plan

Multi-Agent Plan v3.1.0

Multi-perspective parallel planning → Risk assessment → Consensus design → Implementation plan (auto commit)

Automation Mechanisms

⚡ This skill has integrated Claude Code Hooks

Action logging, state tracking, and git commit are all handled automatically by hooks

Only CP1 initialization needs to be executed; the remaining checkpoints execute automatically

Usage

/multi-plan [功能需求]
/multi-plan 建立用戶認證系統 --from-research ID

Flags: --perspectives N | --quick | --deep | --from-research ID

Default 4 perspectives

ID	Name	Model	Focus
`architect`	System Architect	sonnet	Technical feasibility, component design
`risk-analyst`	Risk Analyst	sonnet	Potential risks, failure scenarios
`estimator`	Estimation Expert	haiku	Effort estimation, schedule planning
`ux-advocate`	UX Advocate	haiku	User experience, API design

→ Model routing configuration:shared/config/model-routing.yaml

Execution Flow

CP1: 工作流初始化 ⚡ 手動執行
    python scripts/hooks/init_workflow.py --topic "{feature}" --stage PLAN
    ↓
Phase 0: 需求錨定 → 載入 research 上下文、確認驗收標準
    ↓
Phase 1: Memory 搜尋 → 識別可重用的設計決策
    ↓
Phase 2: 視角分解 → 早期安全/架構檢查
    ↓
Phase 3: MAP（並行規劃）✅ 自動追蹤
    ┌──────────┬──────────┬──────────┬──────────┐
    │ 架構師   │ 風險分析 │ 估算專家 │ UX 倡導  │
    └──────────┴──────────┴──────────┴──────────┘
    [CP2/CP3 由 hooks 自動處理 Agent 狀態追蹤]

    ⚠️ **並行執行關鍵**：
       在單一訊息中發送 4 個 Task 工具呼叫：
       - Task({description: "架構師視角", ...})
       - Task({description: "風險分析視角", ...})
       - Task({description: "估算專家視角", ...})
       - Task({description: "UX 倡導視角", ...})
       這樣才能真正並行執行！

    ⚠️ **強制**：每個 Agent 必須在完成前執行：
       1. mkdir -p .claude/memory/plans/{feature-id}/perspectives/
       2. Write → .claude/memory/plans/{feature-id}/perspectives/{perspective_id}.md
       未執行 Write = 任務失敗，工作流中止
    ↓
Phase 4: REDUCE（共識設計 + 風險緩解）
    ↓
Phase 5: 產出實作計劃 → 品質閘門檢查
    ↓
Phase 6: Memory 存檔
    ↓
CP4: Task Commit ✅ 自動執行
    [寫入 .claude/memory/ 時自動 git commit]

Early Error Interception

Perform preventive checks in Phase 2:

Security（BLOCKER）：

Authentication mechanism defined
Authorization model specified
Data validation strategy defined

Architecture（HIGH）：

No circular dependencies
Clear module boundaries

→ Configuration:shared/quality/early-detection.yaml

CP4: Task Commit

After Memory archiving is complete,CP4 Task Commit must be executed。

Phase 6: Memory 存檔
    ↓
CP4: Task Commit
    ├── git add .claude/memory/plans/{feature-id}/
    └── git commit -m "feat(plan): design {feature} implementation plan"

→ Protocol:shared/git/commit-protocol.md

Quality Gates

Pass conditions (PLAN phase):

✅ All necessary components have designs
✅ Risk assessment completed
✅ Milestones clearly defined
✅ Quality score ≥ 75

→ Gate configuration:shared/quality/gates.yaml

Output Structure

.claude/memory/plans/[feature-id]/
├── meta.yaml               # 元數據
├── perspectives/           # 完整視角報告（MAP 產出，保留）
│   ├── architect.md
│   ├── risk-analyst.md
│   ├── estimator.md
│   └── ux-advocate.md
├── summaries/              # 結構化摘要（REDUCE 產出，供快速查閱）
│   ├── architect.yaml
│   ├── risk-analyst.yaml
│   ├── estimator.yaml
│   └── ux-advocate.yaml
├── synthesis.md            # 共識設計
├── implementation-plan.md  # 實作計劃（主輸出）
├── milestones.md           # 里程碑清單
└── risk-mitigation.md      # 風險緩解策略

⚠️ perspectives/ stores complete reports, summaries/ stores structured summaries; both must be retained.

Agent Capability Restrictions

Perspective Agents should not open Tasks：

Allowed Operations	Description
✅ Read	Read files
✅ Glob/Grep	Search files and content
✅ Explore agent	Lightweight exploration
✅ Bash	Execute commands
✅ WebFetch	Fetch web pages
✅ Write	Write reports
❌ Task	Spawn sub-agents

Action Log

After each tool call completes, record to .claude/workflow/{workflow-id}/logs/actions.jsonl。

Recording Timing：

Success: record tool、input、output_preview、duration_ms、status: success
Failure: record tool、input、error、stderr(if any),status: failed

Key Actions (PLAN phase)：

Action	Recording Focus
Read (read research report)	`file_path`、`output_size`
Task (launch Agent)	`subagent_type`、`prompt` (truncated)、`agent_id`
Write (write plan)	`file_path`、`content_size`

Troubleshooting：

# 查看 PLAN 階段所有失敗行動
jq 'select(.stage == "PLAN" and .status == "failed")' actions.jsonl

# 查看特定視角 Agent 的行動
jq 'select(.agent_id == "agent_architect")' actions.jsonl

→ Log specification:shared/communication/execution-logs.md

Shared Modules

Module	Purpose
coordination/map-phase.md	Parallel coordination
coordination/reduce-phase.md	Aggregation and integration, large file processing
synthesis/cross-validation.md	Cross-validation
quality/gates.yaml	Quality gates
quality/early-detection.yaml	Early interception

Workflow Position

RESEARCH → PLAN → TASKS → IMPLEMENT → REVIEW → VERIFY
             ↑
          你在這裡

Input: can load research reports from the research skill
Output：implementation-plan.md for the tasks skill to use