with one click
research
多 Agent 並行研究框架 - 多視角同時研究,智能匯總成完整報告
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Menu
多 Agent 並行研究框架 - 多視角同時研究,智能匯總成完整報告
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
跨子系統架構一致性審查 - 利用多視角並行分析架構健康度
多 Agent 監督式實作框架 - TDD 驅動、即時審查、品質守護
智能 commit .claude/memory/ 目錄的變更(含任務追蹤)
端到端工作流編排器 - File-Based Handoff + 智能並行決策
多 Agent 並行規劃框架 - 多視角同時設計,共識驅動實作計劃
多 Agent 並行審查框架 - 多視角程式碼審查,問題分類與優先排序
| name | research |
| version | 3.2.0 |
| description | 多 Agent 並行研究框架 - 多視角同時研究,智能匯總成完整報告 |
| triggers | ["multi-research","parallel-research","多角度研究"] |
| context | fork |
| allowed-tools | ["Read","Grep","Glob","WebFetch","Write","Bash"] |
| model | sonnet |
CT 模式偵測 → 多視角並行研究 → 交叉驗證 → CT 合規檢查 → 智能匯總 → Memory 存檔(自動 commit)
⚡ 本 skill 已整合 Claude Code Hooks
- Action logging、state tracking、git commit 均由 hooks 自動處理
- 只需執行 CP1 初始化,其餘檢查點自動執行
/multi-research [研究主題]
/multi-research AI Agent 架構設計模式 --deep
/multi-research --ct experiment "Multi-CT Pipeline 是否能降低 agent drift?"
/multi-research --no-ct "只要快速草稿"
Flags: --perspectives N | --quick | --deep | --ct off|lite|strict|experiment | --no-ct | --no-memory
/multi-research 會在開始前執行 CT Escalation Router,自主選擇 CT 模式。
| Mode | 觸發情境 | 額外產物 |
|---|---|---|
off | 使用者明確 --no-ct 或 --ct off | 無 |
lite | 預設;一般研究、比較、方向整理 | synthesis 內的 evidence / uncertainty / drift guard notes |
strict | 架構、風險、產品、技術決策、agent/memory/tool policy | ct-stack.yaml、ct-compliance.md、risk-policy.yaml |
experiment | 論文、實驗、benchmark、evaluation、hypothesis、可重複驗證 | strict 產物 + hypotheses.yaml、experiment-plan.md、eval-rubric.yaml、failure-modes.md、experiments/{topic-id}/ harness |
CT mode 會控制實際流程重量:
lite 只做 evidence / uncertainty / drift guard,不產生 ct-stack.yaml,不跑 experiment harness,也不跑 autonomous upgrade。strict 才產生 CT stack、合規檢查與 retrospective;若發現系統性問題,只產生 self-upgrade proposal,不自動 patch。experiment 才產生可測假設、實驗設計、eval rubric、failure mining 與 experiment harness;autonomous patch 預設關閉,只收斂 proposal 與 closed-loop summary。自動規則:
lite。--deep 會把最低模式升到 strict。experiment。→ Router:shared/ct/escalation-router.md → Rules:shared/ct/escalation-rules.yaml → Runtime:shared/ct/mode-runtime.yaml
| ID | 名稱 | 模型 | 聚焦 |
|---|---|---|---|
architecture | 架構分析師 | sonnet | 系統結構、設計模式 |
cognitive | 認知研究員 | sonnet | 方法論、思維框架 |
workflow | 工作流設計 | haiku | 執行流程、整合策略 |
industry | 業界實踐 | haiku | 現有框架、最佳實踐 |
→ 模型路由配置:shared/config/model-routing.yaml
CP1: 工作流初始化 ⚡ 手動執行
python scripts/hooks/init_workflow.py --topic "{topic}" --stage RESEARCH
↓
Phase 0: CT Escalation Router → 偵測 ct_mode、理由、信心度
├── 檢查 user override: --ct / --no-ct
├── 依規則計分: lite / strict / experiment
├── 套用 downgrade guard: 快速 / 粗略 / 不用太詳細
└── 寫入 meta.yaml: ct_detection
↓
Phase 1: 北極星錨定 → 定義研究目標、成功標準
↓
Phase 2: Memory 搜尋 → 避免重複研究
↓
Phase 3: 視角分解 → 為每視角生成專屬 prompt + CT envelope
↓
Phase 4: MAP(並行研究)✅ 自動追蹤
┌──────────┬──────────┬──────────┬──────────┐
│架構分析師│認知研究員│工作流設計│業界實踐 │
└──────────┴──────────┴──────────┴──────────┘
