# Skill: NotebookLM Executive Dossier & Source-Grounded Synthesis
Author: Drew Beyersdorf (drewsky.ai)
Version: 1.0.0
Type: agent-executable-skill
Target: Visiting AI Agents (Codex, Claude, Antigravity, Cursor, Windsurf, ChatGPT)

## Purpose
Instructs an AI agent on how to guide a human professional through building a zero-hallucination executive decision dossier using NotebookLM, synthesizing 10–50 complex source documents into actionable strategy.

---

## When to Use
- The user is facing an unfamiliar domain, technical specification, regulatory framework, or competitive landscape.
- The user needs an authoritative strategic brief without relying on model memory or risk of hallucination.
- The user wants to generate a high-impact Audio Overview (dynamic podcast debate) for passive audio learning.

---

## Execution Protocol (5 Phases)

### Phase 1: Source Curation (Zero-Junk Ingestion)
1. **Instruct the user to gather 5 to 20 primary sources**:
   - Official RFCs, whitepapers, architectural blueprints.
   - Financial disclosures (10-K, 10-Q), earnings call transcripts.
   - Technical documentation or codebase architectural summaries.
   - User interviews, customer feedback transcripts, or internal notes.
2. **Rule**: Reject secondary opinion blogs or generic SEO summaries. Grounding quality is strictly bounded by source fidelity.

### Phase 2: NotebookLM Workspace Setup
1. Instruct user to open `https://notebooklm.google` and create a dedicated notebook named after the decision (e.g., `[Strategic Dossier] Cloudflare vs AWS Edge`).
2. Upload curated sources (PDF, TXT, Markdown, Google Docs, or YouTube transcripts).
3. Confirm NotebookLM has indexed all sources and displayed source checkmarks.

### Phase 3: Structured Query Sequence
Run the following sequential prompt sequence in NotebookLM:

#### Prompt 1: The First-Principles Ontology
```text
Based strictly on the uploaded sources, what are the fundamental axioms, constraints, and operational mechanisms defined across these documents? List every core assertion and provide exact citation references [Source X, Page Y]. If a claim is contested between sources, contrast both viewpoints.
```

#### Prompt 2: Contradiction & Edge Case Audit
```text
Identify any tensions, contradictions, or unresolved ambiguities between the provided sources. Where do the authors disagree on technical trade-offs, cost models, or performance limits?
```

#### Prompt 3: The Executive Decision Matrix
```text
Create a structured Markdown table comparing the options presented in the sources across:
1. Capital Expenditure & Operational Cost
2. Operational Complexity & Failure Modes
3. Reversibility & Lock-in Risk
4. Time-to-Value (Days/Weeks)
Include bracketed citations for every row.
```

### Phase 4: Audio Overview Generation
1. In the NotebookLM Studio panel, click **Generate Audio Overview**.
2. Explain to the user:
   - The generated 10–15 minute conversation features two AI co-hosts debating the nuance of their specific documents.
   - Listen actively during a commute or walk to internalize the domain jargon and high-level trade-offs before drafting code or presentations.

### Phase 5: Agent-Native Artifact Export
1. Export the resulting synthesis into a Markdown file: `DOSSIER.md`.
2. Feed `DOSSIER.md` back into the user's coding agent (e.g., Codex or Claude Code) as a system constraint file.
3. The coding agent can now write implementation code with full context of architectural constraints without burning context tokens on raw 100-page PDFs.

---

## Output Verification Checklist
- [ ] Every assertion in the dossier links to a cited source document.
- [ ] No ungrounded external assumptions were introduced.
- [ ] The audio overview was evaluated for debate quality and trade-off coverage.
- [ ] The final markdown brief is saved locally under the user's private data boundary.