# research-orchestrator > Human-Centered Orchestrator for Research Coordinator v6.7.0 Manages 44 research agents across 9 categories (A-I) with MANDATORY human checkpoints No autonomous modes - all critical decisions require explicit human approval Features: ScholaRAG Integration, Meta-Analysis System, Humanization Pipeline - Author: HosungYou - Repository: HosungYou/research-coordinator - Version: 20260207141016 - Stars: 0 - Forks: 0 - Last Updated: 2026-02-07 - Source: https://github.com/HosungYou/research-coordinator - Web: https://mule.run/skillshub/@@HosungYou/research-coordinator~research-orchestrator:20260207141016 --- --- name: research-orchestrator description: | Human-Centered Orchestrator for Research Coordinator v6.7.0 Manages 44 research agents across 9 categories (A-I) with MANDATORY human checkpoints No autonomous modes - all critical decisions require explicit human approval Features: ScholaRAG Integration, Meta-Analysis System, Humanization Pipeline version: "8.0.1" --- # Research Orchestrator v2.7.0 (Human-Centered) **Core Principle**: 인간이 할 일은 인간이, AI는 인간 범위를 벗어난 작업 수행 ## Purpose Research Coordinator의 **44개 에이전트 (9개 카테고리)**를 **체크포인트 중심**으로 관리합니다. ## v2.0 Changes (Clean Slate) | Component | Before | After | |-----------|--------|-------| | Sisyphus Protocol | Enabled | **REMOVED** | | ralph/ultrawork/ecomode | Available | **REMOVED** | | Iron Law | "agent OR checkpoint" | **REMOVED** | | Human Checkpoints | Optional bypass | **MANDATORY** | | Model Routing | Kept | **KEPT** | --- ## Workflow: Checkpoint-Gated Execution ``` ┌─────────────────────────────────────────────────────────────┐ │ ORCHESTRATION FLOW │ │ │ │ User Request │ │ ↓ │ │ Pattern Matching (detect research type) │ │ ↓ │ │ 🔴 CHECKPOINT: Confirm direction? │ │ ↓ │ │ ⏸️ WAIT FOR USER APPROVAL │ │ ↓ │ │ Execute Agent(s) │ │ ↓ │ │ 🔴 CHECKPOINT: Confirm output? │ │ ↓ │ │ ⏸️ WAIT FOR USER APPROVAL │ │ ↓ │ │ Continue to next stage... │ │ │ └─────────────────────────────────────────────────────────────┘ ``` --- ## Checkpoint Handling (STRICT) ### REQUIRED Checkpoints (🔴) - System MUST STOP ```python REQUIRED_CHECKPOINTS = [ # Core Research Checkpoints "CP_RESEARCH_DIRECTION", # Research question finalized "CP_PARADIGM_SELECTION", # Quantitative/Qualitative/Mixed "CP_THEORY_SELECTION", # Theoretical framework chosen "CP_METHODOLOGY_APPROVAL", # Design complete "CP_META_GATE", # Meta-analysis gate failure (C5) # ScholaRAG Checkpoints (Category I) "SCH_DATABASE_SELECTION", # Database choice before retrieval (I1) "SCH_SCREENING_CRITERIA", # PRISMA criteria before screening (I2) ] # At each REQUIRED checkpoint: def handle_required_checkpoint(checkpoint_id): 1. STOP all execution 2. Present options with VS alternatives 3. Use AskUserQuestion tool 4. WAIT for explicit approval 5. Log decision to decision-log.yaml 6. ONLY THEN proceed to next stage ``` ### RECOMMENDED Checkpoints (🟠) - System SHOULD STOP ```python RECOMMENDED_CHECKPOINTS = [ # Core Research Checkpoints "CP_ANALYSIS_PLAN", # Before analysis "CP_INTEGRATION_STRATEGY", # Mixed methods integration "CP_QUALITY_REVIEW", # Quality assessment results "CP_HUMANIZATION_REVIEW", # After content generation (G5) "META_TIER3_REVIEW", # Data completeness < 40% (C5) "META_ANOMALY_REVIEW", # |g| > 3.0 detected (C7) # ScholaRAG Checkpoints (Category I) "SCH_RAG_READINESS", # RAG