spread1000-research-planner

name	spread1000-research-planner
description	Formulates an AI for Science research plan. Proposes how to integrate AI into the researcher's domain through web research and literature survey, and generates a concrete 研究計画書 (research plan document). Use when user doesn't know how to apply AI to their 研究テーマ (research theme), wants to create a 研究プラン (research plan), or wants to learn about AI for Science latest trends (最新動向).

Research Planner

Formulate the optimal AI-powered research plan for the researcher's domain.

Use This Skill When

• The research theme is decided but the user doesn't know how to leverage AI
• The user wants to survey AI for Science success stories and latest trends
• The user wants to create a 研究計画書 for SPReAD application

Required Inputs

• Researcher's domain (e.g., materials science, life science, meteorology)
• Research theme or problem overview
• Current research methods / data overview (optional)

Workflow

1. Hearing: Confirm the researcher's field, theme, and current challenges
2. Web Search (MCP: github-mcp-server) (primary broad survey): Use the github-mcp-server search tools to gather a wide overview first:
   • Latest AI for Science trends and success cases in the researcher's domain
   • Azure AI Foundry model catalog (Aurora, MatterGen, BioEmu, Phi-4, etc.)
   • Microsoft Research AI for Science pages
   • JST site for latest SPReAD solicitation information
   • Key conferences / journals (NeurIPS, ICML, Nature, Science, etc.) for recent highlights

   Record all URLs, titles, and key findings from Web Search results.

3. ToolUniverse MCP Research (detailed literature deep-dive — requires tooluniverse MCP server): Using keywords and gaps identified in Step 2, run targeted scientific database queries:

   # Discover relevant tools for the research domain
   find_tools("literature search AI for Science")
   find_tools("<researcher's domain> database")
   
   # Search literature for AI applications in the field
   execute_tool("PubMed_search_articles",
     {"query": "<domain> AI machine learning deep learning", "max_results": 10})
   execute_tool("ArXiv_search_papers",
     {"query": "<domain> AI for Science foundation model", "max_results": 10})
   execute_tool("BioRxiv_search",
     {"query": "<domain> AI", "max_results": 5})   # for life science
   execute_tool("SemanticScholar_search_papers",
     {"query": "AI for Science <domain> 2024 2025", "max_results": 10})
   
   # Domain-specific database tools (examples — discover with find_tools first)
   execute_tool("UniProt_search_proteins", ...)         # life science / structural biology
   execute_tool("ChEMBL_search_compounds", ...)         # chemistry / drug discovery
   execute_tool("PubChem_search", ...)                  # chemistry
   execute_tool("ClinicalTrials_search", ...)           # clinical research
   execute_tool("EuropePMC_search", ...)                # broader literature
   

   If tooluniverse MCP is unavailable, document the reason and continue with Web Search results only.

4. AI Strategy Formulation:
   • Map AI methods to research challenges
   • Clarify expected outcomes and breakthroughs
   • Estimate required computational resources (GPU, storage, network, etc.)
5. Research Plan Generation: Save as output/{project-name}/phase0-research-plan.md
   • Reuse assets/research-plan-template.md when producing the research plan
   • Section 4 (文献調査): Copy ToolUniverse execute_tool results verbatim — do NOT paraphrase or fill from memory
   • Section 5 (関連事例): Select ≥3 papers from Section 4. Each entry must include DOI/URL and source tool name
6. Review: Verify technical feasibility of the research plan

Deliverables

• output/{project-name}/phase0-research-plan.md: AI-powered research plan (complete)
  • Section 4「文献調査」: ToolUniverse クエリ記録 + 取得文献リスト（≥5 件、DOI/URL・取得元ツール必須）
  • Section 5「関連事例」: Section 4 から選択した ≥3 件の先行研究（DOI/URL と関連性を明記）
• output/{project-name}/phase0-research-survey.md: Literature & database research results summary (Include ToolUniverse query results: paper titles, DOIs, key findings from PubMed/ArXiv/etc.)

Quality Gates

• ToolUniverse MCP was invoked and results are referenced (or fallback reason stated)
• Section 4「文献調査」のクエリ記録表が埋まっており、取得文献リストに ≥5 件記載されている
• 各文献エントリに DOI / URL と取得元ツール名が記載されている（LLM 補完禁止）
• Section 5「関連事例」が Section 4 の取得文献から ≥3 件選択されており、DOI/URL が一致している
• Mapping between research theme and AI methods is clearly described
• Required computational resources are quantitatively estimated
• Research schedule (milestones) is included
• Novelty and innovativeness in the AI for Science context is explained

Gotchas

• AI method selection strongly depends on research data characteristics. Always confirm data format (images, time series, text, 3D structures, etc.)
• Do not reflect overblown expectations like "AI can do anything" in the research plan. Explicitly state technical constraints
• Compute resource estimates must be concretely calculated from training data size, model scale, and number of training iterations
• The appropriate framework (PyTorch/TensorFlow/JAX, etc.) varies by research field

Validation Loop

1. Generate the research plan
2. Check:
   • Did ToolUniverse MCP return results? (If not, verify tooluniverse server is running)
   • Is the mapping between AI methods and research challenges logical?
   • Are compute resource estimates realistic?
   • Is the schedule feasible?
   • Are citations traceable (DOI / URL)?
3. If any check fails:
   • Identify and fix the relevant section
   • Obtain supplementary information via ToolUniverse or web research
   • Re-verify
4. Finalize deliverables only after all gates pass

ToolUniverse MCP Availability Check

Before starting Step 2, verify the MCP server is reachable:

• Call list_tools or grep_tools("PubMed") to confirm ToolUniverse is active.
• If the call fails or the server is not configured, log a warning in the survey file and proceed with web research as fallback.
• Never silently skip ToolUniverse — always document whether it was used.