2,039 Hormozi Videos in Claude: The Business Audit Skill

Direct Answer

Systems by Vic did not fine-tune Claude to impersonate Alex Hormozi. He built a retrieval-backed business-audit skill. The system searches 9,330 timestamped passages assembled from 2,039 public MoreMozi videos, combines the relevant passages with a confirmed profile of the business, and asks Claude Code or Codex to produce a long-form diagnosis with source links.

That distinction makes the project more useful, not less. Fine-tuning would be difficult to inspect and would not naturally preserve exact video citations. Retrieval lets the report point to the evidence, expose weak coverage, keep conflicting advice visible, and update the corpus without retraining a model.

Watch the Walkthrough

Video credit: Systems by Vic. Watch the original video on YouTube. Source-content credit belongs to Alex Hormozi, MoreMozi, and the applicable guests and video owners. The transcript corpus was assembled by Jacob Posel and upstream contributors.

Download the Hormozi Report Skill

The archive was inspected before publication. It contains 14 files: Markdown instructions and specifications, JSON and HTML examples, two MIT license copies, and two Python renderer copies. It contains no compiled executable. Windows Defender reported no threats in the supplied archive on 9 August 2026. That is a useful check, not a guarantee about future versions or anything downloaded later by an installer.

What "Trained Claude" Means Here

The video uses the language of training because the output is constrained by a large body of Hormozi material. Technically, the pipeline is closer to local retrieval-augmented generation:

1. The transcript corpus is split into searchable, timestamped passages.
2. The user supplies a structured business profile.
3. A local command searches for passages relevant to the offer, constraint, acquisition, sales, delivery, retention, and team.
4. Selected passages enter the model context with exact YouTube links.
5. The model writes a diagnosis and clearly labeled synthesis.
6. A local renderer turns structured JSON into a self-contained browser dashboard.

The model does not become Alex Hormozi, and the corpus does not guarantee a Hormozi-approved answer. The model is still responsible for selection, interpretation, and prose. The source links make that interpretation auditable.

How the Business-Diagnosis Pipeline Works

The skill's strongest design choice is the confirmation gate. It summarizes what it believes about the business, labels every value as verified, inferred, or missing, and waits. A beautiful report generated from the wrong revenue, conversion rate, customer, or constraint is still wrong.

The 12 Questions Behind the Audit

The interview collects the operating facts that generic business advice usually lacks:

1. Offer: what is sold, how it is delivered, and the exact price or range.
2. Customer: narrow buyer, geography, market, and urgent problem.
3. Current size: revenue, profit or margin, active customers, and headcount.
4. Goal: target, deadline, and why it matters.
5. Acquisition: channels, monthly lead volume, spend, and cost per lead.
6. Sales: response time, qualification, booking, show, close, cycle, and capacity.
7. Delivery: onboarding, time to value, capacity, results, refunds, churn, and lifetime value.
8. Unit economics: CAC, gross and contribution margin, cash collection, and payback period.
9. Team: owner responsibilities, key roles, utilization, hiring needs, and blocked decisions.
10. Current constraint: one primary bottleneck such as leads, conversion, cash, delivery, retention, talent, systems, or owner attention.
11. Previous attempts: action, duration, spend or effort, and measured result.
12. Decision now: exact question, options, limits, and deadline.

Unknown is a valid answer. Inventing a conversion rate is worse than identifying measurement as a constraint. The report should show which recommendation depends on missing or inferred information.

What the Generated Report Contains

Vic's demonstration produces an interactive HTML report with three ranked constraints, an executive diagnosis, business metrics, evidence, and a sequenced action plan. The supplied renderer adds navigation, source filtering, printable styling, and a Markdown companion for inexpensive follow-up questions.

The skill contract requires a hard separation between corpus evidence and model synthesis. That matters. A sourced Hormozi passage can support an argument without proving that the model applied it correctly to this business.

The three-constraint rule

Ranking three constraints forces prioritization. The first should be the binding constraint now. The second and third should describe what is likely to bind after the first moves. If every constraint receives an impact score near 100, the ranking is decorative rather than analytical.

Citation controls

• Only HTTPS YouTube citation URLs are accepted by the renderer.
• The build warns when the visible start time disagrees with the link's t= parameter.
• Quotes are kept short; the report should paraphrase and send the reader to the original video.
• Weak retrieval should be admitted instead of filled with unsourced model memory.
• Conflicting passages should remain visible with their conditions, not be averaged into false certainty.