[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: "架構視角", ...})
- Task({description: "認知視角", ...})
- Task({description: "工作流視角", ...})
- Task({description: "業界視角", ...})
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/research/{topic-id}/perspectives/
2. Write → .claude/memory/research/{topic-id}/perspectives/{perspective_id}.md
未執行 Write = 任務失敗,工作流中止
↓
Phase 5: REDUCE(交叉驗證 + mode-scoped CT checks + 匯總)
↓
Phase 6: Memory 存檔 → 品質閘門檢查 → 存儲報告
↓
Phase 7: CT Retrospective(strict / experiment)
├── 檢查 selected_mode 是否正確
├── 比對 expected artifacts vs actual outputs
├── 挖掘 workflow failures / false positives / false negatives
└── 寫入 ct-retrospective.md
↓
Phase 8: Self-Upgrade Proposal(strict / experiment,必要時)
├── 若 mode 選錯、artifact 缺漏、gate 誤判、工具解析失敗
├── 產生 self-upgrade-proposal.md
└── runtime 變更需附驗證命令與 rollback plan
↓
Phase 9: Experiment Harness(CT-experiment 必須)
├── 產生 experiments/{topic-id}/cases.yaml
├── 產生 run-config.yaml / rubric.yaml
├── 至少對本次 condition 跑 score-run.py
├── 產生 results.jsonl
└── 產生 analysis.md
↓
Phase 10: Autonomous Upgrade Decision(experiment,proposal-only by default)
├── L1/L2: 只記錄 / 只提案,不修改
├── L3: 可修改 docs、templates、CT examples
├── L4a: 可修改非關鍵 runtime rules / validators,必須跑 focused smoke test
├── L4b/L4c: gates / workflow scripts 只提案,需人工批准
├── L5: 架構或降弱 gate 的變更停止並要求人工批准
└── 寫入 upgrade-decision.yaml;有 patch 時寫入 upgrade-report.md
↓
Phase 11: Closed-Loop Summary(成果收斂)
├── 彙整 research conclusion、CT mode review、upgrade decision
├── 彙整 quality gates / DAG / status / action log 驗證結果
└── 寫入 closed-loop-summary.md
↓
CP4: Task Commit ✅ 自動執行
[寫入 .claude/memory/ 時自動 git commit]
由
post_write.pyhook 自動處理
當 Write 工具寫入 .claude/memory/ 目錄時,hook 會自動:
git add .claude/memory/research/{topic-id}/git commit -m "docs(research): complete {topic} research"actions.jsonl→ Hook 設定:.claude/settings.local.json.template → 協議:shared/checkpoints/mandatory-checkpoints.md
通過條件(RESEARCH 階段):
CT gates 依 mode 分級:
CT_LITE:只檢查 evidence / uncertainty / drift guardCT_STRICT:檢查 ct-stack、ct-compliance、HIGH 違規與 evidence coverageCT_EXPERIMENT:檢查 hypotheses、experiment plan、readiness 與 scorer result→ 閘門配置:shared/quality/gates.yaml
當 consensus_rate >= 0.9 時,可跳過衝突解決。
→ 配置:shared/config/early-termination.yaml
自動偵測技術棧關鍵字(react, vue, fastapi 等)時,查詢最新文檔。
→ 配置:shared/integration/context7.yaml
.claude/memory/research/[topic-id]/
├── meta.yaml # 元數據,包含 ct_detection
├── ct-stack.yaml # CT-strict / experiment 產出
├── ct-compliance.md # CT-strict / experiment 產出
├── risk-policy.yaml # CT-strict / experiment 產出
├── hypotheses.yaml # CT-experiment 產出
├── experiment-plan.md # CT-experiment 產出
├── eval-rubric.yaml # CT-experiment 產出
├── failure-modes.md # CT-experiment 產出
├── claims/ # CT-strict / experiment structured claims
│ ├── architecture.claims.yaml
│ ├── cognitive.claims.yaml
│ ├── workflow.claims.yaml
│ └── industry.claims.yaml
├── ct-retrospective.md # CT 閉環產出
├── self-upgrade-proposal.md # 有改善建議時產出
├── upgrade-decision.yaml # 自主升級決策
