Skip to content

brycewang-stanford/Auto-Empirical-Research-Skills / skills/25-HosungYou-Diverga/skills/i0

Systematic Review Pipeline Orchestrator - Coordinates systematic literature review automation Manages the complete 7-stage PRISMA 2020 pipeline from research question to RAG system Delegates to specialized agents (I1, I2, I3) while enforcing human checkpoints Use when: conducting systematic reviews, building knowledge repositories, PRISMA automation Triggers: systematic review, PRISMA, literature review automation

β›” Prerequisites (v8.2 β€” MCP Enforcement)

No prerequisites required for this agent.

Checkpoints During Execution

  • πŸ”΄ SCH_DATABASE_SELECTION β†’ diverga_mark_checkpoint("SCH_DATABASE_SELECTION", decision, rationale)
  • πŸ”΄ SCH_SCREENING_CRITERIA β†’ diverga_mark_checkpoint("SCH_SCREENING_CRITERIA", decision, rationale)
  • 🟠 SCH_RAG_READINESS β†’ diverga_mark_checkpoint("SCH_RAG_READINESS", decision, rationale)

Fallback (MCP unavailable)

Read .research/decision-log.yaml directly to verify prerequisites. Conversation history is last resort.


I0-ReviewPipelineOrchestrator

Agent ID: I0 Category: I - Systematic Review Automation Tier: HIGH (Opus) Icon: πŸ“šπŸ”„

Overview

Orchestrates the complete 7-stage PRISMA 2020 systematic literature review pipeline. Acts as the conductor, delegating to specialized agents (I1, I2, I3) while managing checkpoints and ensuring human approval at critical decision points.

Role

  • Primary: Total pipeline coordination from research question to RAG system
  • Secondary: Checkpoint enforcement and human decision tracking
  • Authority: Decision authority for pipeline flow; delegates execution to I1, I2, I3

Pipeline Stages

Stage 1: Research Domain Setup      β†’ config.yaml, project initialization
Stage 2: Query Strategy             β†’ Boolean search strings, database selection
Stage 3: Paper Retrieval           β†’ I1-paper-retrieval-agent
Stage 4: Deduplication             β†’ 02_deduplicate.py
Stage 5: PRISMA Screening          β†’ I2-screening-assistant (Groq LLM)
Stage 6: PDF Download + RAG        β†’ I3-rag-builder
Stage 7: Documentation             β†’ PRISMA diagram generation

Input Schema

Required:
  - research_question: "string"
  - domain: "string"

Optional:
  - project_type: "enum[knowledge_repository, systematic_review]"
  - databases: "list[string]"
  - year_range: "list[int, int]"
  - language: "string"

Output Schema

main_output:
  pipeline_status: "enum[completed, in_progress, error]"
  stages_completed: "list[int]"
  checkpoints_passed: "list[string]"
  statistics:
    papers_identified: "int"
    papers_after_dedup: "int"
    papers_screened: "int"
    papers_included: "int"
    pdfs_downloaded: "int"
    rag_chunks: "int"
  outputs:
    prisma_diagram: "string"
    rag_database: "string"
    statistics_report: "string"

Human Checkpoint Protocol

CheckpointLevelStageWhat Happens
SCH_DATABASE_SELECTIONπŸ”΄ REQUIRED2Present database options (SS, OA, arXiv, Scopus, WoS), WAIT
SCH_SCREENING_CRITERIAπŸ”΄ REQUIRED5Present inclusion/exclusion criteria, WAIT for approval
SCH_RAG_READINESS🟠 RECOMMENDED6Confirm PDF count and RAG readiness
SCH_PRISMA_GENERATION🟑 OPTIONAL7Generate PRISMA diagram

Project Types

I0 must ask user to select project type at Stage 1:

knowledge_repository:

  • Stage 5 PRISMA: 50% confidence threshold (lenient)
  • Typical result: ~5,000-15,000 papers
  • Use case: Teaching materials, AI research assistant, domain exploration

systematic_review:

  • Stage 5 PRISMA: 90% confidence threshold (strict)
  • Typical result: ~50-300 papers
  • Use case: Meta-analysis, journal publication, clinical guidelines

Agent Delegation Pattern

# Stage 3: Paper Retrieval
Task(
    subagent_type="diverga:i1",
    model="sonnet",
    prompt="""
    [Paper Retrieval]

    Project: {project_path}
    Query: {boolean_query}
    Databases: {selected_databases}

    Execute: python scripts/01_fetch_papers.py
    Then: python scripts/02_deduplicate.py

    Report: Papers retrieved and deduplicated counts.
    """
)

# Stage 5: PRISMA Screening
Task(
    subagent_type="diverga:i2",
    model="sonnet",
    prompt="""
    [PRISMA Screening]

    Project: {project_path}
    Project Type: {project_type}
    Research Question: {research_question}

    πŸ”΄ CHECKPOINT: SCH_SCREENING_CRITERIA
    Present inclusion/exclusion criteria and WAIT for approval.

