = request.maxIterations + additionalIterations;
         return additionalIterations;
       }
 
-      warn(getLabel('piece.iterationLimit.invalidInput'));
+      out.warn(getLabel('piece.iterationLimit.invalidInput'));
     }
   };
 
@@ -228,14 +323,15 @@ export async function executePiece(
           displayRef.current.flush();
           displayRef.current = null;
         }
-        blankLine();
-        info(request.prompt.trim());
+        out.blankLine();
+        out.info(request.prompt.trim());
         const input = await promptInput(getLabel('piece.iterationLimit.userInputPrompt'));
         return input && input.trim() ? input.trim() : null;
       }
     : undefined;
 
   const engine = new PieceEngine(pieceConfig, cwd, task, {
+    abortSignal: options.abortSignal,
     onStream: streamHandler,
     onUserInput,
     initialSessions: savedSessions,
@@ -245,11 +341,14 @@ export async function executePiece(
     language: options.language,
     provider: options.provider,
     model: options.model,
+    personaProviders: options.personaProviders,
     interactive: interactiveUserInput,
     detectRuleIndex,
     callAiJudge,
     startMovement: options.startMovement,
     retryNote: options.retryNote,
+    taskPrefix: options.taskPrefix,
+    taskColorIndex: options.taskColorIndex,
   });
 
   let abortReason: string | undefined;
@@ -257,6 +356,7 @@ export async function executePiece(
   let lastMovementName: string | undefined;
   let currentIteration = 0;
   const phasePrompts = new Map<string, string>();
+  const movementIterations = new Map<string, number>();
 
   engine.on('phase:start', (step, phase, phaseName, instruction) => {
     log.debug('Phase starting', { step: step.name, phase, phaseName });
@@ -311,7 +411,15 @@ export async function executePiece(
   engine.on('movement:start', (step, iteration, instruction) => {
     log.debug('Movement starting', { step: step.name, persona: step.personaDisplayName, iteration });
     currentIteration = iteration;
-    info(`[${iteration}/${pieceConfig.maxIterations}] ${step.name} (${step.personaDisplayName})`);
+    const movementIteration = (movementIterations.get(step.name) ?? 0) + 1;
+    movementIterations.set(step.name, movementIteration);
+    prefixWriter?.setMovementContext({
+      movementName: step.name,
+      iteration,
+      maxIterations: pieceConfig.maxIterations,
+      movementIteration,
+    });
+    out.info(`[${iteration}/${pieceConfig.maxIterations}] ${step.name} (${step.personaDisplayName})`);
 
     // Log prompt content for debugging
     if (instruction) {
@@ -322,15 +430,18 @@ export async function executePiece(
     const movementIndex = pieceConfig.movements.findIndex((m) => m.name === step.name);
     const totalMovements = pieceConfig.movements.length;
 
-    const quiet = isQuietMode();
-    const prefix = options.taskPrefix;
-    const agentLabel = prefix ? `${prefix}:${step.personaDisplayName}` : step.personaDisplayName;
-    displayRef.current = new StreamDisplay(agentLabel, quiet, {
-      iteration,
-      maxIterations: pieceConfig.maxIterations,
-      movementIndex: movementIndex >= 0 ? movementIndex : 0,
-      totalMovements,
-    });
+    // In parallel mode, StreamDisplay is not used (prefixWriter handles output).
+    // In single mode, StreamDisplay renders stream events directly.
+    if (!prefixWriter) {
+      const quiet = isQuietMode();
+      const agentLabel = step.personaDisplayName;
+      displayRef.current = new StreamDisplay(agentLabel, quiet, {
+        iteration,
+        maxIterations: pieceConfig.maxIterations,
+        movementIndex: movementIndex >= 0 ? movementIndex : 0,
+        totalMovements,
+      });
+    }
 
     // Write step_start record to NDJSON log
     const record: NdjsonStepStart = {
@@ -364,25 +475,26 @@ export async function executePiece(
       displayRef.current.flush();
       displayRef.current = null;
     }
-    blankLine();
+    prefixWriter?.flush();
+    out.blankLine();
 
     if (response.matchedRuleIndex != null && step.rules) {
       const rule = step.rules[response.matchedRuleIndex];
       if (rule) {
         const methodLabel = response.matchedRuleMethod ? ` (${response.matchedRuleMethod})` : '';
-        status('Status', `${rule.condition}${methodLabel}`);
+        out.status('Status', `${rule.condition}${methodLabel}`);
       } else {
-        status('Status', response.status);
+        out.status('Status', response.status);
       }
     } else {
-      status('Status', response.status);
+      out.status('Status', response.status);
     }
 
     if (response.error) {
-      error(`Error: ${response.error}`);
+      out.error(`Error: ${response.error}`);
     }
     if (response.sessionId) {
-      status('Session', response.sessionId);
+      out.status('Session', response.sessionId);
     }
 
