Roast — 5-Angle Idea Panel → One Verdict
Roast is an adversarial idea panel that pressure-tests a business or product idea from five angles — Critic, Champion, Analyst, Investigator, and Customer — then synthesizes a single GO, RESHAPE, or KILL verdict with a concrete 48-hour test to de-risk the riskiest assumption. It produces a decisive, skimmable verdict report.
✨ What it does
- Convenes a 5-angle panel (Critic, Champion, Analyst, Investigator, Customer) in parallel.
- Builds a normalized brief from four inputs (idea, audience, money, edge, constraints).
- Synthesizes panel scores with weighted logic and veto gates into a verdict.
- Designs a cheapest 48-hour test targeting the riskiest assumption.
- Outputs a structured verdict with confidence, risks, upside, and money read.
- Includes references and assets for methodology and examples.
🎯 When to use it
- When you want a brutal, multi-angle second opinion before building something.
- When you need to validate or stress-test a business idea quickly.
- When you're about to invest time or money and want to surface blind spots.
- When you want a clear GO/RESHAPE/KILL call rather than vague advice.
🚀 How to use
Trigger: /roast
Trigger with the /roast slash command or a request to 'roast' or 'pressure-test' an idea. Provide the idea in your prompt; the skill will ask up to 4 clarifying questions (idea, audience, money model, edge, constraints) if not already given. Example prompts:
/roast An AI that drafts grant applications for small nonprofits from a 10-min intake call
Roast my idea: a subscription box for eco-friendly office supplies. I have a small Instagram following.
📄 Output: A verdict report with GO/RESHAPE/KILL, confidence, panel scores, and a cheapest 48-hour test.
📦 Add this skill to Claude Code
# 1. Get the skills repo
git clone --depth 1 https://github.com/alirezarezvani/claude-skills /tmp/claude-skills
# 2. Copy this skill into your project (or ~/.claude/skills for all projects)
mkdir -p .claude/skills
cp -r /tmp/claude-skills/productivity/roast/skills/roast .claude/skills/roastSkill source: productivity/roast/skills/roast/SKILL.md
⚠️ Good to know
The skill relies on the user providing sufficient context; if the brief is too thin, the panel may argue past each other. The Investigator uses web search, but other panelists reason without external data.
❓ FAQ
What verdicts can the Judge give?
The Judge must pick one of three: GO, RESHAPE, or KILL. 'It depends' is not allowed.
How are the five panel scores combined?
The synthesizer weights Customer and Critic heaviest and Champion lightest, applies veto gates (e.g., a Customer who won't pay vetoes a GO), and flags the widest disagreement as the tension to resolve. Scores are not averaged.
What is the most important output?
The cheapest 48-hour test, which validates the riskiest assumption before building anything.
🤖 Overview, features, install steps and FAQ were generated from the project's SKILL.md on Sep 4, 2026. Always check the original source before running commands.