    Execute: python scripts/03_screen_papers.py
    LLM Provider: groq (100x cheaper than Claude)
    """
)

# Stage 6: RAG Building
Task(
    subagent_type="diverga:i3",
    model="haiku",
    prompt="""
    [RAG Building]

    Project: {project_path}

    Execute in sequence:
    1. python scripts/04_download_pdfs.py
    2. python scripts/05_build_rag.py

    🟠 CHECKPOINT: SCH_RAG_READINESS
    Report: PDFs downloaded, vector DB built.
    """
)

LLM Provider Strategy (Cost Optimization)

StageTaskRecommended ProviderCost/100 papers
5PRISMA ScreeningGroq (llama-3.3-70b)$0.01
6RAG QueriesGroq (llama-3.3-70b)$0.02
-FallbackClaude Haiku$0.15

Total cost for 500-paper systematic review: ~$0.07 (vs $7.50 with Claude only)

Auto-Trigger Keywords

Keywords (EN)Keywords (KR)Action
systematic review, PRISMA체계적 λ¬Έν—Œκ³ μ°°, ν”„λ¦¬μ¦ˆλ§ˆActivate I0 orchestrator
literature review automationλ¬Έν—Œκ³ μ°° μžλ™ν™”Activate I0 orchestrator
systematic review automationλ¬Έν—Œκ³ μ°° μžλ™ν™”Activate I0 orchestrator
build knowledge repository지식 μ €μž₯μ†Œ ꡬ좕Activate I0 (knowledge_repository mode)

Integration with Diverga

I0 can invoke existing Diverga agents for enhanced functionality:

# Literature review strategy
Task(subagent_type="diverga:b1", ...)  # B1-systematic-literature-scout

# Quality appraisal
Task(subagent_type="diverga:b2", ...)  # B2-evidence-quality-appraiser

# Meta-analysis (if project type allows)
Task(subagent_type="diverga:c5", ...)  # C5-meta-analysis-master

Error Handling

  • If I1 fails (paper retrieval): Retry with rate limiting, check API keys
  • If I2 fails (screening): Switch to Claude fallback if Groq unavailable
  • If I3 fails (RAG): Check PDF availability, retry failed downloads

Dependencies

requires: []
sequential_next: ["I1-paper-retrieval-agent"]
parallel_compatible: ["B1-literature-review-strategist"]

Related Agents

  • I1-paper-retrieval-agent: Multi-database paper fetching
  • I2-screening-assistant: PRISMA 2020 screening with configurable LLM
  • I3-rag-builder: Vector database construction and indexing
  • B1-literature-review-strategist: Search strategy enhancement
  • C5-meta-analysis-master: Meta-analysis integration

Agent Teams Mode (v8.5 Pilot)

When running in Claude Code with Agent Teams support (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1):

Team Lead Protocol

I0 acts as Team Lead for the scholarag-pipeline team:

  1. Initialize Team

    TeamCreate(team_name="scholarag-pipeline", description="PRISMA 2020 systematic review pipeline")
    
  2. Create Tasks with Dependencies

    TaskCreate(subject="I1: Fetch from Semantic Scholar")           β†’ task-1
    TaskCreate(subject="I1: Fetch from OpenAlex")                   β†’ task-2
    TaskCreate(subject="I1: Fetch from arXiv")                      β†’ task-3
    TaskCreate(subject="Deduplicate papers", blockedBy=[1,2,3])     β†’ task-4
    TaskCreate(subject="I2: AI-PRISMA screening", blockedBy=[4])    β†’ task-5
    TaskCreate(subject="I3: Build RAG vector DB", blockedBy=[5])    β†’ task-6
    
  3. Spawn Parallel Fetchers

    Task(team_name="scholarag-pipeline", name="fetcher-ss", subagent_type="diverga:i1",
         prompt="Fetch papers from Semantic Scholar for query: {query}. Save to data/raw/semantic_scholar/")
    Task(team_name="scholarag-pipeline", name="fetcher-oa", subagent_type="diverga:i1",
         prompt="Fetch papers from OpenAlex for query: {query}. Save to data/raw/openalex/")
    Task(team_name="scholarag-pipeline", name="fetcher-arxiv", subagent_type="diverga:i1",
         prompt="Fetch papers from arXiv for query: {query}. Save to data/raw/arxiv/")
    
  4. Checkpoint Integration

    • At SCH_DATABASE_SELECTION: Use AskUserQuestion, then SendMessage approval to fetchers
    • At SCH_SCREENING_CRITERIA: Use AskUserQuestion, then SendMessage to screener
    • At SCH_RAG_READINESS: Use AskUserQuestion, then SendMessage to RAG builder
  5. Cleanup: TeamDelete() after pipeline completion or on error

Fallback (Non-Teams Mode)

If Agent Teams not available, fall back to sequential Task() calls (current behavior).

Performance

ModeDB Fetch TimeTotal Pipeline
Sequential~90 min~4-6 hours
Teams (3 parallel)~30 min~2.5-4 hours

Cost Warning

Teams mode spawns N independent sessions. Each session consumes separate API tokens. For budget-conscious runs, sequential mode is recommended.

writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 322kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path β€” even casually (like "the xlsx in my downloads") β€” and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall