claude-world/notebooklm-skill

• **Archetype**: domain-skill-pack
• **Stars**: 167
• **Last pushed**: 2026-03-24
• **License**: MIT
• **Discovered**: 2026-04-12
• **Skills found**: 1 (with MCP server providing 13 tools)
• **Source**: gh-search

Summary

A full-stack research-to-content pipeline bridging Google NotebookLM (free research engine with cited Q&A and 10 artifact types) with Claude (content creation). Delivered as a Claude Code skill, MCP server (13 tools), and CLI with named pipeline workflows. The key insight: NotebookLM handles research indexing at zero cost, Claude handles only the creative writing.

Assessment

The multi-phase pipeline pattern (Ingest -> Synthesize -> Create -> Publish) with clear phase boundaries is a textbook orchestration template. The MCP server architecture (FastMCP, 13 tools, async context manager, fuzzy resolution) is a strong reference implementation. The SKILL.md is one of the best-authored in the wild (800+ lines, phase-by-phase instructions, CLI examples for every operation). The dual-interface pattern (CLI + MCP + Python API from one implementation) is worth noting. Depends on notebooklm-py library.

Extraction Priority

• **High**
• Rationale: Most sophisticated Claude Code skill found. The pipeline pattern, MCP architecture, and SKILL.md authoring quality are all directly extractable as process templates and skill patterns.

---

Processes

- Source: SKILL.md pipeline section and scripts/pipeline.py - Placement: specializations/shared/research-pipeline - Inputs: source URLs/PDFs/YouTube, research questions, output format (article/social/slides) - Outputs: research findings with citations, content drafts, optional social posts - Complexity: complex - Notes: Uses NotebookLM as zero-cost research backend; Claude only handles creative writing

• **research-to-content-pipeline**: 4-phase flow: Ingest sources -> Synthesize via cited Q&A and artifact generation -> Claude creates original content -> Optional social publishing

- Source: SKILL.md research section - Placement: specializations/shared/web-research - Inputs: notebook ID, research query, mode (fast 10-30s / deep 1-5min) - Outputs: research report (Markdown) + discovered source URLs - Complexity: moderate - Notes: Discovery amplification pattern -- finds sources beyond what user provided

• **deep-web-research**: Use NotebookLM's web/Drive research to auto-discover and import sources the user didn't know existed

- Source: SKILL.md artifact section - Placement: specializations/shared/artifact-generation - Inputs: notebook ID, artifact type, output directory, language - Outputs: downloaded artifact files - Complexity: moderate - Notes: Polling pattern for external async service; documents known limitations (infographic unreliable)

• **async-artifact-generation**: Generate and download 10 artifact types (podcast, video, slides, report, quiz, flashcards, mind map, infographic, data table, study guide) with async polling

Plugin Ideas

- What install.md would do: Check if notebooklm-py and MCP server are available, interview user about research needs (competitive analysis, market research, technical research, content creation), copy research pipeline processes into .a5c/processes/research/, create /research and /deep-research slash commands, configure MCP server connection in .claude/settings.json - Processes it would copy: research-to-content-pipeline, deep-web-research, artifact-generation - Configs/hooks it would create: .a5c/commands/research.md, .a5c/commands/deep-research.md, MCP server configuration, output directory structure for research artifacts - Source evidence: The repo's complete 4-phase pipeline with named workflows and MCP tool surface

• **research-automation**: A babysitter plugin that sets up automated research workflows for a project

Library Mapping

Plugin Marketplace Mapping

Implicit Procedural Knowledge

- Source: SKILL.md phase descriptions and scripts/pipeline.py named workflows - Placement: specializations/shared/research-pipeline - Why codify: The pipeline has 4 clear phases with typed inputs/outputs and async polling between them. Named workflows (research-to-article, research-to-social, trend-to-content, batch-digest) show different compositions of the same phases. - Sketch: Phase 1 (shell: create notebook, add sources, wait for processing), Phase 2 (agent: formulate research questions, run cited Q&A), Phase 3 (agent: Claude writes original content from research), Breakpoint (human reviews content), Phase 4 (optional shell: publish to platforms)

• **research-pipeline-orchestration**: Full pipeline from source ingestion to published content