Agent-Skills.md

Agent Skills: NotebookLM CLI (nlm)

This skill should be used when the user asks to "create a notebook", "add source to notebook", "generate audio overview", "create podcast", "manage NotebookLM", "nlm", "add PDF to notebook", "list notebooks", "summarize sources", "generate study guide", "create FAQ", "briefing document", "chat with notebook", "generate outline", "research a topic", "deep research", or needs to interact with Google NotebookLM via the nlm CLI tool.

UncategorizedID: edwinhu/workflows/nlm

Repository

edwinhu/workflows
edwinhu
5

Install this agent skill to your local

pnpm dlx add-skill https://github.com/edwinhu/workflows/tree/HEAD/skills/nlm

Skill Files

Browse the full folder contents for nlm.

Download Skill

Loading file tree…

skills/nlm/SKILL.md

Skill Metadata

Name
nlm
Description
This skill should be used when the user asks to "create a notebook", "add source to notebook", "generate audio overview", "create podcast", "manage NotebookLM", "nlm", "add PDF to notebook", "list notebooks", "summarize sources", "generate study guide", "create FAQ", "briefing document", "chat with notebook", "generate outline", "slide deck", "research a topic", "deep research", or needs to interact with Google NotebookLM via the nlm CLI tool.

NotebookLM CLI (nlm)

Manage Google NotebookLM notebooks, sources, notes, chat, and generated content (audio/video overviews, slide decks, apps, reports) via the nlm command-line tool.

Requires: nlm on PATH (nix-managed, built from upstream tmc/nlm). Install/update via cd ~/nix && nix run .#build-switch.

Check: command -v nlm || echo "MISSING: nlm CLI not installed"

Command structure: the CLI uses subcommand groups — nlm <group> <action> (e.g. nlm notebook list, nlm source add, nlm audio create). The older flat aliases (nlm list, nlm add, nlm audio-create, …) were removed in upstream's parser migration. When unsure of exact syntax, run nlm --help or nlm <group> --help — it's authoritative. The references/*.md files predate the migration and may be stale.

Authentication

Before first use, authenticate with Google:

nlm auth login -all

This connects to Chrome via CDP (Chrome DevTools Protocol) to extract cookies from an active NotebookLM session. Credentials are stored in ~/.nlm/env.

Troubleshooting Authentication

If nlm auth fails with "no valid profiles found" or "SESSION_COOKIE_INVALID":

  1. Verify Chrome is running with remote debugging:

    ps aux | grep -E "chrome.*remote-debugging-port=9400"

    If not running, Chrome needs to be started with --remote-debugging-port=9400.

  2. Test CDP connection: curl http://localhost:9400/json/version (should return Chrome version JSON).

  3. Re-authenticate with debug: nlm auth login -debug -all (shows which profiles are checked and why).

  4. Verify: nlm notebook list should list notebooks without errors.

You can also point at a remote CDP socket: nlm auth login -cdp-url ws://localhost:9222, or pick a Google account index with -authuser 1.

Core Commands

Notebooks

nlm notebook list
nlm notebook create "Research Notes"
nlm notebook rename <nb-id> "New Title"
nlm notebook delete <nb-id>
nlm notebook description <nb-id> [text]    # text via arg or stdin; empty clears
nlm notebook emoji <nb-id> <emoji>
nlm notebook featured                      # list featured notebooks
nlm analytics <nb-id>                      # usage time series

Sources

nlm source list <nb-id>
nlm source add <nb-id> <file|url|-> [more...]   # '-' streams stdin as one source
nlm source add <nb-id> https://www.youtube.com/watch?v=VIDEO_ID
nlm source rename <source-id> "New Name"
nlm source delete <nb-id> <source-id|a,b,c|->   # '-' reads newline-delimited IDs from stdin
nlm source refresh <nb-id> <source-id>
nlm source read <source-id> [nb-id]             # print the server-indexed text body
nlm source sync <nb-id> [paths...]              # bundle local files into a synced txtar source (auto-chunks at 5MB)
nlm discover-sources <nb-id> "query"            # discover relevant sources

Notes

nlm note list <nb-id>
nlm note read <nb-id> <note-id>
nlm note create <nb-id> "Title" ["content"]     # content via arg or stdin
nlm note update <nb-id> <note-id> "content" "Title"
nlm note delete <nb-id> <note-id>

