nateherkai/roast-room

Roast Room – AI‑powered business‑decision council

What it is

• A Node‑based web app that lets you feed a business decision to a simulated council of seven famous entrepreneurs/investors. The council is driven by Claude Code sub‑agents that embody the thinking of Jeff Bezos, Paul Graham, Alex Hormozi, Gary Vaynerchuk, Dan Martell, Gino Wickman and Warren Buffett.
• Each “seat” has a brain – a collection of quotes, frameworks and principles extracted from that person’s public writings, talks and interviews. The brains are stored in the repository under .claude/skills/roast-room/brains/.
• Claude Code acts as the chair: it orchestrates rounds of discussion, scores every option according to the weight of each pillar, flags ruin risk, and finally writes a verdict and a PDF report.

How to use it

1. Clone the repo and run npm start. A tiny web server starts on http://localhost:4730 showing the “room”.
2. In the same folder open Claude Code (claude command) and type /roast-room followed by a decision prompt.
3. Provide a one‑line decision, the options you’re weighing, key numbers (revenue, cash, team size), limits (budget, deadline), success criteria and your current leaning.
4. Claude Code (the Chair) asks the council’s questions, the seven agents argue, score the options and produce a short verdict in‑chat and a full markdown report saved under sessions/<date>-<decision>/verdict.md.
5. The web UI replays the session and lets you download a nicely formatted PDF.

Key features

• Seven expert personas covering strategy, product, pricing, marketing, people, operations and finance.
• Prompt‑driven scoring: each seat scores every option; the Chair aggregates the scores, caps the verdict on ruin risk, and forces deeper debate when consensus is too quick.
• Zero‑dependency server: the room runs on plain Node 20, no extra packages needed.
• Claude Code integration: the heavy lifting (LLM calls) happens inside Claude Code, so you only need an Anthropic plan that includes Claude Code (no separate API key). An optional mode (COUNCIL_RUNNER=api) can use your own Anthropic API key for direct calls.
• Replayable sessions: all council runs are saved; you can view past sessions in the UI or download PDFs.
• Transparent sourcing: every quote the agents use is tied to a source file; a helper script checks that quotes match the brain files.

Typical workflow example

/roast-room I run a coaching business at $30k a month with 15 clients at $2,000 each.
Should I raise to $3,500 for everyone, raise for new clients only, or keep $2,000?
Close rate is about 50% and I have $40k in the bank. Success is $45k a month by March.
I'm leaning toward raising for everyone because I'm underpaid.

The council will discuss the pricing impact (Hormozi), market demand (Graham), cash runway (Buffett), etc., then output a scored recommendation.

Why it matters

• Demonstrates how LLM‑driven agents can be instantiated with real‑world expertise by feeding them curated knowledge bases.
• Shows a practical, low‑code way to augment human decision‑making with multi‑perspective AI reasoning.
• All core logic lives in the repository; you can inspect or extend the brains, add new advisors, or change the scoring rubric.

License & credits

• MIT‑licensed code.
• PDF generation uses jsPDF (MIT) and the Montserrat font (SIL Open Font License).
• Built by the AI Automation Society.

---

All details are taken directly from the repository’s README.