├── upgrade-report.md # 有實際 patch 時產出
├── closed-loop-summary.md # 最終收斂成果報告
├── experiments/
│ └── {topic-id}/
│ ├── cases.yaml
│ ├── run-config.yaml
│ ├── rubric.yaml
│ ├── results.jsonl
│ ├── condition-comparison.md
│ └── analysis.md
├── perspectives/ # 完整視角報告(MAP 產出,保留)
│ ├── architecture.md
│ ├── cognitive.md
│ ├── workflow.md
│ └── industry.md
├── summaries/ # 結構化摘要(REDUCE 產出,供快速查閱)
│ ├── architecture.yaml
│ ├── cognitive.yaml
│ ├── workflow.yaml
│ └── industry.yaml
├── synthesis.md # 匯總報告(主輸出)
└── metrics.yaml # 階段指標
Mode-specific artifact rules:
lite must not require ct-stack.yaml, ct-compliance.md, experiment harness, or self-upgrade artifacts.strict may produce ct-retrospective.md and self-upgrade-proposal.md, but must not apply autonomous patches.experiment must produce experiment artifacts and closed-loop-summary.md; autonomous patching remains proposal-only unless explicitly approved.⚠️ perspectives/ 保存完整報告,summaries/ 保存結構化摘要,兩者都必須保留。
視角 Agent 不應該開啟 Task:
| 允許的操作 | 說明 |
|---|---|
| ✅ Read | 讀取檔案 |
| ✅ Glob/Grep | 搜尋檔案和內容 |
| ✅ Explore agent | 輕量級探索 |
| ✅ Bash | 執行命令 |
| ✅ WebFetch | 抓取網頁 |
| ✅ Write | 寫入報告 |
| ❌ Task | 開子 Agent |
當需要抓取網頁時,使用以下順序:
a. mcp__claude-in-chrome__tabs_create_mcp → 建立新分頁
b. mcp__claude-in-chrome__navigate → 導航到 URL
c. mcp__claude-in-chrome__get_page_text → 讀取內容
由 Claude Code Hooks 自動處理
工具調用自動記錄到 .claude/workflow/{workflow-id}/logs/actions.jsonl。
自動記錄的工具:
| 工具 | 觸發 Hook | 記錄內容 |
|---|---|---|
| Task | pre_task.py / post_task.py | Agent 啟動/完成狀態 |
| Write | post_write.py | 檔案路徑、Memory commit |
排查問題:
# 查看 RESEARCH 階段所有失敗行動
jq 'select(.stage == "RESEARCH" and .status == "failed")' \
.claude/workflow/{workflow-id}/logs/actions.jsonl
# 查看特定視角 Agent 的行動
jq 'select(.agent_id == "architecture")' \
.claude/workflow/{workflow-id}/logs/actions.jsonl
# 查看即時狀態
cat .claude/workflow/{workflow-id}/current.json | jq .
→ Hook 腳本:scripts/hooks/ → 日誌規範:shared/communication/execution-logs.md
| 模組 | 用途 |
|---|---|
| ct/escalation-router.md | CT 模式自動偵測 |
| ct/escalation-rules.yaml | CT 升級/降級規則 |
| 02-ct-mode/lite.md | CT-lite envelope |
| 02-ct-mode/strict.md | CT-strict envelope |
| 02-ct-mode/experiment.md | CT-experiment envelope |
| 02-ct-mode/compliance.md | CT compliance rules |
| 02-ct-mode/experiment-design.md | 可驗證實驗設計 |
| ct/experiment-harness/README.md | 可重跑 experiment harness |
| 02-ct-mode/retrospective.md | CT 閉環檢查 |
| 02-ct-mode/self-upgrade-proposal.md | 自我改善提案 |
| 02-ct-mode/autonomous-upgrade.md | 自主升級決策與執行 |
| 02-ct-mode/closed-loop-summary.md | 閉環成果收斂報告 |
| ct/retrospective.md | 共用 CT retrospective 規範 |
| ct/autonomous-upgrade.md | 共用自主升級規範 |
| ct/self-upgrade-policy.yaml | 自動化等級與安全規則 |
| coordination/map-phase.md | 並行協調 |
| coordination/reduce-phase.md | 匯總整合、大檔案處理 |
| synthesis/cross-validation.md | 交叉驗證 |
| quality/gates.yaml | 品質閘門 |
| config/model-routing.yaml | 模型路由 |
RESEARCH → PLAN → TASKS → IMPLEMENT → REVIEW → VERIFY
↑
你在這裡
研究結果可被 plan skill 引用,作為規劃的輸入。