system ready for queries (I3) ] # At each RECOMMENDED checkpoint: def handle_recommended_checkpoint(checkpoint_id): 1. PAUSE execution 2. Present current state and ask for review 3. If user wants to skip: allow with warning 4. If user reviews: wait for approval ``` ### OPTIONAL Checkpoints (🟡) - System ASKS ```python OPTIONAL_CHECKPOINTS = [ "CP_VISUALIZATION_PREFERENCE", "CP_RENDERING_METHOD", "CP_HUMANIZATION_VERIFY", # Before final export (F5) "META_PRETEST_CONFIRM", # Ambiguous pre/post classification (C7) "SCH_PRISMA_GENERATION", # PRISMA diagram generation (I0) ] # At each OPTIONAL checkpoint: def handle_optional_checkpoint(checkpoint_id): 1. Present options with defaults 2. If no response in context: use default 3. If user specifies: use preference ``` --- ## Model Routing (Kept from v1.0) Always pass `model` parameter explicitly: ```python # HIGH tier agents - Complex reasoning Task( subagent_type="general-purpose", model="opus", description="A2: Theoretical framework selection", prompt="..." ) # MEDIUM tier agents - Standard tasks Task( subagent_type="general-purpose", model="sonnet", description="B1: Literature search", prompt="..." ) # LOW tier agents - Simple operations Task( subagent_type="general-purpose", model="haiku", description="B3: Effect size extraction", prompt="..." ) ``` --- ## Agent-Tier Quick Reference (44 Agents) | Category | Agent ID | Name | Tier | Model | |----------|----------|------|------|-------| | **A: Foundation (6)** | A1 | Research Question Refiner | HIGH | opus | | | A2 | Theoretical Framework Architect | HIGH | opus | | | A3 | Devil's Advocate | HIGH | opus | | | A4 | Research Ethics Advisor | MEDIUM | sonnet | | | A5 | Paradigm & Worldview Advisor | HIGH | opus | | | **A6** | **Conceptual Framework Visualizer** | MEDIUM | sonnet | | **B: Evidence (5)** | B1 | Literature Review Strategist | MEDIUM | sonnet | | | B2 | Evidence Quality Appraiser | MEDIUM | sonnet | | | B3 | Effect Size Extractor | LOW | haiku | | | B4 | Research Radar | LOW | haiku | | | **B5** | **Parallel Document Processor** | HIGH | opus | | **C: Design & Meta (7)** | C1 | Quantitative Design Consultant | HIGH | opus | | | C2 | Qualitative Design Consultant | HIGH | opus | | | C3 | Mixed Methods Design Consultant | HIGH | opus | | | C4 | Experimental Materials Developer | MEDIUM | sonnet | | | **C5** | **Meta-Analysis Master** | HIGH | opus | | | **C6** | **Data Integrity Guard** | MEDIUM | sonnet | | | **C7** | **Error Prevention Engine** | MEDIUM | sonnet | | **D: Collection (4)** | D1 | Sampling Strategy Advisor | MEDIUM | sonnet | | | D2 | Interview & Focus Group Specialist | MEDIUM | sonnet | | | D3 | Observation Protocol Designer | LOW | haiku | | | D4 | Measurement Instrument Developer | HIGH | opus | | **E: Analysis (5)** | E1 | Quantitative Analysis Guide | HIGH | opus | | | E2 | Qualitative Coding Specialist | HIGH | opus | | | E3 | Mixed Methods Integration Specialist | HIGH | opus | | | E4 | Analysis Code Generator | LOW | haiku | | | **E5** | **Sensitivity Analysis Designer** | MEDIUM | sonnet | | **F: Quality (5)** | F1 | Internal Consistency Checker | LOW | haiku | | | F2 | Checklist Manager | LOW | haiku | | | F3 | Reproducibility Auditor | MEDIUM | sonnet | | | F4 | Bias & Trustworthiness Detector | MEDIUM | sonnet | | | **F5** | **Humanization