     // Write step_complete record to NDJSON log
@@ -408,8 +520,8 @@ export async function executePiece(
 
   engine.on('movement:report', (_step, filePath, fileName) => {
     const content = readFileSync(filePath, 'utf-8');
-    console.log(`\n📄 Report: ${fileName}\n`);
-    console.log(content);
+    out.logLine(`\n📄 Report: ${fileName}\n`);
+    out.logLine(content);
   });
 
   engine.on('piece:complete', (state) => {
@@ -445,8 +557,8 @@ export async function executePiece(
       : '';
     const elapsedDisplay = elapsed ? `, ${elapsed}` : '';
 
-    success(`Piece completed (${state.iteration} iterations${elapsedDisplay})`);
-    info(`Session log: ${ndjsonLogPath}`);
+    out.success(`Piece completed (${state.iteration} iterations${elapsedDisplay})`);
+    out.info(`Session log: ${ndjsonLogPath}`);
     if (shouldNotify) {
       notifySuccess('TAKT', getLabel('piece.notifyComplete', undefined, { iteration: String(state.iteration) }));
     }
@@ -459,6 +571,7 @@ export async function executePiece(
       displayRef.current.flush();
       displayRef.current = null;
     }
+    prefixWriter?.flush();
     abortReason = reason;
     sessionLog = finalizeSessionLog(sessionLog, 'aborted');
 
@@ -492,8 +605,8 @@ export async function executePiece(
       : '';
     const elapsedDisplay = elapsed ? ` (${elapsed})` : '';
 
-    error(`Piece aborted after ${state.iteration} iterations${elapsedDisplay}: ${reason}`);
-    info(`Session log: ${ndjsonLogPath}`);
+    out.error(`Piece aborted after ${state.iteration} iterations${elapsedDisplay}: ${reason}`);
+    out.info(`Session log: ${ndjsonLogPath}`);
     if (shouldNotify) {
       notifyError('TAKT', getLabel('piece.notifyAbort', undefined, { reason }));
     }
@@ -536,6 +649,7 @@ export async function executePiece(
       reason: abortReason,
     };
   } finally {
+    prefixWriter?.flush();
     sigintCleanup?.();
     if (onAbortSignal && options.abortSignal) {
       options.abortSignal.removeEventListener('abort', onAbortSignal);

src/features/tasks/execute/taskExecution.ts

@@ -52,7 +52,7 @@ function resolveTaskIssue(issueNumber: number | undefined): ReturnType<typeof fe
  * Execute a single task with piece.
  */
 export async function executeTask(options: ExecuteTaskOptions): Promise<boolean> {
-  const { task, cwd, pieceIdentifier, projectCwd, agentOverrides, interactiveUserInput, interactiveMetadata, startMovement, retryNote, abortSignal, taskPrefix } = options;
+  const { task, cwd, pieceIdentifier, projectCwd, agentOverrides, interactiveUserInput, interactiveMetadata, startMovement, retryNote, abortSignal, taskPrefix, taskColorIndex } = options;
   const pieceConfig = loadPieceByIdentifier(pieceIdentifier, projectCwd);
 
   if (!pieceConfig) {
@@ -77,12 +77,14 @@ export async function executeTask(options: ExecuteTaskOptions): Promise<boolean>
     language: globalConfig.language,
     provider: agentOverrides?.provider,
     model: agentOverrides?.model,
+    personaProviders: globalConfig.personaProviders,
     interactiveUserInput,
     interactiveMetadata,
     startMovement,
     retryNote,
     abortSignal,
     taskPrefix,
+    taskColorIndex,
   });
   return result.success;
 }
@@ -101,16 +103,31 @@ export async function executeAndCompleteTask(
   cwd: string,
   pieceName: string,
   options?: TaskExecutionOptions,
-  parallelOptions?: { abortSignal?: AbortSignal; taskPrefix?: string },
+  parallelOptions?: { abortSignal?: AbortSignal; taskPrefix?: string; taskColorIndex?: number },
 ): Promise<boolean> {
   const startedAt = new Date().toISOString();
   const executionLog: string[] = [];
+  const taskAbortController = new AbortController();
+  const externalAbortSignal = parallelOptions?.abortSignal;
+  const taskAbortSignal = externalAbortSignal ? taskAbortController.signal : undefined;
+
+  const onExternalAbort = (): void => {
+    taskAbortController.abort();
+  };
+
+  if (externalAbortSignal) {
+    if (externalAbortSignal.aborted) {
+      taskAbortController.abort();
+    } else {
+      externalAbortSignal.addEventListener('abort', onExternalAbort, { once: true });
+    }
+  }
 
