literary-review

Literary Review

Gather and synthesize knowledge via NotebookLM. Works standalone or as Phase 3 of a research engagement.

NotebookLM CLI reference: ${CLAUDE_PLUGIN_ROOT}/skills/literary-review/references/notebooklm-cli.md NotebookLM scripts: ${CLAUDE_PLUGIN_ROOT}/scripts/notebook-*.sh — use these instead of calling notebooklm directly. They encode the correct CLI syntax.

---

Phase 0 — Resume Detection

Check for an existing research session:

ENGAGEMENT_DIR="${ENGAGEMENT_DIR:-.}"
test -f "$ENGAGEMENT_DIR/.research.json" && python3 -c "
import json
with open('$ENGAGEMENT_DIR/.research.json') as f:
    d = json.load(f)
print(d.get('status', 'none'))
print(d.get('notebook_id', ''))
print(d.get('title', ''))
"

• status: gathering → research still running. Check with notebooklm research status -n NOTEBOOK_ID.
• status: ready → sources imported. Skip to Phase 3 (Source Review).
• status: synthesized → already done. Report and offer to re-query.
• Otherwise → start fresh at Phase 1.

---

Phase 1 — Intake

Extract from the user's message or PI's spawn prompt:

• topic: research subject (notebook title and research query)
• seed_urls: starting point URLs (0 or more)
• focus: specific angle or constraints
• notebook_id: existing NotebookLM notebook to reuse (if provided)

Check for reusable notebooks

Before creating a new notebook, check if a relevant one already exists from a prior engagement. Topics like "Drupal cache optimization" or "CDN integration patterns" are project-agnostic — the same reference material applies across projects.

# List existing notebooks
notebooklm list 2>/dev/null | grep -i "<topic keywords>"

If an existing notebook covers the same domain, reuse it by passing its ID. You can still add project-specific seed URLs as new sources. Only create a new notebook if no relevant one exists.

Defaults:

• Research mode: deep (20+ sources)
• Research source: web
• Notebook title: "Research: <topic>"

Write session state:

python3 -c "
import json
state = {
    'version': '1.0',
    'title': 'Research: TOPIC',
    'topic': 'TOPIC',
    'focus': 'FOCUS',
    'seed_urls': [],
    'notebook_id': '',
    'status': 'creating'
}
with open('$ENGAGEMENT_DIR/.research.json', 'w') as f:
    json.dump(state, f, indent=2)
"

---

Phase 2 — Build the Notebook

Option A: New notebook (no existing notebook)

Use the setup script to create, seed, and fire research in one step:

NOTEBOOK_ID=$(${CLAUDE_PLUGIN_ROOT}/scripts/notebook-setup.sh "Research: TOPIC" \
  --seed-url "URL1" --seed-url "URL2" \
  --research "TOPIC FOCUS" --no-wait)

Update .research.json with the notebook ID and set status to gathering.

Option B: Existing notebook

If the user provided a notebook ID, skip creation. Set it in .research.json and proceed to Phase 3.

Wait for research (if --no-wait was used)

Spawn a background agent to wait:

Agent(
  description="Wait for NotebookLM research",
  subagent_type="general-purpose",
  run_in_background=true,
  prompt="Wait for NotebookLM deep research to complete.
  Run: notebooklm research wait -n NOTEBOOK_ID
  When complete, run: notebooklm source list -n NOTEBOOK_ID --json
  Count sources with status=ready.
  Update $ENGAGEMENT_DIR/.research.json: set status='ready', source_count=N."
)

Tell the user: research is running (15-30 minutes). Offer to add more URLs while waiting.

---

Phase 3 — Source Review

notebooklm source list -n NOTEBOOK_ID

Show the source list. Ask the user/PI:

"Here are the gathered sources. Remove any irrelevant ones? Add any missing URLs?"

Handle removals:

notebooklm source delete SOURCE_ID -n NOTEBOOK_ID

Handle additions:

notebooklm source add "URL" -n NOTEBOOK_ID --json

---

Phase 4 — Synthesis

Run a structured synthesis query using the ask script:

${CLAUDE_PLUGIN_ROOT}/scripts/notebook-ask.sh NOTEBOOK_ID \
  "Summarize the key themes, best practices, and important distinctions across all sources on: TOPIC FOCUS. Structure: (1) core concepts, (2) common patterns, (3) known pitfalls or debates."

Save the synthesis as a notebook note:

${CLAUDE_PLUGIN_ROOT}/scripts/notebook-ask.sh NOTEBOOK_ID \
  "Summarize the key themes, best practices, and important distinctions across all sources on: TOPIC FOCUS. Structure: (1) core concepts, (2) common patterns, (3) known pitfalls or debates." \
  --save-as-note --note-title "Research Summary: TOPIC"

---

Phase 5 — Output

Write 02-literary-review.md to the engagement directory. Structure:

# Literary Review: TOPIC

## Notebook
- ID: NOTEBOOK_ID
- Sources: N curated sources
- Research mode: deep web

## Core Concepts
<from synthesis>

## Common Patterns
<from synthesis>

## Known Pitfalls and Debates
<from synthesis>

## Key Sources
<top 5 sources with brief descriptions>

Update .research.json status to synthesized.

Vault archival (standalone mode only)

If running standalone (not inside research-lab:run):

VAULT_ROOT="$HOME/Vaults/${OBSIDIAN_VAULT_NAME:-Neurons}"
TOPIC_SLUG="<kebab-case-topic>"
DEST="Research/$TOPIC_SLUG/$(date +%Y-%m-%d)-literary-review.md"
mkdir -p "$VAULT_ROOT/$(dirname "$DEST")"
cp "$ENGAGEMENT_DIR/02-literary-review.md" "$VAULT_ROOT/$DEST"

---

User Preferences

• Always deep mode. Fast is for lookups, not research.
• Web by default. Drive only when explicitly asked.
• Seed first — add known URLs before firing research.
• Review before synthesizing — show sources, let the user curate.
• Save synthesis as a notebook note for later retrieval.