NotebookLM Skill Factory#

Orchestrate the full pipeline: NotebookLM source ingestion → structured SKILL.md extraction → write to skills directory → validate → test → iterate.

Prerequisites Check#

Before the pipeline, verify NotebookLM is authenticated:

bash

notebooklm status || notebooklm login

If notebooklm login opens a browser, tell user to complete Google OAuth and press ENTER in their terminal.

Pipeline#

Phase 1: Setup & Source Ingestion#

1. Create a dedicated notebook for this skill: bash

   notebooklm create "Skill: {skill-name}" --json
   
   

   Parse the notebook id from JSON output.

2. Collect sources from user. Ask:

   "What sources should this skill be based on? Give me URLs, local file paths, YouTube links — anything high-quality and specific to this domain."

   If the user doesn't have sources ready, suggest:

   • Official documentation URLs
   • Well-known books/articles on the topic (user can provide PDFs)
   • YouTube tutorials/channels specializing in the area

3. Add sources (use -n <notebook_id> for all subsequent commands): bash

   notebooklm source add "https://..." -n <id> --json
   notebooklm source add ./local-file.pdf -n <id> --json
   
   

   Capture each source_id.

4. Wait for indexing — spawn a background agent to avoid blocking: bash

   notebooklm source wait <source_id> -n <id> --timeout 600
   
   

   For multiple sources, wait them all. If any source fails indexing (exit code 1), log a warning but continue with remaining sources.

Phase 2: Extract SKILL.md#

1. Load the extraction prompt template: Read references/skill-extraction-prompt.md. Replace {USER_INTENT} with the user's original request.

2. Query NotebookLM against all indexed sources: bash

   notebooklm ask "{extraction_prompt}" -n <id> --json
   
   

3. Parse the output into a clean SKILL.md: bash

   echo '<json_output>' | python3 scripts/parse-skill-output.py > /tmp/skill-output.md
   
   

   Or save the JSON to a temp file first, then pipe.

4. Validate basic structure:

   • Starts with --- YAML frontmatter
   • Has name: and description: fields
   • Body is non-empty after frontmatter If validation fails, re-prompt NotebookLM with more specific instructions.

Phase 3: Install & Validate#

1. Create the skill directory: bash

   mkdir -p ~/.claude/skills/{skill-name}
   
   

2. Write the SKILL.md: Move the parsed output to ~/.claude/skills/{skill-name}/SKILL.md

3. Run skill-creator validation (via Skill tool): Invoke skill-creator with: "Validate the skill at ~/.claude/skills/{skill-name}/ and fix any issues found. Run package_skill.py to check for errors."

4. Run skill-vetter security check (via Skill tool): Invoke skill-vetter with: "Review the skill at ~/.claude/skills/{skill-name}/ for security issues."

Phase 4: Test & Iterate#

1. Test the skill by invoking it with a realistic prompt matching its intended use case.

2. Collect failures:

   • Did the skill trigger correctly?
   • Did it produce correct output?
   • Any hallucinations or gaps?

3. If issues found, iterate:

   • Feed the failure description back to NotebookLM: bash

     notebooklm ask "The SKILL.md generated earlier has this issue: {failure}. Based on the sources, rewrite it to fix this. Output complete corrected SKILL.md in a markdown code block." -n <id> --json
     
     

   • Parse again and overwrite SKILL.md

   • Re-test

4. Repeat until the skill passes a real usage test.

Error Recovery#

SituationActionNotebookLM auth expiredRun notebooklm login, retrySource indexing failedSkip that source, warn user, continueExtraction prompt returned emptyCheck if sources are indexed (status=ready), re-query with --new flagGenerated SKILL.md fails validationSend failure description back to NB for rewrite (Phase 4 iteration)NB rate limitedWait 5-10 min, retry once

Source Quality Guidelines#

Remind users:

• One notebook per skill, one theme per notebook
• 3-10 high-quality sources work better than 20+ mixed ones
• Official docs > blog posts > YouTube transcripts
• If the skill covers multiple unrelated domains, create separate notebooks for each
• After generation, user should be the final human reviewer — check tone, logic, completeness

Completing the Pipeline#

When the skill passes testing, report:

• Skill name and path ( ~/.claude/skills/{name}/)
• Number of sources used
• Number of iteration rounds
• Final test result summary