   try {
     const { execCwd, execPiece, isWorktree, branch, baseBranch, startMovement, retryNote, autoPr, issueNumber } = await resolveTaskExecution(task, cwd, pieceName);
 
     // cwd is always the project root; pass it as projectCwd so reports/sessions go there
-    const taskSuccess = await executeTask({
+    const taskRunPromise = executeTask({
       task: task.content,
       cwd: execCwd,
       pieceIdentifier: execPiece,
@@ -118,9 +135,12 @@ export async function executeAndCompleteTask(
       agentOverrides: options,
       startMovement,
       retryNote,
-      abortSignal: parallelOptions?.abortSignal,
+      abortSignal: taskAbortSignal,
       taskPrefix: parallelOptions?.taskPrefix,
+      taskColorIndex: parallelOptions?.taskColorIndex,
     });
+
+    const taskSuccess = await taskRunPromise;
     const completedAt = new Date().toISOString();
 
     if (taskSuccess && isWorktree) {
@@ -189,6 +209,10 @@ export async function executeAndCompleteTask(
 
     error(`Task "${task.name}" error: ${getErrorMessage(err)}`);
     return false;
+  } finally {
+    if (externalAbortSignal) {
+      externalAbortSignal.removeEventListener('abort', onExternalAbort);
+    }
   }
 }
 
@@ -220,7 +244,7 @@ export async function runAllTasks(
     info(`Concurrency: ${concurrency}`);
   }
 
-  const result = await runWithWorkerPool(taskRunner, initialTasks, concurrency, cwd, pieceName, options);
+  const result = await runWithWorkerPool(taskRunner, initialTasks, concurrency, cwd, pieceName, options, globalConfig.taskPollIntervalMs);
 
   const totalCount = result.success + result.fail;
   blankLine();

src/features/tasks/execute/types.ts

@@ -30,6 +30,8 @@ export interface PieceExecutionOptions {
   language?: Language;
   provider?: ProviderType;
   model?: string;
+  /** Per-persona provider overrides (e.g., { coder: 'codex' }) */
+  personaProviders?: Record<string, ProviderType>;
   /** Enable interactive user input during step transitions */
   interactiveUserInput?: boolean;
   /** Interactive mode result metadata for NDJSON logging */
@@ -42,6 +44,8 @@ export interface PieceExecutionOptions {
   abortSignal?: AbortSignal;
   /** Task name prefix for parallel execution output (e.g. "[task-name] output...") */
   taskPrefix?: string;
+  /** Color index for task prefix (cycled mod 4 across concurrent tasks) */
+  taskColorIndex?: number;
 }
 
 export interface TaskExecutionOptions {
@@ -72,6 +76,8 @@ export interface ExecuteTaskOptions {
   abortSignal?: AbortSignal;
   /** Task name prefix for parallel execution output (e.g. "[task-name] output...") */
   taskPrefix?: string;
+  /** Color index for task prefix (cycled mod 4 across concurrent tasks) */
+  taskColorIndex?: number;
 }
 
 export interface PipelineExecutionOptions {

src/infra/codex/client.ts

@@ -23,6 +23,8 @@ import {
 export type { CodexCallOptions } from './types.js';
 
 const log = createLogger('codex-sdk');
+const CODEX_STREAM_IDLE_TIMEOUT_MS = 10 * 60 * 1000;
+const CODEX_STREAM_ABORTED_MESSAGE = 'Codex execution aborted';
 
 /**
  * Client for Codex SDK agent interactions.
@@ -55,6 +57,34 @@ export class CodexClient {
       ? `${options.systemPrompt}\n\n${prompt}`
       : prompt;
 
