577206/knowledge-forge/notebooklm-obsidian-second-brain

NotebookLM × Obsidian Second Brain

Use this skill for the workflow:

Capture → Digest → Distill → Apply → Review → Archive

NotebookLM is the temporary reading room, Obsidian/Markdown is the long-term vault, and the agent is the controlled organizer.

Hard boundaries

• notebooklm-py is unofficial and uses undocumented Google NotebookLM APIs.
• Never ask the user to paste storage_state.json, cookies, Google tokens, or browser session data into chat.
• Never commit .notebooklm/, copied storage_state.json, .env, raw private documents, or generated private vault notes.
• Require explicit user confirmation before uploading private/local documents to NotebookLM.
• Require explicit confirmation before deleting notebooks/sources/artifacts, changing sharing settings, creating public links, or downloading artifacts into a project folder.
• Prefer personal/local workflows. Do not package this as a SaaS that stores other people's Google sessions.

Install / verify

Project-local venv example:

python -m venv .venv-notebooklm
.\.venv-notebooklm\Scripts\python.exe -m pip install -U pip
.\.venv-notebooklm\Scripts\python.exe -m pip install "notebooklm-py[browser]"
.\.venv-notebooklm\Scripts\python.exe -m playwright install chromium
.\.venv-notebooklm\Scripts\notebooklm.exe --help

Auth must be performed by the user:

.\.venv-notebooklm\Scripts\notebooklm.exe login
.\.venv-notebooklm\Scripts\notebooklm.exe auth check --test --json

Treat auth as valid only if JSON contains:

{
  "status": "ok",
  "checks": { "token_fetch": true }
}

Safe CLI commands

Read/status commands are normally safe:

notebooklm auth check --test --json
notebooklm list --json
notebooklm status
notebooklm source list
notebooklm artifact list
notebooklm metadata --json

Write/external commands require confirmation:

notebooklm create "Notebook name"
notebooklm source add "file-or-url"
notebooklm ask -n <notebook_id> "question"
notebooklm generate quiz -n <notebook_id>
notebooklm download quiz -n <notebook_id> --format markdown output.md

Avoid shared context in parallel agents. Prefer explicit -n <notebook_id> instead of relying on notebooklm use.

Agent workflow

1. Confirm scope:

• source files / URLs
• target vault folder
• allowed outputs
• forbidden directories/files
• review criteria

1. Capture:

• create/select NotebookLM notebook
• add only approved sources

1. Digest:

• ask NotebookLM for summaries, questions, key themes, gaps
• optionally generate flashcards/quizzes/mind maps

1. Distill:

• do not paste NotebookLM output raw
• rewrite into durable Markdown: MOC, concept cards, source cards, tasks, open questions

1. Apply:

• produce study plan / research plan / project handoff / quiz set

1. Review:

• mark uncertain claims
• keep source references
• list what needs human checking

1. Archive:

• leave raw imports or generated drafts in inbox/staging unless user approves promotion

Obsidian output pattern

Write to a reviewable inbox/staging area first:

<VAULT>/00_Inbox/
<VAULT>/10_Schoolwork/<Course>/
<VAULT>/20_Research/<Topic>/
<VAULT>/40_Projects/<Project>/AI_Handoff/

Each generated note should include:

---
type: notebooklm-digest
status: draft
source_tool: notebooklm-py
created: YYYY-MM-DD
review_required: true
---
 
# Title
 
## Source scope
 
## Summary
 
## Key concepts
 
## Evidence / citations to verify
 
## Flashcards or quiz items
 
## Open questions
 
## Next actions

MVP acceptance

A minimal successful run means:

• notebooklm CLI installed and --help works
• user completed notebooklm login
• auth check --test --json passes
• one test notebook is created or selected
• one approved source is added
• one question is asked successfully
• one Markdown digest is written into the configured vault inbox/staging folder
• no credentials or private source files are committed