Notebooklm Research

Research Pipeline

When a user asks to research a topic, the agent runs the full pipeline autonomously from start to finish. The pipeline is split across files for readability, but it is one continuous flow. Never stop between steps.

[!IMPORTANT] This is one pipeline, not three optional workflows. Each workflow file has a checklist at the top. After completing each step, check the list and proceed to the next unchecked step. Never present a mid-pipeline summary and ask what to do next. The pipeline decides when to stop, not the agent.

How to execute

1. Read workflows/deep-research.md and execute Steps 1-7 using the checklist
2. Step 7 says "read workflows/report-handoff.md" - do that immediately
3. Execute Steps 8-10 using the checklist
4. Step 10 transitions to the minibook-pipeline extension automatically

Agent Behaviors

Core Rules

• Never blindly copy NotebookLM suggestions into code - evaluate first
• Cross-reference with existing project conventions and skills
• Flag conflicts if NotebookLM contradicts the project's patterns
• Cite the source: "Based on the NotebookLM research, I'm using X because..."
• Prefer project conventions over research suggestions unless the user wants to change
• Suggest NotebookLM for topics that benefit from deep, multi-source research
• Never construct external URLs by guessing from local data (OS username, folder names, file paths). Always verify from the actual source: git remote get-url origin for repo URLs, gh api user -q .login for GitHub usernames

Research Folder Structure

The agent MUST maintain a research/ folder at the project root whenever NotebookLM is used.

project-root/
  research/
    index.md          <- notebook index
    reports/
      [slug]_briefing.md
      [slug]_briefing.md

When to create/update research/index.md:

• After any deep research workflow completes
• After creating/deleting notebooks or adding/removing sources
• On first use of NotebookLM in a new project (scan ALL existing notebooks)

How to build:

1. Create research/ and research/reports/ directories if they don't exist
2. notebook_list to get all notebooks
3. For each: title, URL, source count, creation date
4. notebook_describe for AI-generated summaries
5. Write to research/index.md

Format:

# Research Index

> Auto-generated overview of all research notebooks.
> Last updated: [date]

---

## [Notebook Title]

- **ID:** `[notebook_id]`
- **Sources:** [count]
- **Report:** "[report title]" (if generated)
- **URL:** [url]
- **Key sources:** [list of major source institutions]
- **Consensus findings:** [brief summary]
- **Key conflict:** [brief summary]

---

*[N] notebooks / [total] sources*

Formatting rules: One notebook per section, --- dividers, metadata on separate lines, consensus/conflict summaries when available, italic footer with totals.

[!IMPORTANT] This is a living document. Update in place - don't recreate. Always include ALL notebooks.

Source: nichochar/notebooklm-mcp-cli