+    let idleTimeoutId: ReturnType<typeof setTimeout> | undefined;
+    const streamAbortController = new AbortController();
+    const timeoutMessage = `Codex stream timed out after ${Math.floor(CODEX_STREAM_IDLE_TIMEOUT_MS / 60000)} minutes of inactivity`;
+    let abortCause: 'timeout' | 'external' | undefined;
+
+    const resetIdleTimeout = (): void => {
+      if (idleTimeoutId !== undefined) {
+        clearTimeout(idleTimeoutId);
+      }
+      idleTimeoutId = setTimeout(() => {
+        abortCause = 'timeout';
+        streamAbortController.abort();
+      }, CODEX_STREAM_IDLE_TIMEOUT_MS);
+    };
+
+    const onExternalAbort = (): void => {
+      abortCause = 'external';
+      streamAbortController.abort();
+    };
+
+    if (options.abortSignal) {
+      if (options.abortSignal.aborted) {
+        streamAbortController.abort();
+      } else {
+        options.abortSignal.addEventListener('abort', onExternalAbort, { once: true });
+      }
+    }
+
     try {
       log.debug('Executing Codex thread', {
         agentType,
@@ -62,7 +92,10 @@ export class CodexClient {
         hasSystemPrompt: !!options.systemPrompt,
       });
 
-      const { events } = await thread.runStreamed(fullPrompt);
+      const { events } = await thread.runStreamed(fullPrompt, {
+        signal: streamAbortController.signal,
+      });
+      resetIdleTimeout();
       let content = '';
       const contentOffsets = new Map<string, number>();
       let success = true;
@@ -70,6 +103,7 @@ export class CodexClient {
       const state = createStreamTrackingState();
 
       for await (const event of events as AsyncGenerator<CodexEvent>) {
+        resetIdleTimeout();
         if (event.type === 'thread.started') {
           threadId = typeof event.thread_id === 'string' ? event.thread_id : threadId;
           emitInit(options.onStream, options.model, threadId);
@@ -172,15 +206,27 @@ export class CodexClient {
       };
     } catch (error) {
       const message = getErrorMessage(error);
-      emitResult(options.onStream, false, message, threadId);
+      const errorMessage = streamAbortController.signal.aborted
+        ? abortCause === 'timeout'
+          ? timeoutMessage
+          : CODEX_STREAM_ABORTED_MESSAGE
+        : message;
+      emitResult(options.onStream, false, errorMessage, threadId);
 
       return {
         persona: agentType,
         status: 'blocked',
-        content: message,
+        content: errorMessage,
         timestamp: new Date(),
         sessionId: threadId,
       };
+    } finally {
+      if (idleTimeoutId !== undefined) {
+        clearTimeout(idleTimeoutId);
+      }
+      if (options.abortSignal) {
+        options.abortSignal.removeEventListener('abort', onExternalAbort);
+      }
     }
   }
 

src/infra/codex/types.ts

@@ -21,6 +21,7 @@ export function mapToCodexSandboxMode(mode: PermissionMode): CodexSandboxMode {
 /** Options for calling Codex */
 export interface CodexCallOptions {
   cwd: string;
+  abortSignal?: AbortSignal;
   sessionId?: string;
   model?: string;
   systemPrompt?: string;

src/infra/config/global/globalConfig.ts

@@ -37,6 +37,7 @@ function createDefaultGlobalConfig(): GlobalConfig {
     enableBuiltinPieces: true,
     interactivePreviewMovements: 3,
     concurrency: 1,
+    taskPollIntervalMs: 500,
   };
 }
 