Chat & research

nlm generate-chat <nb-id> "What are the main themes?"   # one-shot, streamed
nlm chat <nb-id> ["prompt"]                             # interactive; one-shot if a prompt is given
nlm chat <nb-id> -f prompt.txt                          # read a long prompt from file ('-' = stdin)
nlm chat <nb-id> --source-match <regex>                 # focus on matching sources (also --source-ids, --label-match, --citations)
nlm research <nb-id> "query" --mode=fast|deep [--md]    # research + import sources (NOTE: notebook-id comes FIRST)

Source → output transforms

All take <nb-id> [source-id...] (omit source ids to use all sources):

| Command | Purpose | |---------|---------| | summarize / briefing-doc | Summary / professional briefing | | study-guide / faq | Key concepts + review questions / FAQ | | outline / toc / timeline | Outline / table of contents / timeline | | rephrase / expand / explain | Reword / elaborate / simplify | | critique / verify | Critique / fact-check | | brainstorm | Ideate from sources | | mindmap <nb-id> <source-id> [...] | Interactive mindmap (opens in browser) |

nlm summarize <nb-id>
nlm study-guide <nb-id> <source-id-1> <source-id-2>

Audio & video overviews

nlm audio create <nb-id> "Focus on key themes, professional tone"   # alias: nlm create-audio
nlm audio list <nb-id>
nlm audio get <nb-id>
nlm audio download <nb-id> [output.mp3]
nlm audio share <nb-id> [--public]
nlm audio delete <nb-id>
nlm audio-suggestions <nb-id>                  # blueprint ideas as JSON lines (pipe to create-audio)

nlm video create <nb-id> "instructions"
nlm video list <nb-id>
nlm video get <nb-id>
nlm video download <nb-id> [output.mp4]

Slide decks, apps, reports (added upstream)

# Slide decks → exportable as PDF / PPTX
nlm deck create [--format detailed|presenter] <nb-id> ["instructions"]
nlm deck download <nb-id> --id <artifact-id> [--format pdf|pptx] [--output file]

# Interactive app artifacts
nlm app create --type prototype|mindmap|canvas <nb-id> ["instructions"]
nlm mindmap create <nb-id> ["instructions"]

# Reports
nlm report-suggestions <nb-id>                 # list valid report types
nlm create-report <nb-id> <report-type> [description] [instructions]
nlm generate-report <nb-id> [--sections ...]

# Manage any generated artifact (decks, apps, reports, etc.)
nlm artifact list <nb-id>
nlm artifact get <artifact-id>
nlm artifact update <artifact-id> [new-title]
nlm artifact delete <artifact-id>

Labels (autolabel clusters)

nlm label list <nb-id>
nlm label generate <nb-id>                     # recompute clusters
nlm label create <nb-id> <name> [emoji]
nlm label attach <nb-id> <label-id> <source-id>
nlm label unlabeled <nb-id>                    # apply existing labels to unlabeled sources

Sharing & other

nlm share <nb-id>                              # share publicly
nlm share-private <nb-id>                      # share privately
nlm generate-guide <nb-id>                     # notebook guide (short summary)
nlm magic <nb-id> [source-id...]               # notebook 'Magic View' synthesis
nlm mcp                                        # run the MCP server on stdin/stdout
nlm account                                    # show/set the authenticated NotebookLM account
nlm refresh                                    # refresh stored auth credentials

Exit codes: 0 ok · 2 bad args · 3 auth required/invalid · 4 not found · 5 precondition (quota, source cap, wrong type) · 6 transient (rate limit/5xx) · 7 busy (still generating).

Workflows

For longer recipes (research → study materials, content analysis, Readwise→NLM import), see references/workflows.md (note: predates the parser migration — translate flat commands to the subcommand form above).

Quick start — automated research:

id=$(nlm notebook create "Topic Research")        # capture the new notebook id from output
nlm research "$id" "your topic" --mode=deep
nlm generate-chat "$id" "What are the key findings?"

Troubleshooting

  • Auth errors (exit 3): nlm auth login -all to re-authenticate; nlm refresh to refresh credentials.
  • Debug mode: add -debug for detailed API interaction logs.
  • Resource busy (exit 7): generation (audio/video/deck/report) is async — re-run the get/list/artifact get command until it's ready.
  • Authoritative syntax: nlm --help and nlm <group> --help always reflect the installed binary.

Environment Variables

  • NLM_AUTH_TOKEN: Authentication token (managed by the auth command)
  • NLM_COOKIES: Authentication cookies (managed by the auth command)
  • NLM_BROWSER_PROFILE: Chrome/Brave profile to use (default: "Default")
  • NLM_AUTHUSER: Google account index for multi-account profiles