nlm-skill

NotebookLM CLI & MCP Expert

Leading words: nlm, NotebookLM, audio overview, studio artifacts, progressive disclosure, phase separation, confirm-required, capture IDs, alias shortcuts, no REPL.

This skill provides comprehensive guidance for using NotebookLM via both the nlm CLI and MCP tools. The body carries the critical rules, the tool detection decision tree, and the 3-phase workflow. Depth lives in references/ behind context pointers.

Tool Detection (CRITICAL — read first)

ALWAYS check which tools are available before proceeding:

1. Check for MCP tools: Look for tools starting with mcp__notebooklm-mcp__*.
2. If BOTH MCP tools AND CLI are available: ASK the user which they prefer.
3. If only MCP tools are available: Use them directly (see references/mcp_tools.md).
4. If only CLI is available: Use nlm CLI commands via Bash (see references/command_reference.md).

has_mcp_tools = check_available_tools()  # mcp__notebooklm-mcp__*
has_cli = check_bash_available()         # Can run nlm commands

if has_mcp_tools and has_cli:
    user_preference = ask_user()
else if has_mcp_tools:
    use_mcp_tools()
else:
    use_cli()

Critical Rules (read first)

1. Always authenticate first: Run nlm login before any operations.
2. Sessions expire in ~20 minutes: Re-run nlm login if commands start failing with auth errors.
3. ⚠️ ALWAYS ASK USER BEFORE DELETE: Deletions are irreversible. Show what will be deleted and warn about permanent data loss.
4. --confirm is REQUIRED: All generation and delete commands need --confirm / -y (CLI) or confirm=True (MCP).
5. Research requires --notebook-id: The flag is mandatory, not positional.
6. Capture IDs from output: Create/start commands return IDs needed for subsequent operations.
7. Use aliases: nlm alias set <name> <uuid> simplifies long UUIDs.
8. Check aliases before creating: nlm alias list first to avoid name conflicts.
9. DO NOT launch REPL: Never nlm chat start — it opens an interactive REPL that AI tools cannot control. Use nlm notebook query for one-shot Q&A.
10. Output format: Default (no flags) = compact + token-efficient. --quiet captures IDs for piping. --json only when parsing fields.
11. Use --help when unsure: nlm <command> --help shows options.