@@ -105,11 +106,13 @@ export class GlobalConfigManager {
       minimalOutput: parsed.minimal_output,
       bookmarksFile: parsed.bookmarks_file,
       pieceCategoriesFile: parsed.piece_categories_file,
+      personaProviders: parsed.persona_providers,
       branchNameStrategy: parsed.branch_name_strategy,
       preventSleep: parsed.prevent_sleep,
       notificationSound: parsed.notification_sound,
       interactivePreviewMovements: parsed.interactive_preview_movements,
       concurrency: parsed.concurrency,
+      taskPollIntervalMs: parsed.task_poll_interval_ms,
     };
     validateProviderModelCompatibility(config.provider, config.model);
     this.cachedConfig = config;
@@ -170,6 +173,9 @@ export class GlobalConfigManager {
     if (config.pieceCategoriesFile) {
       raw.piece_categories_file = config.pieceCategoriesFile;
     }
+    if (config.personaProviders && Object.keys(config.personaProviders).length > 0) {
+      raw.persona_providers = config.personaProviders;
+    }
     if (config.branchNameStrategy) {
       raw.branch_name_strategy = config.branchNameStrategy;
     }
@@ -185,6 +191,9 @@ export class GlobalConfigManager {
     if (config.concurrency !== undefined && config.concurrency > 1) {
       raw.concurrency = config.concurrency;
     }
+    if (config.taskPollIntervalMs !== undefined && config.taskPollIntervalMs !== 500) {
+      raw.task_poll_interval_ms = config.taskPollIntervalMs;
+    }
     writeFileSync(configPath, stringifyYaml(raw), 'utf-8');
     this.invalidateCache();
   }

src/infra/config/loaders/index.ts

@@ -13,6 +13,7 @@ export {
   listPieces,
   listPieceEntries,
   type MovementPreview,
+  type FirstMovementInfo,
   type PieceDirEntry,
   type PieceSource,
   type PieceWithSource,

src/infra/config/loaders/pieceLoader.ts

@@ -21,6 +21,7 @@ export {
   listPieces,
   listPieceEntries,
   type MovementPreview,
+  type FirstMovementInfo,
   type PieceDirEntry,
   type PieceSource,
   type PieceWithSource,

src/infra/config/loaders/pieceParser.ts

@@ -280,6 +280,7 @@ export function normalizePieceConfig(
     maxIterations: parsed.max_iterations,
     loopMonitors: normalizeLoopMonitors(parsed.loop_monitors, pieceDir, sections, context),
     answerAgent: parsed.answer_agent,
+    interactiveMode: parsed.interactive_mode,
   };
 }
 

src/infra/config/loaders/pieceResolver.ts

@@ -8,7 +8,7 @@
 import { existsSync, readFileSync, readdirSync, statSync } from 'node:fs';
 import { join, resolve, isAbsolute } from 'node:path';
 import { homedir } from 'node:os';
-import type { PieceConfig, PieceMovement } from '../../../core/models/index.js';
+import type { PieceConfig, PieceMovement, InteractiveMode } from '../../../core/models/index.js';
 import { getGlobalPiecesDir, getBuiltinPiecesDir, getProjectConfigDir } from '../paths.js';
 import { getLanguage, getDisabledBuiltins, getBuiltinPiecesEnabled } from '../global/globalConfig.js';
 import { createLogger, getErrorMessage } from '../../../shared/utils/index.js';
@@ -219,24 +219,10 @@ function buildMovementPreviews(piece: PieceConfig, maxCount: number): MovementPr
     const movement = movementMap.get(currentName);
     if (!movement) break;
 
-    let personaContent = '';
-    if (movement.personaPath) {
-      try {
-        personaContent = readFileSync(movement.personaPath, 'utf-8');
-      } catch (err) {
-        log.debug('Failed to read persona file for preview', {
-          path: movement.personaPath,
-          error: getErrorMessage(err),
-        });
-      }
-    } else if (movement.persona) {
-      personaContent = movement.persona;
-    }
-
     previews.push({
       name: movement.name,
       personaDisplayName: movement.personaDisplayName,
-      personaContent,
+      personaContent: readMovementPersona(movement),
       instructionContent: movement.instructionTemplate,
       allowedTools: movement.allowedTools ?? [],
       canEdit: movement.edit === true,
@@ -250,26 +236,86 @@ function buildMovementPreviews(piece: PieceConfig, maxCount: number): MovementPr
   return previews;
 }
 