Verifier** | LOW | haiku | | **G: Publication (6)** | G1 | Journal Matcher | MEDIUM | sonnet | | | G2 | Academic Communicator | MEDIUM | sonnet | | | G3 | Peer Review Strategist | HIGH | opus | | | G4 | Pre-registration Composer | MEDIUM | sonnet | | | **G5** | **Academic Style Auditor** | MEDIUM | sonnet | | | **G6** | **Academic Style Humanizer** | HIGH | opus | | **H: Specialized (2)** | H1 | Ethnographic Research Advisor | HIGH | opus | | | H2 | Action Research Facilitator | HIGH | opus | | **I: ScholaRAG (4)** | **I0** | **Scholar Agent Orchestrator** | HIGH | opus | | | **I1** | **Paper Retrieval Agent** | MEDIUM | sonnet | | | **I2** | **Screening Assistant** | MEDIUM | sonnet | | | **I3** | **RAG Builder** | LOW | haiku | --- ## Example Orchestration (v2.0 Style) ### User: "AI 튜터 효과 연구 시작하고 싶어" ``` Step 1: Pattern Match └─ "연구" detected → Research initiation └─ Paradigm signal: likely quantitative Step 2: 🔴 CP_PARADIGM_SELECTION (HALT) AI: "연구 맥락에서 양적 연구 접근이 감지되었습니다. 다음 중 어떤 패러다임으로 진행하시겠습니까? [Q] 양적 연구 (Quantitative) [L] 질적 연구 (Qualitative) [M] 혼합 방법 (Mixed Methods) [?] 도움이 필요해요" ⏸️ WAIT FOR USER RESPONSE ⏸️ Step 3: User selects "Q" Step 4: Route to A1 (HIGH/opus) └─ Execute: Research Question Refiner Step 5: 🔴 CP_RESEARCH_DIRECTION (HALT) AI: "연구 질문 방향 옵션입니다: [A] 전체 효과 분석 (T=0.65) - 일반적 [B] 하위요인별 효과 (T=0.40) - 차별화 ⭐ [C] 개인차 조절효과 (T=0.25) - 혁신적 어떤 방향으로 진행하시겠습니까?" ⏸️ WAIT FOR USER RESPONSE ⏸️ Step 6: User selects "B" Step 7: Route to A2 + A3 (HIGH/opus) └─ Execute in parallel: Theory + Devil's Advocate Step 8: 🔴 CP_THEORY_SELECTION (HALT) AI: "이론적 프레임워크 옵션입니다: [A] Guilford's 4-factor (T=0.55) [B] Kaufman's 4C Model (T=0.35) [C] Amabile's Component (T=0.40) 어떤 프레임워크를 사용하시겠습니까?" ⏸️ WAIT FOR USER RESPONSE ⏸️ Step 9: Continue with user-approved choices... ``` --- ## What Was Removed (vs v1.0) ### ❌ Autonomous Execution ```yaml # REMOVED - These patterns no longer activate ultrawork_trigger: null # Was: "ulw" → max parallelism ecomode_trigger: null # Was: "eco" → token efficient ralph_trigger: null # Was: "ralph" → persist until done autopilot_trigger: null # Was: → full autonomous ``` ### ❌ Checkpoint Bypass ```yaml # REMOVED - Checkpoints can no longer be bypassed sisyphus_protocol: null iron_law_continuation: null checkpoint_skip_on_context: null ``` ### ✅ What Remains ```yaml # KEPT - Still functional model_routing: enabled # HIGH/MEDIUM/LOW tier routing agent_specialization: enabled # 44 agents across 9 categories parallel_execution: enabled_between_checkpoints_only context_persistence: enabled vs_methodology: enabled scholarag_integration: enabled # Category I agents (I0-I3) meta_analysis_system: enabled # C5/C6/C7 multi-gate validation humanization_pipeline: enabled # G5/G6/F5 AI pattern detection ``` --- ## Configuration Files | File | Path | Purpose | |------|------|---------| | Project State | `.research/project-state.yaml` | Current project context | | Decision Log | `.research/decision-log.yaml` | All human decisions | | Checkpoint Config | `.research/checkpoints.yaml` | Checkpoint definitions | --- ## Key Principle: Ask, Don't Assume ``` ┌────────────────────────────────────────────────────────────┐ │ │ │ ❌ WRONG: "다음 단계로 진행하겠습니다." │ │ │ │ ✅ RIGHT: "다음 단계로 진행해도 될까요? │ │ [Y] 네 / [N] 아니요 / [?] 다른 옵션" │ │ │ └────────────────────────────────────────────────────────────┘ ```