Notebooklm Content

NotebookLM Content Skill

Create high-quality infographics and slide decks from your blog posts using Google NotebookLM, branded to CryptoFlex LLC standards.

When This Skill Activates

• User says /notebooklm-content
• User asks to "create an infographic from a blog post"
• User asks to "make slides from a blog post"
• User asks to "generate NotebookLM content"
• User mentions creating visual assets from blog content

Prerequisites

• NotebookLM MCP server (notebooklm-mcp-cli) configured in ~/.claude.json
• Authenticated via nlm login with chrisjohnson@cryptoflexllc.com
• First-time setup: run nlm login in your terminal to authenticate

Usage

/notebooklm-content <blog-post-slug-or-path> [options]

Arguments

• <blog-post-slug-or-path>: The blog post to create content from. Can be:
  • A slug (e.g., my-first-24-hours-with-claude-code)
  • A full path to an MDX file
  • "latest" to use the most recent published post

Options

• --type infographic (default): Generate a branded infographic (PNG)
• --type slides: Generate a branded slide deck (PDF + PPTX)
• --type both: Generate both infographic and slide deck

Video Overviews (new in notebooklm-mcp-cli 0.7.0) are also available, including the cinematic format driven by a full creative brief. This skill's brand-QA pipeline is tuned for static assets (infographics, slides), so video generation is not wired into --type here. To produce a Video Overview from a notebook, use the general notebooklm-assistant agent (studio_create with artifact_type="video", video_format="cinematic", brief via focus_prompt).

• --orientation landscape|portrait|square: Infographic orientation (default: landscape)
• --detail concise|standard|detailed: Infographic detail level (default: detailed)
• --style professional|editorial|scientific: Infographic visual style (default: professional)
• --slide-format detailed|presenter: Slide deck format (default: detailed)

Examples

/notebooklm-content my-first-24-hours-with-claude-code --type both
/notebooklm-content latest --type infographic --style editorial
/notebooklm-content ~/GitProjects/cryptoflexllc/src/content/blog/building-blog-with-ai.mdx --type slides

What Happens

1. Reads the specified blog post from the cryptoflexllc repo
2. Uses NotebookLM MCP tools to create a notebook and add the post content as a source
3. Primes the notebook with CryptoFlex LLC branding guidelines via MCP query
4. Generates the requested content type(s) via MCP studio tools (5-15 min per asset)
5. Downloads output to ~/GitProjects/cryptoflexllc/content-assets/notebooklm/
6. Runs QA review: spelling, accuracy, brand compliance, DLP scanning
7. Revises if needed (max 2 cycles)
8. Reports results with file paths and QA summary

Output Location

Generation output lands in the gitignored working area:

~/GitProjects/cryptoflexllc/content-assets/notebooklm/
  <post-slug>/
    infographic.png
    slides.pdf
    slides.pptx
    slides/
      slide-01.png
      slide-02.png
      ...

Curated, embed-ready assets are then copied into the published tree under public/blog/<slug>/. See "Embedding Slides Into the Post" below.

Implementation

This skill delegates to the notebooklm-content agent:

Agent(
  prompt="Follow the instructions in ~/.claude/agents/notebooklm-content.md.
         Blog post: <resolved-path>
         Content type: <type>
         Options: <options>",
  subagent_type="general-purpose",
  model="sonnet",
  name="notebooklm-content"
)

Authentication

If not yet logged in or cookies have expired, run in your terminal:

nlm login

This opens a Chromium browser window for Google sign-in with chrisjohnson@cryptoflexllc.com. Cookie sessions expire every 2-4 weeks and require re-authentication.

Relationship to Blog Post Pipeline

This skill is separate from the /blog-post command. It does not run as part of the blog production pipeline. You invoke it independently after a blog post is drafted or published.

The slide deck is first-class article content, not external-only collateral. The primary use is to curate select slides and embed them directly into the post (see "Embedding Slides Into the Post" below). Social and presentation reuse is secondary:

• Primary: Embed curated slides into the blog post as in-line article visuals, and link the full deck near the end of the post.
• Secondary: LinkedIn posts and social media, presentation materials, internal documentation.

The blog-post captain does not call this skill. You invoke it independently when you want visual content derived from a post. When a post has a companion deck, the captain incorporates the curated slides per this skill's convention rather than treating them as external-only.

Embedding Slides Into the Post

After generation and QA, curate select slides into the article. Slides are primary article content: choose the few that add visual value beyond the prose and callouts.

1. Select, do not dump. Embed only the 4-6 strongest slides, each mapping to a distinct article section. Skip the title slide (the infographic is already the cover), any before/after slide when the post already has an equivalent custom SVG diagram or comparison table, and pure-takeaway or summary slides that merely restate existing callouts. Quality over completeness.
2. Semantic filenames in public/. Copy each chosen slide from content-assets/notebooklm/<slug>/slides/slide-NN.png to public/blog/<slug>/<semantic-name>.png using a descriptive kebab-case name, not slide-NN. The infographic remains the cover image at public/blog/<slug>/infographic.png.
3. Embed with a lead-in. In the post .mdx, place each slide as a plain markdown image ![rich descriptive alt text](/blog/<slug>/<semantic-name>.png) in its relevant section, preceded by a one-sentence in-voice prose lead-in that sets up what the reader is about to see. Alt text must be genuinely descriptive for accessibility and SEO, matching the pattern in existing posts like notebooklm-content-pipeline and home-network-mission-control-dashboard-log-lake-panel.
4. Link the full deck. Copy slides.pdf to public/blog/<slug>/slides.pdf and link it once near the end of the post: "If you'd rather skim this as slides, the deck is here: slide deck (PDF)."

Limitations

• NotebookLM generation takes 5-15 minutes per asset
• Infographic style is influenced but not fully controlled by instructions (NotebookLM makes its own design choices)
• Cookie auth expires every 2-4 weeks (run nlm login to refresh; nlm login --check verifies the current session). If the MCP reports auth_status: stale, re-auth; unverified is a transient network error, not an auth failure.
• Uses reverse-engineered Google APIs via notebooklm-mcp-cli (may break without notice). Keep current with uv tool upgrade notebooklm-mcp-cli, then reconnect the notebooklm MCP server.
• Maximum 50 sources per notebook
• Free tier rate limit: approximately 50 queries per day

Known stall patterns

• Slide decks can appear stuck at in_progress for 25+ minutes before completing. This is the expected pattern, not a failure. Do not cancel or retry. Keep polling. Most common on first-time or large notebook runs; subsequent requests tend to complete faster.
• download_artifact can succeed while studio_status still reports in_progress. Observed on a 71+ minute wait (and a separate 75 minute case). Do not gate the download on a completed status. Call download_artifact periodically; if it returns bytes, accept the artifact and move on. If it errors, keep polling and retry. Have a manual fallback (custom SVG diagram component) ready so a post can ship even without the NotebookLM asset.

Troubleshooting

Python Virtualenv Orphaned After uv/Python Updates

If the MCP server reports ENOENT or fails to start after updating the uv tool or its linked Python interpreter, the notebooklm-mcp-cli binary itself may still exist on disk — the virtualenv uv manages for it is what's orphaned. uv tool upgrade notebooklm-mcp-cli (see Limitations above) does not fix this case, since there's no working environment left to upgrade.

Fix: reinstall the tool's virtualenv, then reconnect the MCP server:

uv tool install --force notebooklm-mcp-cli