+/**
+ * Read persona content from a movement.
+ * When personaPath is set, reads from file (returns empty on failure).
+ * Otherwise uses inline persona string.
+ */
+function readMovementPersona(movement: PieceMovement): string {
+  if (movement.personaPath) {
+    try {
+      return readFileSync(movement.personaPath, 'utf-8');
+    } catch (err) {
+      log.debug('Failed to read persona file', {
+        path: movement.personaPath,
+        error: getErrorMessage(err),
+      });
+      return '';
+    }
+  }
+  return movement.persona ?? '';
+}
+
+/** First movement info for persona mode */
+export interface FirstMovementInfo {
+  /** Persona prompt content */
+  personaContent: string;
+  /** Persona display name */
+  personaDisplayName: string;
+  /** Allowed tools for this movement */
+  allowedTools: string[];
+}
+
 /**
  * Get piece description by identifier.
- * Returns the piece name, description, workflow structure, and optional movement previews.
+ * Returns the piece name, description, workflow structure, optional movement previews,
+ * piece-level interactive mode default, and first movement info for persona mode.
  */
 export function getPieceDescription(
   identifier: string,
   projectCwd: string,
   previewCount?: number,
-): { name: string; description: string; pieceStructure: string; movementPreviews: MovementPreview[] } {
+): {
+  name: string;
+  description: string;
+  pieceStructure: string;
+  movementPreviews: MovementPreview[];
+  interactiveMode?: InteractiveMode;
+  firstMovement?: FirstMovementInfo;
+} {
   const piece = loadPieceByIdentifier(identifier, projectCwd);
   if (!piece) {
     return { name: identifier, description: '', pieceStructure: '', movementPreviews: [] };
   }
+
+  const previews = previewCount && previewCount > 0
+    ? buildMovementPreviews(piece, previewCount)
+    : [];
+
+  const firstMovement = buildFirstMovementInfo(piece);
+
   return {
     name: piece.name,
     description: piece.description ?? '',
     pieceStructure: buildWorkflowString(piece.movements),
-    movementPreviews: previewCount && previewCount > 0
-      ? buildMovementPreviews(piece, previewCount)
-      : [],
+    movementPreviews: previews,
+    interactiveMode: piece.interactiveMode,
+    firstMovement,
+  };
+}
+
+/**
+ * Build first movement info for persona mode.
+ * Reads persona content from the initial movement.
+ */
+function buildFirstMovementInfo(piece: PieceConfig): FirstMovementInfo | undefined {
+  const movement = piece.movements.find((m) => m.name === piece.initialMovement);
+  if (!movement) return undefined;
+
+  return {
+    personaContent: readMovementPersona(movement),
+    personaDisplayName: movement.personaDisplayName,
+    allowedTools: movement.allowedTools ?? [],
   };
 }
 

src/infra/providers/codex.ts

@@ -27,6 +27,7 @@ function isInsideGitRepo(cwd: string): boolean {
 function toCodexOptions(options: ProviderCallOptions): CodexCallOptions {
   return {
     cwd: options.cwd,
+    abortSignal: options.abortSignal,
     sessionId: options.sessionId,
     model: options.model,
     permissionMode: options.permissionMode,

src/infra/providers/types.ts

@@ -20,6 +20,7 @@ export interface AgentSetup {
 /** Runtime options passed at call time */
 export interface ProviderCallOptions {
   cwd: string;
+  abortSignal?: AbortSignal;
   sessionId?: string;
   model?: string;
   allowedTools?: string[];

src/shared/i18n/labels_en.yaml

@@ -24,6 +24,17 @@ interactive:
       continue: "Continue editing"
     cancelled: "Cancelled"
     playNoTask: "Please specify task content: /play <task>"
+    personaFallback: "No persona available for the first movement. Falling back to assistant mode."
+  modeSelection:
+    prompt: "Select interactive mode:"
+    assistant: "Assistant"
+    assistantDescription: "Ask clarifying questions before generating instructions"
+    persona: "Persona"
+    personaDescription: "Converse as the first agent's persona"
+    quiet: "Quiet"
+    quietDescription: "Generate instructions without asking questions"
+    passthrough: "Passthrough"
+    passthroughDescription: "Pass your input directly as task text"
   previousTask:
     success: "✅ Previous task completed successfully"
     error: "❌ Previous task failed: {error}"

src/shared/i18n/labels_ja.yaml

@@ -24,6 +24,17 @@ interactive:
       continue: "会話を続ける"
     cancelled: "キャンセルしました"
     playNoTask: "タスク内容を指定してください: /play <タスク内容>"
+    personaFallback: "先頭ムーブメントにペルソナがありません。アシスタントモードにフォールバックします。"
+  modeSelection:
+    prompt: "対話モードを選択してください:"
+    assistant: "アシスタント"
+    assistantDescription: "確認質問をしてから指示書を作成"
+    persona: "ペルソナ"
+    personaDescription: "先頭エージェントのペルソナで対話"
+    quiet: "クワイエット"
+    quietDescription: "質問なしでベストエフォートの指示書を生成"
+    passthrough: "パススルー"
+    passthroughDescription: "入力をそのままタスクとして渡す"
   previousTask:
     success: "✅ 前回のタスクは正常に完了しました"
     error: "❌ 前回のタスクはエラーで終了しました: {error}"