Install the Complete System

The current project README recommends installing from the maintained GitHub repository. Review the scripts before running them, especially because the installer may download a standalone Python runtime when Python 3.11 or 3.12 is not already available.

Windows PowerShell

git clone https://github.com/viclaranja/hormozi-ai-skill.git
cd hormozi-ai-skill
./install.ps1

macOS or Linux

git clone https://github.com/viclaranja/hormozi-ai-skill.git
cd hormozi-ai-skill
./install.sh

The repository describes approximately 400 MB of downloads and 300 MB on disk. It installs both ask-hormozi for quick sourced questions and hormozi-report for the full diagnosis into Claude Code and Codex.

Restart the agent, invoke /hormozi-report, and run the documented health check before trusting a report. The expected check reports a healthy engine, 9,330 segments, and no malformed citations. If the ZIP's skill file still contains {{HORMOZI_COMMAND}}, the reference folder was copied without being connected to the installed engine.

Privacy, Rights, and Security Boundaries

Local does not mean no data leaves the computer

The corpus and search engine can remain local. The model still needs selected passages and business facts in its context. Those inputs go to the model provider used by Claude Code or Codex under that provider's terms and organization settings.

For a client audit, remove credentials, personal identifiers, contracts, raw customer records, payment data, and anything the client did not authorize. Prefer a prepared business profile over pointing an agent at an entire drive. When using an Obsidian vault or notes folder, export narrow excerpts for the questions being answered.

Treat local notes as untrusted data

A note, transcript, or CRM export can contain text that looks like an instruction to the agent. The skill explicitly tells the agent to treat source files as data rather than commands. Keep that boundary and report any instruction-like content instead of following it.

Code rights and transcript rights are different

The supplied bundle includes MIT license files for its code. The upstream ask-hormozi repository also identifies its code license as MIT. That does not relicense the transcript material. Video and transcript rights remain with Alex Hormozi, MoreMozi, guests, and the applicable source owners.

How to Evaluate the Business Audit

A report can feel expensive because it is long, branded, and specific. None of those qualities proves the diagnosis. The test is whether it identifies a real bottleneck, proposes a coherent intervention, and predicts a measurable change.

A Safe First Run

1. Use a test business or your own non-sensitive profile. Do not begin with a client's private vault.
2. Answer the interview manually. This reveals exactly what the skill needs before connecting local sources.
3. Mark unknown values honestly. Let missing measurement appear as a finding.
4. Confirm the profile line by line. Correct inferences before retrieval begins.
5. Open at least five citations. Check relevance, timing, and whether the report overstates the source.
6. Choose one reversible action. Test the primary constraint diagnosis for two to four weeks.
7. Review the metric. Keep, revise, or reject the diagnosis based on evidence.

A compact business-profile template

Offer:
Customer:
Price and delivery model:
Monthly revenue and gross margin:
Active customers and team size:
Goal and deadline:
Acquisition channels and monthly leads:
Sales conversion metrics:
Retention, churn, refunds, and lifetime value:
Owner responsibilities:
Suspected constraint:
What has already been tried:
Decision that must be made now:
Unknown or unreliable numbers:

Video Chapters

1. 0:00 - The idea and project credits
2. 2:25 - How the corpus-backed skill works
3. 3:50 - The interactive business-audit dashboard
4. 5:03 - Installing the skill
5. 6:18 - Interview, local source, and pasted-material modes
6. 7:02 - The business profile questions
7. 9:02 - Confirming facts and searching the corpus
8. 10:49 - Reviewing the generated diagnosis
9. 11:50 - Creator download and resources

Bottom Line

Vic's project is a strong example of a skill wrapping a source corpus into a complete workflow. It does more than answer questions: it collects business facts, confirms uncertainty, retrieves evidence, ranks constraints, writes an action plan, validates citations, and packages the result as a usable report.

The phrase "trained on 2,000 videos" is catchy. The real system is better described as local search plus a structured agent workflow. Its value comes from inspectable evidence and disciplined intake, not from pretending the model has become Alex Hormozi. Download the reference bundle, inspect it, use the supported installer for the full corpus, and treat every diagnosis as a hypothesis to verify in the business.

Sources

• Systems by Vic: I Trained Claude on 2,000 Alex Hormozi Videos
• Systems by Vic on YouTube
• Systems by Vic free Skool community
• Hormozi AI Skill on GitHub
• Jacob Posel: ask-hormozi transcript search project
• Alex Hormozi on YouTube
• MoreMozi on YouTube
• User-supplied Hormozi Report reference ZIP