main_scientific_research

Prompt file imported from albertozafra7/MAESTRO (.github/prompts/main_scientific_research.prompt.md). Fill in {{topic}} before use. Copyright stays with the author.

Initialize a new literature review project using the Research Planner and then execute the initial ingestion pipeline using the orchestrator.

Step 1 — Workspace Initialization (Planner Agent)

Use the "Research Planner and Workspace Initializer Agent" to:

• Create a new research workspace for the topic: "{{topic}}"

• Generate the full directory structure under: /research/<topic_name>/

• Output the mkdir command and a description of the structure

• Copy or replicate the LaTeX template from templates/latex-ieee-review/ into: literature/latex/ (rename main.tex → main_review.tex; copy sections/ subdirectory)

• Report:

  • what was created
  • what is missing (if any)
  • readiness for next steps

Step 2 — Planning and Delegation (Orchestrator Agent)

Once the workspace is initialized, hand off to the "Literature Review Orchestrator".

The orchestrator must:

1. Analyze the topic
2. Define the initial workflow plan
3. Identify:
   • required inputs (PDFs, keywords, etc.)
   • active agents for this stage
4. Produce a pipeline plan limited to the initial ingestion phase

Output format:

• task breakdown
• delegation per agent
• expected outputs
• file locations

Do NOT execute ingestion yet.

Step 4 — PDF Ingestion Execution

Now execute the ingestion stage.

Use the "PDF Ingestion and Markdown Conversion Agent" with:

Input folder: /research/<topic_name>/literature/pdfs/

The agent must:

• read all PDFs in the folder
• convert each into Markdown
• extract metadata conservatively
• store results in:
  • /research/<topic_name>/literature/markdown/
  • /research/<topic_name>/literature/metadata/

If no PDFs are found:

• stop execution
• report clearly
• suggest next actions (e.g., add PDFs or run discovery agent)

Step 5 — Update Spreadsheet

After ingestion, use the "Literature Spreadsheet Agent" to:

• read Markdown + metadata outputs
• generate a structured spreadsheet dataset

Store outputs in:

• /research/<topic_name>/literature/spreadsheet/literature_review.csv
• /research/<topic_name>/literature/spreadsheet/literature_review.json

Constraints:

• no fabricated metadata
• missing fields remain empty

Step 6 — Final Status Report

The orchestrator must summarize:

• ✅ completed steps
• 📁 files created
• ⚠️ missing inputs (if any)
• ▶️ recommended next steps

Suggested next steps may include:

• running the Paper Discovery Agent
• running Trend Analysis
• starting Visual Synthesis
• drafting LaTeX

Execution Order

Strict order:

1. Planner Agent
2. Orchestrator (planning mode)
3. Paper Discovery Agent (optional, if no PDFs)
4. PDF Ingestion Agent
5. Spreadsheet Agent
6. Trend Analysis Agent
7. Visual Synthesis Agent
8. Scientific Writer Agent
9. Orchestrator (summary mode)

Execution Mode

Follow strict agent delegation:

• Never merge roles
• Always call the correct agent
• Stop on missing inputs instead of guessing
• Do not skip steps.
• Do not merge agent roles.
• Ask for feedback after each stage