src/shared/ui/TaskPrefixWriter.ts

@@ -0,0 +1,118 @@
+/**
+ * Line-buffered, prefixed writer for task-level parallel execution.
+ *
+ * When multiple tasks run concurrently (takt run --concurrency N), each task's
+ * output must be identifiable and line-aligned to prevent mid-line interleaving.
+ * This class wraps process.stdout.write with line buffering and a colored
+ * `[taskName]` prefix on every non-empty line.
+ *
+ * Design mirrors ParallelLogger (movement-level) but targets task-level output:
+ * - Regular log lines (info, header, status) get the prefix
+ * - Stream output gets line-buffered then prefixed
+ * - Empty lines are passed through without prefix
+ */
+
+import { stripAnsi } from '../utils/text.js';
+
+/** ANSI color codes for task prefixes (cycled by task index) */
+const TASK_COLORS = ['\x1b[36m', '\x1b[33m', '\x1b[35m', '\x1b[32m'] as const;
+const RESET = '\x1b[0m';
+
+export interface TaskPrefixWriterOptions {
+  /** Task name used in the prefix */
+  taskName: string;
+  /** Color index for the prefix (cycled mod 4) */
+  colorIndex: number;
+  /** Override process.stdout.write for testing */
+  writeFn?: (text: string) => void;
+}
+
+export interface MovementPrefixContext {
+  movementName: string;
+  iteration: number;
+  maxIterations: number;
+  movementIteration: number;
+}
+
+/**
+ * Prefixed line writer for a single parallel task.
+ *
+ * All output goes through `writeLine` (complete lines) or `writeChunk`
+ * (buffered partial lines). The prefix `[taskName]` is prepended to every
+ * non-empty output line.
+ */
+export class TaskPrefixWriter {
+  private readonly taskPrefix: string;
+  private readonly writeFn: (text: string) => void;
+  private movementContext: MovementPrefixContext | undefined;
+  private lineBuffer = '';
+
+  constructor(options: TaskPrefixWriterOptions) {
+    const color = TASK_COLORS[options.colorIndex % TASK_COLORS.length];
+    const taskLabel = options.taskName.slice(0, 4);
+    this.taskPrefix = `${color}[${taskLabel}]${RESET}`;
+    this.writeFn = options.writeFn ?? ((text: string) => process.stdout.write(text));
+  }
+
+  setMovementContext(context: MovementPrefixContext): void {
+    this.movementContext = context;
+  }
+
+  private buildPrefix(): string {
+    if (!this.movementContext) {
+      return `${this.taskPrefix} `;
+    }
+
+    const { movementName, iteration, maxIterations, movementIteration } = this.movementContext;
+    return `${this.taskPrefix}[${movementName}](${iteration}/${maxIterations})(${movementIteration}) `;
+  }
+
+  /**
+   * Write a complete line with prefix.
+   * Multi-line text is split and each non-empty line gets the prefix.
+   */
+  writeLine(text: string): void {
+    const cleaned = stripAnsi(text);
+    const lines = cleaned.split('\n');
+
+    for (const line of lines) {
+      if (line === '') {
+        this.writeFn('\n');
+      } else {
+        this.writeFn(`${this.buildPrefix()}${line}\n`);
+      }
+    }
+  }
+
+  /**
+   * Write a chunk of streaming text with line buffering.
+   * Partial lines are buffered until a newline arrives, then output with prefix.
+   */
+  writeChunk(text: string): void {
+    const cleaned = stripAnsi(text);
+    const combined = this.lineBuffer + cleaned;
+    const parts = combined.split('\n');
+
+    const remainder = parts.pop() ?? '';
+    this.lineBuffer = remainder;
+
+    for (const line of parts) {
+      if (line === '') {
+        this.writeFn('\n');
+      } else {
+        this.writeFn(`${this.buildPrefix()}${line}\n`);
+      }
+    }
+  }
+
+  /**
+   * Flush any remaining buffered content.
+   */
+  flush(): void {
+    if (this.lineBuffer !== '') {
+      this.writeFn(`${this.buildPrefix()}${this.lineBuffer}\n`);
+      this.lineBuffer = '';
+    }
+  }
+
+}