NotebookLM — manual setup for project work

There is no automated NotebookLM integration in this repo’s workflow. docs/TOOLING.md’s tool evaluation assessed the automation path — PleasePrompto/notebooklm-skill — and found it bootstraps a stealth, anti-bot-evading Chrome instance with --no-sandbox and persists your Google session cookies to disk: a credential-bearing network round-trip, the sharpest violation of this project’s on-device-by-default doctrine (docs/PRIVACY.md) in the current tooling landscape.

Status update 2026-07-17: on the owner’s explicit override, that skill was installed globally (files-only — its pip/Chrome bootstrap runs on the owner’s first use) for personal use outside this project; see docs/TOOLING.md → “Global skills — installed 2026-07-17” (second wave).

Status update 2026-08-12: the skill was later moved out of the global skills directory into the parked-skills tree, and then removed altogether during a tooling audit, because a second, overlapping integration had been wired in the meantime: the notebooklm MCP server (notebooklm-mcp, registered user-globally in ~/.claude.json). The two duplicated each other, and the MCP costs no per-session context because its tools load on demand. The MCP carries the same browser-automation profile the assessment above rejected — a headless, stealth-enabled Chrome session against a Google login — so it remains an owner-override tool for personal use only, and it is currently unauthenticated (get_health reports authenticated: false, zero notebooks), therefore inert.

The project rule is unchanged through all of this: SenseBridge work uses the manual, five-minute browser workflow below — no scraping, no stored credentials — and neither the skill nor the MCP is ever invoked from a SenseBridge session.

Setup (manual, ~5 minutes)

1. Go to notebooklm.google.com and create a notebook named “SenseBridge”.
2. Add sources — NotebookLM accepts pasted URLs, uploaded files, and pasted text. Use the repo docs below as sources; re-add after major doc changes (NotebookLM does not watch a live repo).
3. Ask NotebookLM questions grounded in exactly those sources, or generate its study guide / audio overview features from them.

Recommended sources, by topic

Add this repo’s own docs first — they carry the SenseBridge-specific constraints (safety-framing, on-device doctrine, accessibility gate) that generic framework docs won’t know about:

• docs/ARCHITECTURE.md, docs/PRODUCT.md, docs/TESTING.md, docs/TOOLING.md, docs/ENVIRONMENT.md, docs/ACCESSIBILITY.md, docs/SAFETY-FRAMING.md, docs/PRIVACY.md, AGENTS.md, CLAUDE.md

Then add official docs for whichever part of the stack you’re ramping up on — most of this list describes the planned Next.js/React migration, not the current state (website/ is still static HTML/CSS; see docs/TOOLING.md’s “Marketing website” row):

Why this isn’t wired into Claude, Continue, or any agent

NotebookLM has no public API — nothing here (or anywhere) could programmatically push sources into it without either screen-scraping the web UI (the rejected path above) or a human doing it by hand. Claude, Continue, and this repo’s own skills already have direct, local access to every doc listed above — reach for docs/TOOLING.md, the .agents/skills/ skills, or the documentation skill for AI-agent-facing teaching. NotebookLM is a human study aid layered on top, not an agent context source.

Need help? See SUPPORT.md.