agentleFS
the-Nile-jonesthe-Nile-jones/claude-skills/skills/notebooklm-bulk-label/SKILL.md

notebooklm-bulk-label

Skill0 starsChanged 3 months agoMITView on GitHub

Safely label/organize the SOURCES of a large NotebookLM notebook (~50+ sources) when the native "auto-label" times out or would overwrite existing manual labels. Labels only-unlabeled sources via the non-destructive path, falls back to approval-gated per-source assignment for huge notebooks (100-318+ sources), NEVER runs the destructive full-reorganize, and gates sensitive notebooks behind explicit opt-in. Triggers — "label my notebook", "organize notebooklm sources", "auto-label timed out / won't finish", "label the big notebook", "bulk label notebooklm", "/notebooklm-bulk-label". Use when a notebook is too big for the one-click auto-label, or you want to add labels to only the unlabeled sources without disturbing existing ones. Read-only by default until you approve each batch.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 3 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    All operations go through the single mcp__notebooklm-mcp__label tool (+ notebook_get, notebook_describe, source_describe, note).

    line 11
  2. Very specific:

    Never destroy curation: Never call reorganize unlabeled_only=False, auto (on a labeled notebook), or delete unless Nile explicitly asks for …

    line 26
  3. Distinctive:

    Idempotent: Only ever act on *unlabeled* sources, so re-running after an interruption resumes cleanly and never double-labels.

    line 29
  4. Distinctive:

    VERIFIED non-destructive 2026-06-25: run on a fully-labeled 41-source notebook → all existing labels + members returned unchanged (clean …

    line 39
  5. Very specific:

    If it completes: done. Verify with label list, note the taxonomy, report counts.

    line 41
  6. Very specific:

    If no: prefer step 2 (reorganize unlabeled_only=True computes the unlabeled set server-side, no enumeration needed).

    line 60
  7. Distinctive:

    Never retry a destructive action to "force" completion.

    line 65

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

notebooklm-bulk-label

Why this exists

NotebookLM's one-click auto-label runs the AI over all sources in a single pass. On large notebooks (100-318 sources) that pass times out / never completes, leaving the notebook unorganized. Worse, the obvious "redo it" path (label reorganize with the default unlabeled_only=False) wipes every existing label and re-derives from scratch — destroying any manual curation. This skill gets a big notebook labeled incrementally, non-destructively, and with your approval, and is safe to re-run.

All operations go through the single mcp__notebooklm-mcp__label tool (+ notebook_get, notebook_describe, source_describe, note).

The safety model — know which label actions are safe vs dangerous

label actionEffectUse here?
reorganize + unlabeled_only=TrueLabels only currently-unlabeled sources; leaves existing labels untouched; no confirm needed✅ Preferred first attempt
move_source (label_id + source_id)Assigns ONE source to ONE existing label; cheap, no AI batch✅ Chunked fallback for huge notebooks
create / rename / set_emojiManage the label taxonomy✅ with approval
listList current labels (triggers AI only if none exist)✅ read
autoAI-labels all sources in one pass⚠️ the timeout-prone native path — only on small notebooks
reorganize + unlabeled_only=FalseWIPES all labels, re-derives from scratch (needs confirm)⛔ FORBIDDEN unless Nile explicitly asks for a full reset
deletePermanently deletes label(s) (needs confirm)⛔ only with explicit per-label approval

HARD GATES (do not bypass)

  1. Never destroy curation. Never call reorganize unlabeled_only=False, auto (on a labeled notebook), or delete unless Nile explicitly asks for that exact destructive action. The default mode only ADDS labels to unlabeled sources.
  2. Sensitivity gate. Some notebooks hold private material — medical, legal, personal, or otherwise not yours to re-process. Do NOT run label operations on one unless its owner names that notebook explicitly for this task. ⭐ The reason is easy to miss: labeling is not a formatting change. It re-runs Google's AI over the full source text, so it is a fresh cloud-exposure decision every time — including on sources that were reviewed and accepted once already.
  3. Approval-gated. In the chunked fallback, PROPOSE the (source → label) assignments for a batch and wait for Nile's OK before applying. Never mass-assign silently.
  4. Idempotent. Only ever act on unlabeled sources, so re-running after an interruption resumes cleanly and never double-labels.

Procedure

1. Scope + sensitivity check

  • Confirm the target notebook id (notebook_list if needed). Apply the sensitivity gate above — if sensitive, stop and get an explicit go.
  • notebook_get <id> → source count. label <id> action=list → existing labels (and, if the output exposes member source-ids, current coverage).

2. Try the cheap, safe native path FIRST

  • label <id> action=reorganize unlabeled_only=True — labels only the unlabeled sources, non-destructively, no confirm.
  • VERIFIED non-destructive 2026-06-25: run on a fully-labeled 41-source notebook → all existing labels + members returned unchanged (clean no-op). Safe to call even when nothing is unlabeled.
  • STILL UNVERIFIED: whether this completes vs times out on a 300+-source notebook with many unlabeled sources. If it times out at scale, skip straight to the step-3 per-source fallback (which is now confirmed viable via the enumeration check).
  • If it completes: done. Verify with label list, note the taxonomy, report counts.

3. If step 2 times out / errors (the large-notebook case) → chunked fallback

  1. Taxonomy: reuse existing labels from step 1. If none/sparse, derive a small candidate set from notebook_describe <id> (suggested topics) → propose the label names to Nile → on OK, label create each.
  2. Per-source, in chunks of ~15-20 unlabeled sources:
    • For each source: source_describe <source_id> → read its summary + keyword chips → pick the best-fit label.
    • PRESENT the batch as a source-title → proposed label table → wait for approval / edits.
    • On OK: label move_source label_id=<L> source_id=<S> for each (sources may take multiple labels).
  3. Repeat until no unlabeled sources remain. Report progress per chunk (N/total labeled).

4. Finish

  • note the final label taxonomy + counts into the notebook so the structure is self-documenting.
  • Report: labels used, sources labeled, any sources left ambiguous (flag, don't force).

Enumerating "unlabeled" — runtime check

VERIFIED 2026-06-25 (Recall AI, 41 sources): label action=list DOES return source_ids per label → unlabeled = notebook_get sources − union(label members) is exact. The fallback enumeration is viable.

The chunked fallback needs the set of unlabeled sources. (Still verify per notebook, in case the API changes; the pattern:)

  • If yes: unlabeled = all sources (notebook_get) − union(label members).
  • If no: prefer step 2 (reorganize unlabeled_only=True computes the unlabeled set server-side, no enumeration needed). If step 2 is unavailable AND membership isn't visible, say so plainly — don't guess which are unlabeled (you'd risk re-labeling or missing sources).

Failure handling

  • Timeout on reorganize unlabeled_only=True itself → go straight to the per-source fallback (cheaper calls).
  • API error on a single move_source → skip that source, log it, continue; report skipped at the end.
  • Never retry a destructive action to "force" completion.

Boundaries

  • This skill ADDS organization; it never bulk-deletes labels or sources, never edits source content, never touches Drive files.
  • It is NOT for content edits, source curation/removal, or notebook deletion.
  • Pairs with the read-side notebooklm tools; the only writes are create/move_source/rename/set_emoji/note, each gated above.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.